P4 Prescott adding pipeline stages?

[Yousuf Khan]

Rumours are that Intel is adding additional pipeline stages to the Prescott,
even higher than its existing 20 stages. I can't see it happening in this
late stage of the game, perhaps half a year from now. It will possibly be
getting rid of the internal double clocked ALUs too, and replacing it with
two-stage single-clocked ALUs.

http://www.neowin.net/comments.php?id=16264&category=main

Yousuf Khan

[Robert Myers]

On Thu, 01 Jan 2004 18:07:01 GMT, "Yousuf Khan"
<removethisspam.bjsk90.removethispam@hotmail.com> wrote:
Rumours are that Intel is adding additional pipeline stages to the Prescott,even higher than its existing 20 stages. I can't see it happening in thislate stage of the game, perhaps half a year from now. It will possibly begetting rid of the internal double clocked ALUs too, and replacing it withtwo-stage single-clocked ALUs.http://www.neowin.net/comments.php?id=16264&category=main

Since Intel engineers came out with simulations showing fifty pipeline
stages as plausible, only the naive would be shocked that at some
point in the life of the Pentium four Intel would increase the number
of pipeline stages.

"That, however, will have an impact on the performance of the
processor, if it turns out to be true. Branch prediction misses will
cause a lot more trouble, for example. And, again, if this is correct,
it will mean that any performance increases Intel might have got, the
increased internal tables could well hold up additional instructions
which are shuffling through the processor. Where does Intel go from
here with the Pentium 4 architecture? It's evident that if Intel wants
to keep up not only with AMD's Opteron but also with chips such as its
own Pentium M, a very serious redesign of a flagship desktop processor
must be on the cards."

Yada. Yada. Yada. When the cost of pipeline stalls went up, ISV's
learned out to cope with them. Telling the world to use AMD
processors because they were easier to code for was not an option, and
it's still not an option. While hobbyists have been playing with
their Athlons, ISV's have been learning to cope with the P4.
Deepening the pipeline doesn't take them off in any new direction. It
just makes them work harder.

RM

[Yousuf Khan]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:14u8vvkfckfce8fkf738u6nstltbeuvfj6@4ax.com... "That, however, will have an impact on the performance of the processor, if it turns out to be true. Branch prediction misses will cause a lot more trouble, for example. And, again, if this is correct, it will mean that any performance increases Intel might have got, the increased internal tables could well hold up additional instructions which are shuffling through the processor. Where does Intel go from here with the Pentium 4 architecture? It's evident that if Intel wants to keep up not only with AMD's Opteron but also with chips such as its own Pentium M, a very serious redesign of a flagship desktop processor must be on the cards." Yada. Yada. Yada. When the cost of pipeline stalls went up, ISV's learned out to cope with them. Telling the world to use AMD processors because they were easier to code for was not an option, and it's still not an option. While hobbyists have been playing with their Athlons, ISV's have been learning to cope with the P4. Deepening the pipeline doesn't take them off in any new direction. It just makes them work harder.

However, it does make some sense that at some point in the future, Intel may
have to seriously take a look at replacing the Pentium 4 with the Pentium M.

Yousuf Khan

[Keith R. Williams]

In article <14u8vvkfckfce8fkf738u6nstltbeuvfj6@4ax.com>,
rmyers@rustuck.com says... On Thu, 01 Jan 2004 18:07:01 GMT, "Yousuf Khan" <removethisspam.bjsk90.removethispam@hotmail.com> wrote:Rumours are that Intel is adding additional pipeline stages to the Prescott,even higher than its existing 20 stages. I can't see it happening in thislate stage of the game, perhaps half a year from now. It will possibly begetting rid of the internal double clocked ALUs too, and replacing it withtwo-stage single-clocked ALUs.http://www.neowin.net/comments.php?id=16264&category=main Since Intel engineers came out with simulations showing fifty pipeline stages as plausible, only the naive would be shocked that at some point in the life of the Pentium four Intel would increase the number of pipeline stages. "That, however, will have an impact on the performance of the processor, if it turns out to be true. Branch prediction misses will cause a lot more trouble, for example. And, again, if this is correct, it will mean that any performance increases Intel might have got, the increased internal tables could well hold up additional instructions which are shuffling through the processor. Where does Intel go from here with the Pentium 4 architecture? It's evident that if Intel wants to keep up not only with AMD's Opteron but also with chips such as its own Pentium M, a very serious redesign of a flagship desktop processor must be on the cards." Yada. Yada. Yada. When the cost of pipeline stalls went up, ISV's learned out to cope with them. Telling the world to use AMD processors because they were easier to code for was not an option, and it's still not an option. While hobbyists have been playing with their Athlons, ISV's have been learning to cope with the P4. Deepening the pipeline doesn't take them off in any new direction. It just makes them work harder.

I suppose ISVs enjoy coding around the lack of the barrel shifter
and integer multiplier too?

Yada, indeed.

--
Keith

[Robert Myers]

On Thu, 1 Jan 2004 20:32:26 -0500, Keith R. Williams
<krw@attglobal.net> wrote:

<snip>I suppose ISVs enjoy coding around the lack of the barrel shifterand integer multiplier too?Yada, indeed.

When the VP in Charge of Dimwittery wanders by to ask why the "Intel
Inside" label on boxes is slowly being replaced by this weird Green
label that says AMD, you might want to skip the stack depth, barrel
shifter, and integer multiplier and tell him that some assistant who
has long ago been sacked made the mistake and you'll see to it that no
one does anything that foolish again.

If you're working for someone who would think ill of you because you'd
rather let the rest of the world do the experimenting, I would step
very carefully around the workplace. You never know what strange
ideas such an employer might have.

That's just the way it goes. When AMD figures _that_ part out, I
might invest in some AMD stock.

RM

[Yousuf Khan]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:6ql9vvcrtduv9edviol2cm7t6a3bpo00a9@4ax.com... That's just the way it goes. When AMD figures _that_ part out, I might invest in some AMD stock.

You mean you'd invest in AMD stock the day that AMD figures out that it is
has all of the cards stacked against it, and that it should just stop
competing against anything, and just stop producing products all together?
:-)

Doesn't sound like sound investment strategy to me, but hell either did the
DOTCOMmerce strategy, but it worked for them for several years.

Yousuf Khan

[George Macdonald]

On Fri, 02 Jan 2004 07:00:54 GMT, "Yousuf Khan"
<removethisspam.bjsk90.removethispam@hotmail.com> wrote:
"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:6ql9vvcrtduv9edviol2cm7t6a3bpo00a9@4ax.com... That's just the way it goes. When AMD figures _that_ part out, I might invest in some AMD stock.You mean you'd invest in AMD stock the day that AMD figures out that it ishas all of the cards stacked against it, and that it should just stopcompeting against anything, and just stop producing products all together?:-)Doesn't sound like sound investment strategy to me, but hell either did theDOTCOMmerce strategy, but it worked for them for several years.

Hmmm, I think it's another of Robert's Riddles.:-) I can't for the life of
me figure what AMD has to do to gain his confidence.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[Robert Myers]

On Fri, 02 Jan 2004 05:45:54 -0500, George Macdonald
<fammacd=!SPAM^nothanks@tellurian.com> wrote:
On Fri, 02 Jan 2004 07:00:54 GMT, "Yousuf Khan"<removethisspam.bjsk90.removethispam@hotmail.com> wrote:"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:6ql9vvcrtduv9edviol2cm7t6a3bpo00a9@4ax.com... That's just the way it goes. When AMD figures _that_ part out, I might invest in some AMD stock.You mean you'd invest in AMD stock the day that AMD figures out that it ishas all of the cards stacked against it, and that it should just stopcompeting against anything, and just stop producing products all together?:-)Doesn't sound like sound investment strategy to me, but hell either did theDOTCOMmerce strategy, but it worked for them for several years.Hmmm, I think it's another of Robert's Riddles.:-) I can't for the life ofme figure what AMD has to do to gain his confidence.

It's the marketplace that's the riddle, not me. "IBM doesn't make the
best computers," it was once said, "they just sell the most of them."

What would AMD have to do to gain the confidence of the marketplace?
For instance, what combination of circumstances would give OEM's the
breathing room to and the money to promote, say, Opteron, the way its
technical merits suggest it ought to be promoted?

They know, like I know, that if Opteron became a real threat to
Intel's market dominance, Intel would do whatever is necessary with
money and marketing to change that. And woe unto the long term
prospects of the OEM that got out of line.

The other part of the story is that, for the moment at least, Intel
needs AMD as a credible competitor they can point to so no one can
accuse them of being a monopoly. Therefore, even given the
opportunity, Intel would not crush AMD.

It's just not an emotional issue for me. Microsoft, I will admit, has
become an emotional issue for me, so I can understand how it feels.
For the most part Intel has used/abused its market dominance much more
benignly than, say, IBM did when it had the chance and certainly more
benignly than Microsoft does now.

RM

[Robert Redelmeier]

Robert Myers <rmyers@rustuck.com> wrote: What would AMD have to do to gain the confidence of the marketplace?

Which market? The OEMs (even informal) who actually buy chips,
or the consumers who buy systems?

The OEMs have to be looking carefully at eMachines overtaking
Gateway. Value obviously counts to the customer, even though
everyone [Dell] wants to reduce product variations for simpler
logistical and support.

Globally, perhaps 50% of PCs are white-box. These folks
defy all predictions and persist. Mostly by being carefully
attuned to their customers and flexible.

End users are even simpler. They buy AMD for value, and
Intel for MHz bragging rights (hopefully longer replacement
cycles) and a hope of better stability and support.

For instance, what combination of circumstances would give OEM's the breathing room to and the money to promote, say, Opteron, the way its technical merits suggest it ought to be promoted?

On 64 bit specifically, MS-Win64 would help. Beyond that,
OEM have to be waiting for Intel's response. If they pull
a rabbit out of their hats (unlikely), OEMs don't want to be
left stranded with a dead product line.

-- Robert

[Sebastian Kaliszewski]

"Yousuf Khan" <removethisspam.bjsk90.removethispam@hotmail.com> wrote in
message news:9lZIb.251985$ea%.240495@news01.bloor.is.net.cable.rogers.com... Rumours are that Intel is adding additional pipeline stages to the
Prescott, even higher than its existing 20 stages. I can't see it happening in this late stage of the game, perhaps half a year from now. It will possibly be getting rid of the internal double clocked ALUs too, and replacing it with two-stage single-clocked ALUs. http://www.neowin.net/comments.php?id=16264&category=main Yousuf Khan


It was already present in Intel's presentations quite some time ago -- It
was Intel presentation, where they said about more pipeline stages.

rgds
Sebastian

[Yousuf Khan]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:qmravvsmapchfp8p36brqbstegkfajrhsr@4ax.com... What would AMD have to do to gain the confidence of the marketplace? For instance, what combination of circumstances would give OEM's the breathing room to and the money to promote, say, Opteron, the way its technical merits suggest it ought to be promoted?

Sun seems to be getting ready to do just that right now. IBM & HP are still
taking the keep it quiet approach.
They know, like I know, that if Opteron became a real threat to Intel's market dominance, Intel would do whatever is necessary with money and marketing to change that. And woe unto the long term prospects of the OEM that got out of line.

You mean Intel isn't already doing that? Could've fooled me.
The other part of the story is that, for the moment at least, Intel needs AMD as a credible competitor they can point to so no one can accuse them of being a monopoly. Therefore, even given the opportunity, Intel would not crush AMD.

I find this theory a little hard to believe: that Intel could be that
generous towards AMD. If it were truly interested in keeping some
competition alive, then it wouldn't just promote one competitor. It would
attempt to foster multiple competitors with an equal marketshare as each
other, such as Transmeta and VIA. With AMD as big as it is, it becomes
dangerous. If AMD, VIA and Transmeta were as big as each other, then they'd
be busier fighting amongst themselves for the tablescraps, giving Intel
quite a bit of breathing room.
It's just not an emotional issue for me. Microsoft, I will admit, has become an emotional issue for me, so I can understand how it feels. For the most part Intel has used/abused its market dominance much more benignly than, say, IBM did when it had the chance and certainly more benignly than Microsoft does now.

I don't know, I know I'd feel Intel is much more benign when it's got 50%
marketshare instead of 80%. It's been done before, Airbus did it to Boeing,
Compaq did it to IBM.

Yousuf Khan

[Yousuf Khan]

"Sebastian Kaliszewski" <sk@bez.spamu.z.pl> wrote in message
news:bt435l$j0a$1@bozon2.softax.com.pl... It was already present in Intel's presentations quite some time ago -- It was Intel presentation, where they said about more pipeline stages.

What presentation was that?

Yousuf Khan

[George Macdonald]

On Fri, 02 Jan 2004 08:36:34 -0500, Robert Myers <rmyers@rustuck.com>
wrote:
On Fri, 02 Jan 2004 05:45:54 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:On Fri, 02 Jan 2004 07:00:54 GMT, "Yousuf Khan"<removethisspam.bjsk90.removethispam@hotmail.com> wrote:"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:6ql9vvcrtduv9edviol2cm7t6a3bpo00a9@4ax.com...> That's just the way it goes. When AMD figures _that_ part out, I> might invest in some AMD stock.You mean you'd invest in AMD stock the day that AMD figures out that it ishas all of the cards stacked against it, and that it should just stopcompeting against anything, and just stop producing products all together?:-)Doesn't sound like sound investment strategy to me, but hell either did theDOTCOMmerce strategy, but it worked for them for several years.Hmmm, I think it's another of Robert's Riddles.:-) I can't for the life ofme figure what AMD has to do to gain his confidence.It's the marketplace that's the riddle, not me. "IBM doesn't make thebest computers," it was once said, "they just sell the most of them."

Your apparent(?) suggestion that the purchase of AMD systems somehow
indicated a reluctance on the part of employees to "experiment" seemed like
a riddle to me and I still don't know what AMD has to actually do.
What would AMD have to do to gain the confidence of the marketplace?For instance, what combination of circumstances would give OEM's thebreathing room to and the money to promote, say, Opteron, the way itstechnical merits suggest it ought to be promoted?

So if AMD were rich they could indulge in the same marketing umm practices
as Intel?
They know, like I know, that if Opteron became a real threat toIntel's market dominance, Intel would do whatever is necessary withmoney and marketing to change that. And woe unto the long termprospects of the OEM that got out of line.

This is the status quo as far as Intel's money and marketing and greenmail
tactics. I believe that Opteron *is* a real threat.
The other part of the story is that, for the moment at least, Intelneeds AMD as a credible competitor they can point to so no one canaccuse them of being a monopoly. Therefore, even given theopportunity, Intel would not crush AMD.It's just not an emotional issue for me. Microsoft, I will admit, hasbecome an emotional issue for me, so I can understand how it feels.For the most part Intel has used/abused its market dominance much morebenignly than, say, IBM did when it had the chance and certainly morebenignly than Microsoft does now.

I see no benign here at all. One is/was as bad as the other, though IBM is
the one who was the best corporate citizen with their pure research
activities, much of which was freely available to the world for the asking.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[Robert Myers]

On Fri, 02 Jan 2004 18:12:05 -0500, George Macdonald
<fammacd=!SPAM^nothanks@tellurian.com> wrote:
On Fri, 02 Jan 2004 08:36:34 -0500, Robert Myers <rmyers@rustuck.com>wrote:On Fri, 02 Jan 2004 05:45:54 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:On Fri, 02 Jan 2004 07:00:54 GMT, "Yousuf Khan"<removethisspam.bjsk90.removethispam@hotmail.com> wrote:

<snip>
Hmmm, I think it's another of Robert's Riddles.:-) I can't for the life ofme figure what AMD has to do to gain his confidence.It's the marketplace that's the riddle, not me. "IBM doesn't make thebest computers," it was once said, "they just sell the most of them."Your apparent(?) suggestion that the purchase of AMD systems somehowindicated a reluctance on the part of employees to "experiment" seemed likea riddle to me and I still don't know what AMD has to actually do.

Did the statement, "Nobody ever got fired for buying IBM" require
explanation to you when it was at least mostly true? Right now, the
safe bet is to buy boxes from Dell that say "Intel Inside," because
that's what most customers do, and that's what the buzz says smart
customers do.

It doesn't matter if it's the best decision for people to make or not.
It's the safe decision, and people have too many other things to think
about. If it doesn't make a big difference to the bottom line, the
safe choice is the right choice, and Intel and Dell between them will
make sure that it doesn't make a big difference to the bottom line.
What would AMD have to do to gain the confidence of the marketplace?For instance, what combination of circumstances would give OEM's thebreathing room to and the money to promote, say, Opteron, the way itstechnical merits suggest it ought to be promoted?So if AMD were rich they could indulge in the same marketing umm practicesas Intel?

You don't think they would?
They know, like I know, that if Opteron became a real threat toIntel's market dominance, Intel would do whatever is necessary withmoney and marketing to change that. And woe unto the long termprospects of the OEM that got out of line.This is the status quo as far as Intel's money and marketing and greenmailtactics. I believe that Opteron *is* a real threat.

I'll bet Intel thinks it's a real threat, too, but I'm sure that their
response to it will be only partly technical. They'll do whatever
they can on the technical front (which may be limited; it's happened
before), and do the rest with pricing magic and marketing.

<snip>
It's just not an emotional issue for me. Microsoft, I will admit, hasbecome an emotional issue for me, so I can understand how it feels.For the most part Intel has used/abused its market dominance much morebenignly than, say, IBM did when it had the chance and certainly morebenignly than Microsoft does now.I see no benign here at all. One is/was as bad as the other, though IBM isthe one who was the best corporate citizen with their pure researchactivities, much of which was freely available to the world for the asking.

IBM is still a pretty good corporate customer, but they played to win
and they still play to win. IBM is in a head-to-head battle with
Microsoft over web services for coprorate environments at the moment,
and they don't very much care how what they do impacts everybody else.
Between them (Microsoft and IBM), the loser will be
platform-independent software.

AMD as a corporate customer? I have no way of knowing, because
they've had so little impact on my own little world that I'd have no
way of judging. For my own needs, which are technical computing,
Intel has been outstanding in terms of the support it makes available.
It plainly wants me and the kind of work I am interested in as a
customer.

Intel understands something that really gets my attention: lots of
numbers crunched plenty cheap. But isn't Opteron the winner in the
price/performance category for that kind of work right now? That's a
subject for another thread.

RM

[Stacey]

Robert Myers wrote:
On Fri, 02 Jan 2004 05:45:54 -0500, George Macdonald <fammacd=!SPAM^nothanks@tellurian.com> wrote:On Fri, 02 Jan 2004 07:00:54 GMT, "Yousuf Khan"<removethisspam.bjsk90.removethispam@hotmail.com> wrote:"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:6ql9vvcrtduv9edviol2cm7t6a3bpo00a9@4ax.com...> That's just the way it goes. When AMD figures _that_ part out, I> might invest in some AMD stock.You mean you'd invest in AMD stock the day that AMD figures out that itis has all of the cards stacked against it, and that it should just stopcompeting against anything, and just stop producing products alltogether?:-)Doesn't sound like sound investment strategy to me, but hell either didthe DOTCOMmerce strategy, but it worked for them for several years.Hmmm, I think it's another of Robert's Riddles.:-) I can't for the lifeof me figure what AMD has to do to gain his confidence. It's the marketplace that's the riddle, not me. "IBM doesn't make the best computers," it was once said, "they just sell the most of them." What would AMD have to do to gain the confidence of the marketplace? For instance, what combination of circumstances would give OEM's the breathing room to and the money to promote, say, Opteron, the way its technical merits suggest it ought to be promoted?

If AMD made their own chipsets and mobo's. What killed AMD from main stream
use was the flakey via chipsets in the early days and that stigma is still
with them.
--

Stacey

[Spam Me Please]

>>>>> "stacey" == stacey <fotocord@yahoo.com> writes:

stacey> If AMD made their own chipsets and mobo's. What killed AMD
stacey> from main stream use was the flakey via chipsets in the early
stacey> days and that stigma is still with them. --

This is true. VIA left a very bad impression at least for me and the
really strange bugs the early via chipsets for AMD. I have said that I
will never buy another via motherboard, but I may in the future. It is
surprising if you believe the web sites that the best amd64
motherboards have the via chipset. I guess via has learned some
lessons.

Whatever.

[Stacey]

Spamme Now wrote:
>>> "stacey" == stacey <fotocord@yahoo.com> writes: stacey> If AMD made their own chipsets and mobo's. What killed AMD stacey> from main stream use was the flakey via chipsets in the early stacey> days and that stigma is still with them. -- This is true. VIA left a very bad impression at least for me and the really strange bugs the early via chipsets for AMD. I have said that I will never buy another via motherboard, but I may in the future. It is surprising if you believe the web sites that the best amd64 motherboards have the via chipset. I guess via has learned some lessons.


My guess is they just haven't found the bugs yet. Some of those early Via
boards pulled great benchmark numbers, but were flakey as hell in real use.
I'll -never- use another Via board!

Even these Nvidia chipsets people rave about seem to have flakey issues when
used with video editing cards? You can blame the cards all you want but
when they work fine on an intel board and flake out on a Via/Nvidia, where
is the problem and who is going to fix it? You think they are going to
design cards around non-intel chipsets?

--

Stacey

[Yousuf Khan]

"Spamme Now" <spamme@edge.net> wrote in message
news:87d6a1smrp.fsf@spamme.edge.net... This is true. VIA left a very bad impression at least for me and the really strange bugs the early via chipsets for AMD. I have said that I will never buy another via motherboard, but I may in the future. It is surprising if you believe the web sites that the best amd64 motherboards have the via chipset. I guess via has learned some lessons.

I bought my first VIA chipset motherboard based on the recommendations of a
website (early Tom's Hardware, as a matter of fact). The reviews assured me
that all of the features that VIA had, that there was no way I could go
wrong. Little did I know at that time that no amount of features on a piece
of paper actually made up for a bad implementation in an actual product.

I bought a few more VIAs thinking that my first experience was just a fluke;
I thought that VIA must've surely learned from their early mistakes and
corrected them. I was beginning to think there was never a hope of finding a
stable system, then I got a SIS, and I will never look back.

I've had a similar experience with ATI video chipsets. The basic chipsets
themselves are just fine, it's the bonebrained software writers who design
the drivers for them that should be tossed to the lowest reaches of hell. I
still remember the drivers that kept getting buggier the newer they got. But
each successive video driver did come with a lot of useless convenience
features.

Yousuf Khan

[Tony Hill]

On Thu, 01 Jan 2004 21:37:42 -0500, Robert Myers <rmyers@rustuck.com>
wrote:On Thu, 1 Jan 2004 20:32:26 -0500, Keith R. Williams<krw@attglobal.net> wrote:<snip>I suppose ISVs enjoy coding around the lack of the barrel shifterand integer multiplier too?Yada, indeed.When the VP in Charge of Dimwittery wanders by to ask why the "IntelInside" label on boxes is slowly being replaced by this weird Greenlabel that says AMD, you might want to skip the stack depth, barrelshifter, and integer multiplier and tell him that some assistant whohas long ago been sacked made the mistake and you'll see to it that noone does anything that foolish again.

Or you say that the new systems are saving the company money both in
up front costs and by reducing maintenance costs through their better
driver support (from nVidia, not AMD, but the boss probably doesn't
care much about that). Money talks, and even the VP in Charge of
Dimwittery listens.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

[The little lost angel]

On Sat, 03 Jan 2004 07:23:13 GMT, "Yousuf Khan"
<removethisspam.bjsk90.removethispam@hotmail.com> wrote:
I bought a few more VIAs thinking that my first experience was just a fluke;I thought that VIA must've surely learned from their early mistakes andcorrected them. I was beginning to think there was never a hope of finding astable system, then I got a SIS, and I will never look back.

Well, I got my first VIA/FIC because of THG too. But it worked fine
for me even when I did funny things to the voltage selectors :pPpP

Similarly, my next VIA/AOpen board worked fine for me too. So is my
current board VIA/MSI. Then again, I had few problems with 98SE too...
:P

I agree with you on the crap ATI drivers however :PppP

--
L.Angel: I'm looking for web design work.
If you need basic to med complexity webpages at affordable rates, email me :)
Standard HTML, SHTML, MySQL + PHP or ASP, Javascript.
If you really want, FrontPage & DreamWeaver too.
But keep in mind you pay extra bandwidth for their bloated code

[George Macdonald]

On Fri, 02 Jan 2004 19:27:36 -0500, Robert Myers <rmyers@rustuck.com>
wrote:
On Fri, 02 Jan 2004 18:12:05 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:<snip>
It's the marketplace that's the riddle, not me. "IBM doesn't make thebest computers," it was once said, "they just sell the most of them."Your apparent(?) suggestion that the purchase of AMD systems somehowindicated a reluctance on the part of employees to "experiment" seemed likea riddle to me and I still don't know what AMD has to actually do.Did the statement, "Nobody ever got fired for buying IBM" requireexplanation to you when it was at least mostly true? Right now, thesafe bet is to buy boxes from Dell that say "Intel Inside," becausethat's what most customers do, and that's what the buzz says smartcustomers do.

Well back in those days I used to cheer for CDC, Univac less and
Burroughs... was a furniture mfr.:-) It seems then that Yousuf was right -
you think that what AMD should do is just find another line of work?
It doesn't matter if it's the best decision for people to make or not.It's the safe decision, and people have too many other things to thinkabout. If it doesn't make a big difference to the bottom line, thesafe choice is the right choice, and Intel and Dell between them willmake sure that it doesn't make a big difference to the bottom line.

Hmmm you may have hit on something there which hadn't dawned on me before:
(Michael) Dell is the driving force behind Yamhill????

It took a while but eventually, IBM was chased out of the PC market - it
can happen again and Dell is no IBM... but then again the PC could
disappear too.

Dell got a lot of mileage out of the dude character in their advertising
for the home market; I can't figure who their latest effort, with the
bullying sgt. major, is aimed at - could be a misstep. Telephone support
from India - another misstep. How many can they afford?
What would AMD have to do to gain the confidence of the marketplace?For instance, what combination of circumstances would give OEM's thebreathing room to and the money to promote, say, Opteron, the way itstechnical merits suggest it ought to be promoted?So if AMD were rich they could indulge in the same marketing umm practicesas Intel?You don't think they would?

Sure they would... if they had the money.
They know, like I know, that if Opteron became a real threat toIntel's market dominance, Intel would do whatever is necessary withmoney and marketing to change that. And woe unto the long termprospects of the OEM that got out of line.This is the status quo as far as Intel's money and marketing and greenmailtactics. I believe that Opteron *is* a real threat.I'll bet Intel thinks it's a real threat, too, but I'm sure that theirresponse to it will be only partly technical. They'll do whateverthey can on the technical front (which may be limited; it's happenedbefore), and do the rest with pricing magic and marketing.

As far as I see Intel has already done the technical response... with
Prescott. Have you looked at Hans de Vries' articles?
<snip>IBM is still a pretty good corporate customer, but they played to winand they still play to win. IBM is in a head-to-head battle withMicrosoft over web services for coprorate environments at the moment,and they don't very much care how what they do impacts everybody else.Between them (Microsoft and IBM), the loser will beplatform-independent software.

I bet Larry E. is quaking in his boots.</sarcasm off>
AMD as a corporate customer? I have no way of knowing, becausethey've had so little impact on my own little world that I'd have noway of judging. For my own needs, which are technical computing,Intel has been outstanding in terms of the support it makes available.It plainly wants me and the kind of work I am interested in as acustomer.Intel understands something that really gets my attention: lots ofnumbers crunched plenty cheap. But isn't Opteron the winner in theprice/performance category for that kind of work right now? That's asubject for another thread.

It certainly looks like 2004 is going to be very interesting.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[George Macdonald]

On Fri, 02 Jan 2004 23:18:42 -0500, stacey <fotocord@yahoo.com> wrote:
Spamme Now wrote:>>>> "stacey" == stacey <fotocord@yahoo.com> writes: stacey> If AMD made their own chipsets and mobo's. What killed AMD stacey> from main stream use was the flakey via chipsets in the early stacey> days and that stigma is still with them. -- This is true. VIA left a very bad impression at least for me and the really strange bugs the early via chipsets for AMD. I have said that I will never buy another via motherboard, but I may in the future. It is surprising if you believe the web sites that the best amd64 motherboards have the via chipset. I guess via has learned some lessons.My guess is they just haven't found the bugs yet. Some of those early Viaboards pulled great benchmark numbers, but were flakey as hell in real use.I'll -never- use another Via board!

I've never had any trouble with VIA based systems - no more than with some
of Intel's earlier chipsets... but I never tried the VIAs with any
specialty add-in cards. Note also that many of the VIA "issues" were as
much to do with inept users as anything else.
Even these Nvidia chipsets people rave about seem to have flakey issues whenused with video editing cards?

A fairly thorough search at Google doesn't turn up anybody raving about
"flakey issues" with nForce(x) chipsets for video editing. You're on the
record as proclaiming that the same chipsets do not support IDE
busmastering when they plainly do!
You can blame the cards all you want butwhen they work fine on an intel board and flake out on a Via/Nvidia, whereis the problem and who is going to fix it? You think they are going todesign cards around non-intel chipsets?

This has been beaten to death here in the past but there's more to making a
chipset/mbrd than a specification - every implementation has its quirks.
The volume card mfrs seem to have pretty much accomodated the VIA quirks as
well as the Intel ones; there *is* the patch for VIA to do with truncated
PCI transfers - never had to use it myself. For nVidia, I think you need
to show more evidence or withdraw that statement.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[Robert Myers]

On Sun, 04 Jan 2004 02:39:06 -0500, George Macdonald
<fammacd=!SPAM^nothanks@tellurian.com> wrote:
On Fri, 02 Jan 2004 19:27:36 -0500, Robert Myers <rmyers@rustuck.com>wrote:On Fri, 02 Jan 2004 18:12:05 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:

<snip>
Well back in those days I used to cheer for CDC, Univac less andBurroughs... was a furniture mfr.:-) It seems then that Yousuf was right -you think that what AMD should do is just find another line of work?

They're pursuing the only obvious corporate strategy open to them,
which is to build on the x86 franchise. Yousuf didn't take me
seriously when I said it, but Intel _needs_ a credible competitor
(something that apparently didn't seep into the all-encompassing brain
of Bill G.).

I don't think there is room for Intel, AMD, and IBM as major players.
Intel's real goal in life is to persuade IBM, not AMD, to get out of
the business. That way, Intel will have two of the best marketers of
boxes (Dell and IBM) selling boxes with its chips, and it really won't
have to care all that much what AMD does.

The x86 franchise has a ceiling (it won't be a mainframe chip), and
the rest of the world is going to figure out how to compete for the
lower end of the market. AMD is destined to live out its life as the
Chrysler Corporation.
It doesn't matter if it's the best decision for people to make or not.It's the safe decision, and people have too many other things to thinkabout. If it doesn't make a big difference to the bottom line, thesafe choice is the right choice, and Intel and Dell between them willmake sure that it doesn't make a big difference to the bottom line.Hmmm you may have hit on something there which hadn't dawned on me before:(Michael) Dell is the driving force behind Yamhill????

Maybe. Between Intel and Dell, they should certainly have the
resources to figure out what they absolutely must do to sell boxes.
It took a while but eventually, IBM was chased out of the PC market - itcan happen again and Dell is no IBM... but then again the PC coulddisappear too.

In its current incarnation, I think it will. Who is to say that Sony
(with IBM's technology) isn't the bigger threat to both Intel and AMD
in the home market?

<snip>
IBM is still a pretty good corporate customer, but they played to winand they still play to win. IBM is in a head-to-head battle withMicrosoft over web services for coprorate environments at the moment,and they don't very much care how what they do impacts everybody else.Between them (Microsoft and IBM), the loser will beplatform-independent software.I bet Larry E. is quaking in his boots.</sarcasm off>

So the losers might be Microsoft, IBM, _and_ platform-independent
software.
It certainly looks like 2004 is going to be very interesting.

Couldn't help but be an improvement on 2003.

RM

[Robert Redelmeier]

George Macdonald <fammacd=!SPAM^nothanks@tellurian.com> wrote: I've never had any trouble with VIA based systems - no more than with some of Intel's earlier chipsets... but I never tried the VIAs with any specialty add-in cards. Note also that many of the VIA "issues" were as much to do with inept users as anything else.

This matches my experience. Used lots systems, and never
had any trouble traceable to the VIA chipset. And I test hard.

There was a problem a while back that many people attributed to
VIA interaction with a sound-card. Around the same time, I had
HD read errors that turned out to be a disk that wouldn't tolerate
4.73V (slightly offspec). Replaced the PSU and all was fine.

PEBKAC includes misdiagnosis.

-- Robert author `cpuburn` moving to http://pages.sbcglobal.net/redelm

[Guest]

Robert Myers <rmyers@rustuck.com> writes:
On Sun, 04 Jan 2004 02:39:06 -0500, George Macdonald <fammacd=!SPAM^nothanks@tellurian.com> wrote:On Fri, 02 Jan 2004 19:27:36 -0500, Robert Myers <rmyers@rustuck.com>wrote:On Fri, 02 Jan 2004 18:12:05 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote: <snip>Well back in those days I used to cheer for CDC, Univac less andBurroughs... was a furniture mfr.:-) It seems then that Yousuf was right -you think that what AMD should do is just find another line of work? They're pursuing the only obvious corporate strategy open to them, which is to build on the x86 franchise. Yousuf didn't take me seriously when I said it, but Intel _needs_ a credible competitor (something that apparently didn't seep into the all-encompassing brain of Bill G.). I don't think there is room for Intel, AMD, and IBM as major players. Intel's real goal in life is to persuade IBM, not AMD, to get out of the business. That way, Intel will have two of the best marketers of boxes (Dell and IBM) selling boxes with its chips, and it really won't have to care all that much what AMD does. The x86 franchise has a ceiling (it won't be a mainframe chip), and
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

What is a "mainframe" anyway? x86 is already being used in a lot of stuff,
like high end workstations, webservers, clusters, etc. What is left of the
market? It seems to me that Opteron will break the "4G" memory barrier, which
is one of the main things holding big applications back. That, combined with
large IDE RAID disk arrays should do it. What else do we need? The other
thing worth noting is the development of 3D graphics chips. And that is all
happening on x86 now. I can't see any need for SGI or Sun, unless they are
price competitive with PCs.

the rest of the world is going to figure out how to compete for the lower end of the market. AMD is destined to live out its life as the Chrysler Corporation.It doesn't matter if it's the best decision for people to make or not.It's the safe decision, and people have too many other things to thinkabout. If it doesn't make a big difference to the bottom line, thesafe choice is the right choice, and Intel and Dell between them willmake sure that it doesn't make a big difference to the bottom line.Hmmm you may have hit on something there which hadn't dawned on me before:(Michael) Dell is the driving force behind Yamhill???? Maybe. Between Intel and Dell, they should certainly have the resources to figure out what they absolutely must do to sell boxes.It took a while but eventually, IBM was chased out of the PC market - itcan happen again and Dell is no IBM... but then again the PC coulddisappear too. In its current incarnation, I think it will. Who is to say that Sony (with IBM's technology) isn't the bigger threat to both Intel and AMD in the home market? <snip>IBM is still a pretty good corporate customer, but they played to winand they still play to win. IBM is in a head-to-head battle withMicrosoft over web services for coprorate environments at the moment,and they don't very much care how what they do impacts everybody else.Between them (Microsoft and IBM), the loser will beplatform-independent software.I bet Larry E. is quaking in his boots.</sarcasm off> So the losers might be Microsoft, IBM, _and_ platform-independent software.It certainly looks like 2004 is going to be very interesting. Couldn't help but be an improvement on 2003. RM

[daytripper]

On 04 Jan 2004 14:47:14 -0500, Mannr@uwaterloo.ca wrote:What is a "mainframe" anyway? x86 is already being used in a lot of stuff,like high end workstations, webservers, clusters, etc. What is left of themarket? It seems to me that Opteron will break the "4G" memory barrier, whichis one of the main things holding big applications back. [snipped]

Whoa, Nellie!

And precisely what '"4G" memory barrier' would that be, again?

/daytripper (And why wasn't I told about this before? ;-)

[Yousuf Khan]

<Mannr@uwaterloo.ca> wrote in message
news:yua8ykn4htp.fsf@tapir.uwaterloo.ca... What is a "mainframe" anyway? x86 is already being used in a lot of
stuff, like high end workstations, webservers, clusters, etc. What is left of
the market? It seems to me that Opteron will break the "4G" memory barrier,
which is one of the main things holding big applications back. That, combined
with large IDE RAID disk arrays should do it. What else do we need? The other thing worth noting is the development of 3D graphics chips. And that is
all happening on x86 now. I can't see any need for SGI or Sun, unless they
are price competitive with PCs.

It's not just the amount of memory a processor can address, nor whether it
has RAID disks. And it has nothing to do with 3D graphics chips. Mainframes
are first and foremost, batch job processing units. They should be able to
run tons of batch jobs silently in the background, with minimal user
interactivity. In fact, with mainframes, they farm the job of user
interaction out to slave processors, whcih free up the CPU as much as
possible for batch processing.

Some of the Unix supersystems out there now, such as the HP Superdome, or
the Sun Starcat F15K, which are able to be partitioned out into multiple
independent sub-servers, are getting close to mainframe functionality
themselves.

So I don't see why x86 can't become a mainfame processor, like Robert has
said, it's simply a matter of adding hardware.

Yousuf Khan

[Robert Myers]

On 04 Jan 2004 14:47:14 -0500, Mannr@uwaterloo.ca wrote:
Robert Myers <rmyers@rustuck.com> writes:

<snip>
The x86 franchise has a ceiling (it won't be a mainframe chip), and ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^What is a "mainframe" anyway? x86 is already being used in a lot of stuff,like high end workstations, webservers, clusters, etc. What is left of themarket? It seems to me that Opteron will break the "4G" memory barrier, whichis one of the main things holding big applications back. That, combined withlarge IDE RAID disk arrays should do it. What else do we need? The otherthing worth noting is the development of 3D graphics chips. And that is allhappening on x86 now. I can't see any need for SGI or Sun, unless they areprice competitive with PCs.

<snip>

Perhaps you should ask someone who knows more about mainframes and
about x86 at the bare iron level than I do. Since daytripper saw the
question and didn't answer, I don't know that anyone better suited to
answer the question will.

What mainframes are still sold are sold on the premise that they can
be partitioned and virtualized to run multiple operating systems as if
they were multiple machines. Intel regularly pitches IBM on how it
could meet those needs with Itanium should IBM see the light and
decide that it could better use its resources elsewhere than in trying
to compete with Intel.

Not to say that someone couldn't fix it somehow, but the current x86
architecture just does not lend itself to being virtualized in the way
that a mainframe would require.

IBM, SGI, and HP _already_ sell big cache-coherent boxes built on
Itania. I woudln't hold my breath waiting for such boxes aimed at the
OLTP market to appear (technical computing might be another story).

As to high-end workstations, you have an odd notion of what high-end
means. High end means $7K+, and that's a market that x86 has not
penetrated but that Itanium has (albeit with mixed reviews).

The 4G "memory barrier" is purely a programming issue. From the POV
of the end user, it does not exist.

RM

[Yousuf Khan]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:7k5gvvgbbvbl25ld4hnqvdlpensf0sqsif@4ax.com... They're pursuing the only obvious corporate strategy open to them, which is to build on the x86 franchise.

Well, why not? Overall, there is nothing wrong with x86 as an architecture.
They're fixing some of the most obvious remaining problems with it, such as
the lack of GP registers, and cleaning out the last vestiges of
segment-based memory management (although segments had good purposes too).
They're even updating the I/O bus scheme, such as Hypertransport.
I don't think there is room for Intel, AMD, and IBM as major players. Intel's real goal in life is to persuade IBM, not AMD, to get out of the business. That way, Intel will have two of the best marketers of boxes (Dell and IBM) selling boxes with its chips, and it really won't have to care all that much what AMD does.

IBM is hardly a big player in PCs anymore.
The x86 franchise has a ceiling (it won't be a mainframe chip), and the rest of the world is going to figure out how to compete for the lower end of the market. AMD is destined to live out its life as the Chrysler Corporation.

Not sure why x86 can't be a mainframe chip. Some previous mainframe chips
were hardly standout chips in their own right. Besides the whole Cray Red
Storm project has the smell of a large-scale mainframe to me. It has
separate i/o nodes, and processing nodes, etc. Cray is even turning it into
a commercial venture, which it calls Strider, where mini-Red Storms will be
sold commercially.
Maybe. Between Intel and Dell, they should certainly have the resources to figure out what they absolutely must do to sell boxes.

It seems to me that Dell (along with others) believes the best way for it to
sell boxes is to sell a different kind of box -- namely tv sets.

Yousuf Khan

[Felger Carbon]

"George Macdonald" <fammacd=!SPAM^nothanks@tellurian.com> wrote in
message news:a5jevvs5rkvojibj5tgc6rnbd528s52spa@4ax.com... (Michael) Dell is the driving force behind Yamhill????

I realize that that was a joke, George. But most companies pay
attention to their largest customer - and Dell is in fact Intel's
largest (and most loyal?) customer.

[Felger Carbon]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:7k5gvvgbbvbl25ld4hnqvdlpensf0sqsif@4ax.com... The x86 franchise has a ceiling (it won't be a mainframe chip), and the rest of the world is going to figure out how to compete for the lower end of the market.

<assume tongue-in-cheek mode>

Right. The x86 can't compete with Motorola's 68K or Nat Semi's 32K.

And the x86 can't compete with them super RISC chips like Alpha and
the 29K.

And the x86 can't compete with the super-good, modern Itanic. Intel
won't let it. Yet.

And ***of course*** the x86 will never become the heart of a
mainframe, economics or no economics!

<exit tongue-in-cheek mode>

[Robert Myers]

On Sun, 04 Jan 2004 21:01:50 GMT, "Felger Carbon" <fmsfnf@jfoops.net>
wrote:
"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:7k5gvvgbbvbl25ld4hnqvdlpensf0sqsif@4ax.com... The x86 franchise has a ceiling (it won't be a mainframe chip), and the rest of the world is going to figure out how to compete for the lower end of the market.<assume tongue-in-cheek mode>Right. The x86 can't compete with Motorola's 68K or Nat Semi's 32K.And the x86 can't compete with them super RISC chips like Alpha andthe 29K.And the x86 can't compete with the super-good, modern Itanic. Intelwon't let it. Yet.And ***of course*** the x86 will never become the heart of amainframe, economics or no economics!<exit tongue-in-cheek mode>
The processor is such a small part of the economics of a mainframe
that "the processor is less expensive" doesn't really cut it. You
know that.

The processor is just one piece of the puzzle. You will often hear
people say, "If only AMD had it's own compiler." Well, they don't,
and that's part of the point. They don't have a compiler not because
they don't think it would be helpful, but because they just don't have
the money. Not only don't they have a compiler, they don't have lots
of other tools and goodies that Intel can provide to its ISV's (even
money, if the need is great enough). Life just isn't fair.

As far as I can tell, Alpha was the most promising microarchitecture
ever. Intel killed it, not because they thought Itanium could do
better, but because Intel has a strategy built around the Itanium
instruction set. I'm handicapping the future based on my own
understanding of money and markets, not on my understanding of
technology. If technology were the decisive factor, DEC would still
be in business.

RM

[daytripper]

On Sun, 04 Jan 2004 15:27:40 -0500, Robert Myers <rmyers@rustuck.com> wrote:
On 04 Jan 2004 14:47:14 -0500, Mannr@uwaterloo.ca wrote:Robert Myers <rmyers@rustuck.com> writes:<snip> The x86 franchise has a ceiling (it won't be a mainframe chip), and ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^What is a "mainframe" anyway? x86 is already being used in a lot of stuff,like high end workstations, webservers, clusters, etc. What is left of themarket? It seems to me that Opteron will break the "4G" memory barrier, whichis one of the main things holding big applications back. That, combined withlarge IDE RAID disk arrays should do it. What else do we need? The otherthing worth noting is the development of 3D graphics chips. And that is allhappening on x86 now. I can't see any need for SGI or Sun, unless they areprice competitive with PCs.<snip>Perhaps you should ask someone who knows more about mainframes andabout x86 at the bare iron level than I do. Since daytripper saw thequestion and didn't answer, I don't know that anyone better suited toanswer the question will.

Sorry, I got lost in the mythical "4GB limit" thing, lost sight of the ball
;-)
What mainframes are still sold are sold on the premise that they canbe partitioned and virtualized to run multiple operating systems as ifthey were multiple machines. Intel regularly pitches IBM on how itcould meet those needs with Itanium should IBM see the light anddecide that it could better use its resources elsewhere than in tryingto compete with Intel.Not to say that someone couldn't fix it somehow, but the current x86architecture just does not lend itself to being virtualized in the waythat a mainframe would require.

I'm curious - the existence of partitionable implementations would not fit my
perception of what (still) *drives* mainframes. Seems like that puts the cart
before the horse...
IBM, SGI, and HP _already_ sell big cache-coherent boxes built onItania. I woudln't hold my breath waiting for such boxes aimed at theOLTP market to appear (technical computing might be another story).As to high-end workstations, you have an odd notion of what high-endmeans. High end means $7K+, and that's a market that x86 has notpenetrated but that Itanium has (albeit with mixed reviews).

You could be right about that, but me and my Parhelia triple 21" headed dual
3.2 ghz Xeon with 4GB of DDR333 and the u320 10krpm raid and dual gigabit
hoses are sorely hurt anyway.
We're cheap - certainly under $7K. But we're not easy ;-)

So...what are we doing with these mythical high-end boxen of which you speak -
real time very large field 3D visualization of WMD tests?

Oh - wait!
The Patriot Act requires me to ask: do you carry an almanac?
No?
Ok. Sorry about the interruption ;-)
The 4G "memory barrier" is purely a programming issue. From the POVof the end user, it does not exist.

We have a winner.

----

I am sure there are more qualified people to set the boundary between
"mainframe" and "something else that's wicked fast and less expensive". Since
the late 80's the line blurred from my perspective - I mean, compared to an
ES9000, was an 8 processor Aquarius a mainframe or a "super mini" - and if the
latter, what exactly is *that*? Is a 32 processor EV68 a mainframe or
"something else"?

The "marketing of continuity" by the very few companies still in the
"mainframe" business is its ever-diminishing grace. That, and a whole lot of
give-backs, you can bet your bippy on that.

The development of partitionable implementations was simply a defensive form
of life support, temporary succor for a segment that will, for sure, go away
within this decade, as the hardware gets ever cheaper, and the last instances
of hoary dust-encrusted single-image NeglectWare (you know who you are) are
finally rewritten for the 21st century.

One can stack quad Xeon bricks, each with 16GB of memory, a handful of
giga-nics and some scary fast disk arrays, and generate WAY more computes per
Kilobuck than can be found in the Big Iron lot. And virtualization can work to
manage huge and relatively cheap server farms as one coherent "something
else..." if that's your bag.

With open source operating systems rapidly evolving to where they can be found
in the same glass-walled rooms next to $proprietary $code running on
$proprietary $architectures, and the ability to virtualize one hair-raisingly
powerful beast with parts from rather lowly origins, I'd say classic
mainframes - if they still exist - ought to be in Death Watch mode.

Once the weasels in expensive suits stop throwing money down the tubes at
their alleged "sure thing" and rewrite their mission-critical code to bring it
up a few decades, it's Adios Amigo to the last rusty dinosaurs...

/daytripper (Google this one sometime in 2010 ;-)

[Nate Edel]

Robert Myers <rmyers@rustuck.com> wrote: seriously when I said it, but Intel _needs_ a credible competitor (something that apparently didn't seep into the all-encompassing brain of Bill G.).

Microsoft has apparantly assumed Apple would work that way in the past. ISTR
Microsoft making a large-ish ($150M) investment in Apple around 1998 or so,
when Steve Jobs came back.

--
Nate Edel http://www.nkedel.com/

"I say we take off and nuke the entire site from orbit. That's the only way
to be sure." -- Ripley, _Aliens_

[daytripper]

On Sun, 4 Jan 2004 15:11:45 -0800, archmage@sfchat.org (Nate Edel) wrote:
Robert Myers <rmyers@rustuck.com> wrote: seriously when I said it, but Intel _needs_ a credible competitor (something that apparently didn't seep into the all-encompassing brain of Bill G.).Microsoft has apparantly assumed Apple would work that way in the past. ISTRMicrosoft making a large-ish ($150M) investment in Apple around 1998 or so,when Steve Jobs came back.

Context is everything: Microsoft made that investment as a legal strategy, as
they were neck deep in court for antitrust/predatory tactics/etc, and keeping
Apple alive gave them an almost plausible defense against monopoly charges...

/daytripper

[Robert Myers]

On Sun, 04 Jan 2004 22:37:32 GMT, daytripper
<day_trippr@REMOVEyahoo.com> wrote:
On Sun, 04 Jan 2004 15:27:40 -0500, Robert Myers <rmyers@rustuck.com> wrote:On 04 Jan 2004 14:47:14 -0500, Mannr@uwaterloo.ca wrote:Robert Myers <rmyers@rustuck.com> writes:

<snip>I'm curious - the existence of partitionable implementations would not fit myperception of what (still) *drives* mainframes. Seems like that puts the cartbefore the horse...
Could be just my personal sampling bias, but every time I read about a
"mainframe" being sold, and occasionally when I hear someone let slip
that they actually spent money on one of those things and are using
it, I always hear server consolidation (running several servers on the
same box) and the resulting ease of administration mentioned as a
major issue. If I *can't* partition, I run the risk of having a very
big box brought down by some subset that's having trouble.

The ability to virtualize is apparently one of the things that still
sells z-series boxes. Since z-series boxes are one of the most
profitable items in the business, I find it hard to believe that
anyone would build a mainframe line around a chip that can't be
partitioned cleanly, there being very little point in having a "value"
mainframe series and a "full-featured" mainframe series.

Intel hasn't said it, but since they know that some unbelievable level
of binary compatibility is what IBM is still able to offer, I believe
that the ability to virtualize Itanium into (say) a Power chip has
been a part of their plans from day one. I don't know that Transmeta
didn't start its business with the intent of selling or licensing the
technology to Intel, knowing full well that Intel had an intent to go
there. It could be _just_ a coincidence, that Transmeta is
virtualizing a VLIW core, you know.

<snip>
As to high-end workstations, you have an odd notion of what high-endmeans. High end means $7K+, and that's a market that x86 has notpenetrated but that Itanium has (albeit with mixed reviews).You could be right about that, but me and my Parhelia triple 21" headed dual3.2 ghz Xeon with 4GB of DDR333 and the u320 10krpm raid and dual gigabithoses are sorely hurt anyway.We're cheap - certainly under $7K. But we're not easy ;-)So...what are we doing with these mythical high-end boxen of which you speak -real time very large field 3D visualization of WMD tests?
I wouldn't be surprised to find that there are more than a few Sun
Sparc and SGI workstations hidden behind the fences at US WMD
factories, but the consumers I had in mind and know about are people
doing what people imagine engineers still doing: designing,
simulating, and (still very compute intensive) 3-D visualization:
Aerospace, GE, GM, people doing oil exploration. I assume there's a
market in the medical imaging business, but I don't in fact know what
people use.

The ability to run a single thread very fast is still important in
those applications because that's how many of those applications are
still written. If people can't figure out (or are unwilling to figure
out) how to write multi-threaded code, the only thing left is for the
processor to do it for them, and Itanium is halfway down that road
already.

The big iron people in comp.arch sometimes ask, "Who cares about
single thread performance, anyway." That is to say, of what relevance
are the eye-popping SPECfp scores that Itanium has been able to
produce? They are of great relevance in situations where you have an
expensive engineer sitting there waiting for single-threaded code to
produce a result--the workstation market, in other words.

<snip>One can stack quad Xeon bricks, each with 16GB of memory, a handful ofgiga-nics and some scary fast disk arrays, and generate WAY more computes perKilobuck than can be found in the Big Iron lot. And virtualization can work tomanage huge and relatively cheap server farms as one coherent "somethingelse..." if that's your bag.
People who want cheap server farms are already buying them. One thing
that IBM can sell that most people can't is that they have decades of
experience with automating tasks where people can make mistakes,
operator error being a more common cause of system failure than
hardware failures in high-end systems.
With open source operating systems rapidly evolving to where they can be foundin the same glass-walled rooms next to $proprietary $code running on$proprietary $architectures, and the ability to virtualize one hair-raisinglypowerful beast with parts from rather lowly origins, I'd say classicmainframes - if they still exist - ought to be in Death Watch mode.
They still exist. IBM still sells them. Fujitsu just announced a new
line--built around Itania.

RM

[Stacey]

George Macdonald wrote:
On Fri, 02 Jan 2004 23:18:42 -0500, stacey <fotocord@yahoo.com> wrote:
Even these Nvidia chipsets people rave about seem to have flakey issueswhen used with video editing cards? A fairly thorough search at Google doesn't turn up anybody raving about "flakey issues" with nForce(x) chipsets for video editing. You're on the record as proclaiming that the same chipsets do not support IDE busmastering when they plainly do!

I was looking for a specific driver but discovered they work fine with the
default one in win98with the right settings, I made a mistake?
You can blame the cards all you want butwhen they work fine on an intel board and flake out on a Via/Nvidia, whereis the problem and who is going to fix it? You think they are going todesign cards around non-intel chipsets? For nVidia, I think you need to show more evidence or withdraw that statement.

http://www.videoguys.com/system.htm

These guys sell lots of real time video cards and had this to say on their
site about chipsets:

Recommended motherboards

WARNING: We do NOT recommend motherboards based on the nVidia chipsets. We
are running into a wide range of issues with them. While they may be OK for
gaming, they are not for NLE.

VIA Chipsets
We don't really recommend these chipsets, because of PCI bottelnecks. If
your motherboard uses the VIA chipset, it is important that you download
and install the latest 4-in-1 driver, version 4.23 or later. You can
download this from your motherboard vendor's sight. Note: Installing the
latest 4-in-1 driver will improve your system performance, but it may not
solve your NLE compatibility issues. If you have a choice, do not get a VIA
chipset.

I suppose you can work around the issues they have buy why fight with them?
I'm glad you've -never- had a problem with a Via based system but I don't
think they are worth fooling with myself from my past experience. I build
some video NLE systems and the intel ones never have the problems that the
via systems I've worked on (gave up on?) have.


--

Stacey

[Tony Hill]

On Sun, 04 Jan 2004 09:01:01 -0500, Robert Myers <rmyers@rustuck.com>
wrote:The x86 franchise has a ceiling (it won't be a mainframe chip), and

No, x86 probably won't be a mainframe chip because mainframe chips
have become a REAL niche market. There's no money in making mainframe
hardware, only in mainframe software and support. Neither AMD or
Intel want to get into that business.
the rest of the world is going to figure out how to compete for thelower end of the market. AMD is destined to live out its life as theChrysler Corporation.

So you meant they'll become one of the world's largest suppliers of
top-end processors (eg Mercedes), BIG servers for those that need to
move lots of data (eg Freightliner trucks), as well as selling some of
the most popular household processors in North America (eg the Dodge
Ram and Caravan), not to mention leading the way in ultra-portable
chips (eg the Smart cars that are popular in Europe).

Hmm.. I think AMD would be quite happy to live out it's life as the
DaimlerChrysler Corporation, especially if that meant that Intel was
Ford! :>
Hmmm you may have hit on something there which hadn't dawned on me before:(Michael) Dell is the driving force behind Yamhill????Maybe. Between Intel and Dell, they should certainly have theresources to figure out what they absolutely must do to sell boxes.

It shouldn't take all that many resources. They need to focus on what
customers need and want, rather than on what Intel wants. This seems
to have been Intel's main problem in recent years, they've
concentrated so much on what they think is an ideal design rather than
listening to what end users actually want. Of course, there are some
signs that this is changing, case-in-point being the Pentium M. This
is exactly what people were looking for in a notebook processor.
Fairly fast chip but with decently low power consumption.

The Itanium, however, is not what customers want or need. Even with
things like SSE (SSE2/SSE3) Intel seems to have missed the mark
somewhat. They've got all sorts of instructions in there that seem to
have little use, but then they left out some features that would seem
to be quite important.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

[George Macdonald]

On Sun, 04 Jan 2004 16:27:16 -0500, Robert Myers <rmyers@rustuck.com>
wrote:
On Sun, 04 Jan 2004 21:01:50 GMT, "Felger Carbon" <fmsfnf@jfoops.net>wrote:"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:7k5gvvgbbvbl25ld4hnqvdlpensf0sqsif@4ax.com... The x86 franchise has a ceiling (it won't be a mainframe chip), and the rest of the world is going to figure out how to compete for the lower end of the market.<assume tongue-in-cheek mode>Right. The x86 can't compete with Motorola's 68K or Nat Semi's 32K.And the x86 can't compete with them super RISC chips like Alpha andthe 29K.And the x86 can't compete with the super-good, modern Itanic. Intelwon't let it. Yet.And ***of course*** the x86 will never become the heart of amainframe, economics or no economics!<exit tongue-in-cheek mode>The processor is such a small part of the economics of a mainframethat "the processor is less expensive" doesn't really cut it. Youknow that.The processor is just one piece of the puzzle. You will often hearpeople say, "If only AMD had it's own compiler." Well, they don't,and that's part of the point. They don't have a compiler not becausethey don't think it would be helpful, but because they just don't havethe money. Not only don't they have a compiler, they don't have lotsof other tools and goodies that Intel can provide to its ISV's (evenmoney, if the need is great enough). Life just isn't fair.

Red herring. If you take a look at M$'s compilers' umm, outputs, it's
kinda obvious that excellence is not obligatory for what computers are most
used for. As for Intel's compiler, I have to ask do they really have
one?... one which is generally useful for giving the best results *and*
dependability over a large repertoire (rather than benchmarks) of software?
If they do it must be rather recent. I guess with the transfer of the
Digital compilers I'd jope they actually have some "stability" now and the
whole package now rather than a plugin.

If AMD wants a "good" compiler, well matched to their software, how many
people do they need? How much money would it take? Hell there aren't that
many people around with the talent and experience for the job anyway -
typically a compiler team is quite small and if managed properly doesn't
cost that much. With open source software the options are even better.
They could do worse than shove a few $$ into OpenWatcom... and now, with
Opteron begging for success, would be a good time to do that. Time for
some action here I think.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[George Macdonald]

On Sun, 04 Jan 2004 09:01:01 -0500, Robert Myers <rmyers@rustuck.com>
wrote:
On Sun, 04 Jan 2004 02:39:06 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:<snip>I don't think there is room for Intel, AMD, and IBM as major players.Intel's real goal in life is to persuade IBM, not AMD, to get out ofthe business. That way, Intel will have two of the best marketers ofboxes (Dell and IBM) selling boxes with its chips, and it really won'thave to care all that much what AMD does.

Depends on the timeframe I suppose but I don't see any way that IBM is
going to drop its CPU development in my binoculars. I'm not even sure how
much "cross-breeding" there is between the various processors they make but
even Intel cannot afford to build the prototype(s).
The x86 franchise has a ceiling (it won't be a mainframe chip), andthe rest of the world is going to figure out how to compete for thelower end of the market. AMD is destined to live out its life as theChrysler Corporation.

The initial rate of uptake of Opteron into what are basically prototype
mainframes is really quite amazing when you think about it. It has the
register count, it has impressive IPC, it has the fastest raw DRAM
interface around, it has a fast serial, scalable pipe to the outside world
and would appear to be easy as a building block. Replace the ZIF
socket/pins you need for the PC market and there's no telling where it
could go. Chrsyler?... Ford?... GM?... they're all in deep caca... maybe a
Renault?:-)

<<snip>>It took a while but eventually, IBM was chased out of the PC market - itcan happen again and Dell is no IBM... but then again the PC coulddisappear too.In its current incarnation, I think it will. Who is to say that Sony(with IBM's technology) isn't the bigger threat to both Intel and AMDin the home market?

I believe that Sony will probably (continue to) suffer from the same
disease as Apple. It appears that Intel is betting on the central home
computer and Wi-Fi to drive that market - that model appeals to me but it's
hard to judge the mass market.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[George Macdonald]

On Sun, 04 Jan 2004 21:01:49 GMT, "Felger Carbon" <fmsfnf@jfoops.net>
wrote:
"George Macdonald" <fammacd=!SPAM^nothanks@tellurian.com> wrote inmessage news:a5jevvs5rkvojibj5tgc6rnbd528s52spa@4ax.com... (Michael) Dell is the driving force behind Yamhill????I realize that that was a joke, George. But most companies payattention to their largest customer - and Dell is in fact Intel'slargest (and most loyal?) customer.

Well no I wasn't really joking - maybe a smelly old lump of bait but if
Michael Dell has the vision he's credited with, he has to be curious about
Opteron and what it could do for him. Then again that's from my
perspective and I'm pretty much convinced that Prescott == Yamhill. Even
if Intel x86-64 doesn't actually appear the design *might* be such that it
adapts to a next level of Hyper Threading.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[Yousuf Khan]

"daytripper" <day_trippr@REMOVEyahoo.com> wrote in message
news:5hehvvo55rlf2ic3de4m9hvouf21q2a0vu@4ax.com... On Sun, 4 Jan 2004 15:11:45 -0800, archmage@sfchat.org (Nate Edel) wrote:Robert Myers <rmyers@rustuck.com> wrote: seriously when I said it, but Intel _needs_ a credible competitor (something that apparently didn't seep into the all-encompassing brain of Bill G.).Microsoft has apparantly assumed Apple would work that way in the past.
ISTRMicrosoft making a large-ish ($150M) investment in Apple around 1998 or
so,when Steve Jobs came back. Context is everything: Microsoft made that investment as a legal strategy,
as they were neck deep in court for antitrust/predatory tactics/etc, and
keeping Apple alive gave them an almost plausible defense against monopoly
charges...

They also did it for Corel.

Yousuf Khan

[George Macdonald]

On Mon, 05 Jan 2004 02:14:41 -0500, stacey <fotocord@yahoo.com> wrote:
George Macdonald wrote: On Fri, 02 Jan 2004 23:18:42 -0500, stacey <fotocord@yahoo.com> wrote:Even these Nvidia chipsets people rave about seem to have flakey issueswhen used with video editing cards? A fairly thorough search at Google doesn't turn up anybody raving about "flakey issues" with nForce(x) chipsets for video editing. You're on the record as proclaiming that the same chipsets do not support IDE busmastering when they plainly do!I was looking for a specific driver but discovered they work fine with thedefault one in win98with the right settings, I made a mistake?

IME it was kinda obvious: you stick the mbrd CD into the drive and let the
thing install the appropriate .INF files... and drivers where necessary.
Even Intel chipsets have required as much.
You can blame the cards all you want butwhen they work fine on an intel board and flake out on a Via/Nvidia, whereis the problem and who is going to fix it? You think they are going todesign cards around non-intel chipsets? For nVidia, I think you need to show more evidence or withdraw that statement.http://www.videoguys.com/system.htmThese guys sell lots of real time video cards and had this to say on theirsite about chipsets:Recommended motherboards WARNING: We do NOT recommend motherboards based on the nVidia chipsets. Weare running into a wide range of issues with them. While they may be OK forgaming, they are not for NLE.

I prefer a bit of precision in such critiques. E.g. there was apparently
an issue with the Firewire interface which was fixed with a .INF file
update which caused the OS to recognize the port properly. This kind of
general damnation of what is, for me, an excellent chipset with the least
trouble I've ever had on a Win(whatever) install, is not good enough.
VIA ChipsetsWe don't really recommend these chipsets, because of PCI bottelnecks. Ifyour motherboard uses the VIA chipset, it is important that you downloadand install the latest 4-in-1 driver, version 4.23 or later. You candownload this from your motherboard vendor's sight. Note: Installing thelatest 4-in-1 driver will improve your system performance, but it may notsolve your NLE compatibility issues. If you have a choice, do not get a VIAchipset.I suppose you can work around the issues they have buy why fight with them?I'm glad you've -never- had a problem with a Via based system but I don'tthink they are worth fooling with myself from my past experience. I buildsome video NLE systems and the intel ones never have the problems that thevia systems I've worked on (gave up on?) have.

Quite frankly, the above sites "recommendations", though they promise that
they "plan on updating it every few months" is a bunch of (mostly) outdated
gibbersih: Cyrix P120/P150?... Win95?

Then apparently all VIA chipsets and all nVidia chipsets? No model numbers
at all?... all lumped into the same boat? Then we have in relation to AMD
K6-II 300+: "We did run into some problems with early versions of the VIA
chipset. New motherboards and computers with the VIA chipset are working
properly." This is not makng sense at all.

Sorry but the credibility level is not very good.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[Yousuf Khan]

"George Macdonald" <fammacd=!SPAM^nothanks@tellurian.com> wrote in message
news:9rvhvvsksu1rjpa7o1t4l14t9a6hdc6fjt@4ax.com... If AMD wants a "good" compiler, well matched to their software, how many people do they need? How much money would it take? Hell there aren't
that many people around with the talent and experience for the job anyway - typically a compiler team is quite small and if managed properly doesn't cost that much. With open source software the options are even better. They could do worse than shove a few $$ into OpenWatcom... and now, with Opteron begging for success, would be a good time to do that. Time for some action here I think.

Well, looks like they're shoving a few $$ into Portland Group compilers,
instead.

http://theinquirer.net/?article=10292

Yousuf Khan

[Robert Myers]

On Mon, 05 Jan 2004 02:54:43 -0500, George Macdonald
<fammacd=!SPAM^nothanks@tellurian.com> wrote:

<snip>Red herring. If you take a look at M$'s compilers' umm, outputs, it'skinda obvious that excellence is not obligatory for what computers are mostused for. As for Intel's compiler, I have to ask do they really haveone?... one which is generally useful for giving the best results *and*dependability over a large repertoire (rather than benchmarks) of software?If they do it must be rather recent. I guess with the transfer of theDigital compilers I'd jope they actually have some "stability" now and thewhole package now rather than a plugin.

Intel's compiler is a descendent of the Portland Group Compiler. In
terms of the speed of the code it is able to produce, it has come
along quite nicely. I don't have a SPEC license, so I use different
benchmarks. I don't know about M$ compilers because I don't use them.
I've heard different stories. Compiler development was always planned
as an integral part of IA-64, where it is anything but a red herring.
Dunno about the money, but somebody's doing an awful lot of work.
If AMD wants a "good" compiler, well matched to their software, how manypeople do they need? How much money would it take? Hell there aren't thatmany people around with the talent and experience for the job anyway -typically a compiler team is quite small and if managed properly doesn'tcost that much. With open source software the options are even better.They could do worse than shove a few $$ into OpenWatcom... and now, withOpteron begging for success, would be a good time to do that. Time forsome action here I think.

ORC is a better bet. I've been told that a (proprietary, not under
AMD control) backend for Opteron is in the works.

RM

[Robert Myers]

On Mon, 05 Jan 2004 02:54:43 -0500, George Macdonald
<fammacd=!SPAM^nothanks@tellurian.com> wrote:
On Sun, 04 Jan 2004 09:01:01 -0500, Robert Myers <rmyers@rustuck.com>wrote:On Sun, 04 Jan 2004 02:39:06 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:<snip>I don't think there is room for Intel, AMD, and IBM as major players.Intel's real goal in life is to persuade IBM, not AMD, to get out ofthe business. That way, Intel will have two of the best marketers ofboxes (Dell and IBM) selling boxes with its chips, and it really won'thave to care all that much what AMD does.Depends on the timeframe I suppose but I don't see any way that IBM isgoing to drop its CPU development in my binoculars.

I don't either, but I'm not sure it wouldn't be the smart thing for
them to do.
I'm not even sure howmuch "cross-breeding" there is between the various processors they make buteven Intel cannot afford to build the prototype(s).
I think the plan always was to have Itanium behind a code-morphing
front-end for many applications. Nice of Transmeta to do so much R&D
for them. I guess the problem is harder than they planned on, but
what R&D project doesn't turn out that way?
The x86 franchise has a ceiling (it won't be a mainframe chip), andthe rest of the world is going to figure out how to compete for thelower end of the market. AMD is destined to live out its life as theChrysler Corporation.The initial rate of uptake of Opteron into what are basically prototypemainframes is really quite amazing when you think about it. It has theregister count, it has impressive IPC, it has the fastest raw DRAMinterface around, it has a fast serial, scalable pipe to the outside worldand would appear to be easy as a building block. Replace the ZIFsocket/pins you need for the PC market and there's no telling where itcould go. Chrsyler?... Ford?... GM?... they're all in deep caca... maybe aRenault?:-)

AMD may or may not be in trouble. They're betting on Opteron big time
with the new fab they're building. Only time will tell.
<<snip>>It took a while but eventually, IBM was chased out of the PC market - itcan happen again and Dell is no IBM... but then again the PC coulddisappear too.In its current incarnation, I think it will. Who is to say that Sony(with IBM's technology) isn't the bigger threat to both Intel and AMDin the home market?I believe that Sony will probably (continue to) suffer from the samedisease as Apple. It appears that Intel is betting on the central homecomputer and Wi-Fi to drive that market - that model appeals to me but it'shard to judge the mass market.
Hard? Impossible, I think. Lots of smart money has been sucked into
big black holes by people betting on who was going to grab the future
of home entertainment. So far, the answer is no one.

RM

[Tony Hill]

On Mon, 05 Jan 2004 11:53:09 GMT, "Yousuf Khan"
<ABCbjsk90DEF@GHIhotmailJKL.com> wrote:"George Macdonald" <fammacd=!SPAM^nothanks@tellurian.com> wrote in messagenews:9rvhvvsksu1rjpa7o1t4l14t9a6hdc6fjt@4ax.com... If AMD wants a "good" compiler, well matched to their software, how many people do they need? How much money would it take? Hell there aren'tthat many people around with the talent and experience for the job anyway - typically a compiler team is quite small and if managed properly doesn't cost that much. With open source software the options are even better. They could do worse than shove a few $$ into OpenWatcom... and now, with Opteron begging for success, would be a good time to do that. Time for some action here I think.Well, looks like they're shoving a few $$ into Portland Group compilers,instead.http://theinquirer.net/?article=10292

That sort of thing is probably the best idea for a company like AMD.
Writing their own compiler is too big of a task for too little of a
return. Very few people are likely to run an AMD compiler unless it's
a real bang-up compiler that produces great code for both AMD and
Intel processors. Even Intel's compiler doesn't really get used for
much of anything except SPEC CPU submissions.

Most code out there for x86 actually comes from just two compilers,
Microsoft Visual Studios for Windows stuff and GCC for all other
operating systems. AMD's best bet is to throw a bit of money at both
of those, and maybe at a few of the small compiler guys (like the PGI
compiler) in order to improve AMD-specific optimizations.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

[Ed]

On Mon, 05 Jan 2004 10:09:42 -0500, Robert Myers <rmyers@rustuck.com>
wrote:

AMD may or may not be in trouble. They're betting on Opteron big timewith the new fab they're building. Only time will tell.


1995, AMD could be in trouble. They're betting on K5 big time.
1997, AMD could be in trouble. They're betting on K6 big time.
1999, AMD could be in trouble. They're betting on K7 big time.
2003, AMD could be in trouble. They're betting on K8 big time.

Let me save you time and trouble right now.....
2006, AMD could be in trouble. They're betting on K9 big time!

LOL!
Ed

[Robert Myers]

On Mon, 05 Jan 2004 18:10:12 GMT, Tony Hill <hilla_nospam_20@yahoo.ca>
wrote:
Even Intel's compiler doesn't really get used formuch of anything except SPEC CPU submissions.

More and more of the heaviest duty number-crunching stuff is done
under Linux, speed is important, and icc generally outperforms gcc for
a P4 by a substantial margin on Real Code (tm), not just on Spec
submissions.

While you can still get an unsupported version of icc for Linux, you
have to pay money for the Intel Math Kernel Library, which means that
Intel has enough demand that it feels no particular compulsion to
goose things along by giving it away. Who knows, maybe people were
using MKL with gcc, but I kind of doubt it.

RM

[George Macdonald]

On Mon, 05 Jan 2004 10:00:10 -0500, Robert Myers <rmyers@rustuck.com>
wrote:
On Mon, 05 Jan 2004 02:54:43 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:<snip>Red herring. If you take a look at M$'s compilers' umm, outputs, it'skinda obvious that excellence is not obligatory for what computers are mostused for. As for Intel's compiler, I have to ask do they really haveone?... one which is generally useful for giving the best results *and*dependability over a large repertoire (rather than benchmarks) of software?If they do it must be rather recent. I guess with the transfer of theDigital compilers I'd jope they actually have some "stability" now and thewhole package now rather than a plugin.Intel's compiler is a descendent of the Portland Group Compiler. Interms of the speed of the code it is able to produce, it has comealong quite nicely. I don't have a SPEC license, so I use differentbenchmarks. I don't know about M$ compilers because I don't use them.I've heard different stories. Compiler development was always plannedas an integral part of IA-64, where it is anything but a red herring.Dunno about the money, but somebody's doing an awful lot of work.

Yeah well for IA-64, the compiler is kinda part of the hardware
architecture. The IA-32 Intel compiler has had a terrible reputation for
production use; for that use the Digital compilers were tops.
If AMD wants a "good" compiler, well matched to their software, how manypeople do they need? How much money would it take? Hell there aren't thatmany people around with the talent and experience for the job anyway -typically a compiler team is quite small and if managed properly doesn'tcost that much. With open source software the options are even better.They could do worse than shove a few $$ into OpenWatcom... and now, withOpteron begging for success, would be a good time to do that. Time forsome action here I think.ORC is a better bet. I've been told that a (proprietary, not underAMD control) backend for Opteron is in the works.

Why? Because of its Itanium heritage? Watcom had very useable compilers;
it would be a shame to see them disappear.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

[Yousuf Khan]

"Tony Hill" <hilla_nospam_20@yahoo.ca> wrote in message
news:a6e3417408fd4489bd4d102f1784a6b9@news.1usenet.com... Most code out there for x86 actually comes from just two compilers, Microsoft Visual Studios for Windows stuff and GCC for all other operating systems. AMD's best bet is to throw a bit of money at both of those, and maybe at a few of the small compiler guys (like the PGI compiler) in order to improve AMD-specific optimizations.

Haven't been in the programming game for ages now, so my knowledge of their
popular trends is now limited. I would've thought that the Microsoft and GNU
stuff would be the most popular, but you keep hearing how they "suck" and
how no one who is serious would use so'n'so compiler. So it's hard to tell
where the programming world ends up. So it would seem to me that it's much
more important to the programmers that they get a good development interface
for themselves rather than them worrying endlessly about how optimized their
programs are.

Yousuf Khan

[Robert Myers]

On Mon, 05 Jan 2004 19:04:08 -0500, George Macdonald
<fammacd=!SPAM^nothanks@tellurian.com> wrote:
On Mon, 05 Jan 2004 10:00:10 -0500, Robert Myers <rmyers@rustuck.com>wrote:On Mon, 05 Jan 2004 02:54:43 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:

<snip>
If AMD wants a "good" compiler, well matched to their software, how manypeople do they need? How much money would it take? Hell there aren't thatmany people around with the talent and experience for the job anyway -typically a compiler team is quite small and if managed properly doesn'tcost that much. With open source software the options are even better.They could do worse than shove a few $$ into OpenWatcom... and now, withOpteron begging for success, would be a good time to do that. Time forsome action here I think.ORC is a better bet. I've been told that a (proprietary, not underAMD control) backend for Opteron is in the works.Why? Because of its Itanium heritage? Watcom had very useable compilers;it would be a shame to see them disappear.

I never looked at the internals of Watcom. ORC has a modern,
hierarchical, sophisticated, and well-documented intermediate
representation that makes its easy to focus on different levels of
organization when optimizing. I don't know all the ancestry, but
there is a clear relationship between ORC and academic compiler
efforts (SUIF, IMPACT). Among all the efforts, including the Chinese
Academy of Sciences, there is a huge body of research to draw on.
Maybe there is a corresponding level of effort around WATCOM, and I'm
just not aware of it.

RM

[Tony Hill]

On Mon, 05 Jan 2004 14:05:08 -0500, Robert Myers <rmyers@rustuck.com>
wrote:On Mon, 05 Jan 2004 18:10:12 GMT, Tony Hill <hilla_nospam_20@yahoo.ca>wrote:Even Intel's compiler doesn't really get used formuch of anything except SPEC CPU submissions.More and more of the heaviest duty number-crunching stuff is doneunder Linux, speed is important, and icc generally outperforms gcc fora P4 by a substantial margin on Real Code (tm), not just on Specsubmissions.

Hmm, I think that my brain snipped out the bit I was going to write
about HPC using Intel's compiler. Yeah, it does get used for that,
but that's a pretty negligible part of the software generated.
While you can still get an unsupported version of icc for Linux, youhave to pay money for the Intel Math Kernel Library, which means thatIntel has enough demand that it feels no particular compulsion togoose things along by giving it away. Who knows, maybe people wereusing MKL with gcc, but I kind of doubt it.

Sure, for those where using one compiler over another could shave
hours or even days off a calculation, Intel's compiler gets used. But
honestly, that's less than 0.1% of the population. Look around at the
apps that are commonly used on home and business desktop systems.
You'll find mostly Microsoft compilers for Windows machines and GCC
for Linux and the BSDs. The odd-ball commercial Unix systems (does
anyone still have one of those on their desktop? :> ) would mostly be
running their own particular version of compiler.

Intel's compiler doesn't really get used for much.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

[Tony Hill]

On Tue, 06 Jan 2004 01:12:59 GMT, "Yousuf Khan"
<ABCbjsk90DEF@GHIhotmailJKL.com> wrote:"Tony Hill" <hilla_nospam_20@yahoo.ca> wrote in messagenews:a6e3417408fd4489bd4d102f1784a6b9@news.1usenet.com... Most code out there for x86 actually comes from just two compilers, Microsoft Visual Studios for Windows stuff and GCC for all other operating systems. AMD's best bet is to throw a bit of money at both of those, and maybe at a few of the small compiler guys (like the PGI compiler) in order to improve AMD-specific optimizations.Haven't been in the programming game for ages now, so my knowledge of theirpopular trends is now limited. I would've thought that the Microsoft and GNUstuff would be the most popular, but you keep hearing how they "suck" and

Sure, plenty of people talk about how badly they suck, though most
people don't bother checking first. Here's some tests a guy did
comparing a few recent versions of GCC vs. ICC:

http://www.coyotegulch.com/reviews/intel_comp/intel_gcc_bench2.html


You'll notice that most of the tests are pretty close between the two,
though occasionally one compiler will just blow the other one out of
the water (3 to 4 times faster) for a particular test. One
interesting thing of note is that sometimes the older compilers do
better than the newer ones (eg GCC 3.0.x besting 3.2.x and/or ICC 6.x
beating 7.x). This would tend to agree with some comments that Linus
Trovalds has been quoted as saying recently, that compiler writers are
trying too hard to find nifty tricks and they often end up doing as
more harm than good.

I've seen similar tests that compare ICC to Microsoft Visual C++, and
the results aren't all that different, at least since version 7 of MS
Visual Studios. Before that their compiler was a bit weak, but the
newest versions seem to usually be pretty close.
how no one who is serious would use so'n'so compiler. So it's hard to tellwhere the programming world ends up. So it would seem to me that it's muchmore important to the programmers that they get a good development interfacefor themselves rather than them worrying endlessly about how optimized theirprograms are.

Yup, that's about the long and the short of it. And I, for one, think
that this is the right choice. I'd MUCH rather have a good
application that runs bug-free than have an application that is a few
percent faster but arrives later and has more bugs. For the HPC
market, it might be worthwhile to spend a bit more time with compiler
optimizations, but the everyone else in the world, getting good apps
is what counts.

Besides, usually the biggest improvements in speed come from algorithm
changes rather than any compiler improvements.

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

[Robert Myers]

On Mon, 05 Jan 2004 12:22:05 -0600, Ed <nobox@hotmail.com> wrote:
On Mon, 05 Jan 2004 10:09:42 -0500, Robert Myers <rmyers@rustuck.com>wrote:AMD may or may not be in trouble. They're betting on Opteron big timewith the new fab they're building. Only time will tell.1995, AMD could be in trouble. They're betting on K5 big time.1997, AMD could be in trouble. They're betting on K6 big time.1999, AMD could be in trouble. They're betting on K7 big time.2003, AMD could be in trouble. They're betting on K8 big time.Let me save you time and trouble right now.....2006, AMD could be in trouble. They're betting on K9 big time!LOL!

Possibly a source of humor for you, but not for everybody. A
relatively small number of people work on the kinds of codes I work
on. No army of open source hackers is going to rewrite highly
specialized software that's been optimized for hardware that no longer
exists. Rewriting software for the architecture of the moment gets
old fast.

RM

[Yousuf Khan]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:hqmkvvku1omnlnuu76gbjvdp923icpsttu@4ax.com... Possibly a source of humor for you, but not for everybody. A relatively small number of people work on the kinds of codes I work on. No army of open source hackers is going to rewrite highly specialized software that's been optimized for hardware that no longer exists. Rewriting software for the architecture of the moment gets old fast.

Well, isn't the architecture of the moment IA64, while the old-fashioned
architecture is the AMD64?

Yousuf Khan

[Ed]

On Tue, 06 Jan 2004 07:48:23 GMT, "Yousuf Khan"
<ABCbjsk90DEF@GHIhotmailJKL.com> wrote:
"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:hqmkvvku1omnlnuu76gbjvdp923icpsttu@4ax.com... Possibly a source of humor for you, but not for everybody. A relatively small number of people work on the kinds of codes I work on. No army of open source hackers is going to rewrite highly specialized software that's been optimized for hardware that no longer exists. Rewriting software for the architecture of the moment gets old fast.Well, isn't the architecture of the moment IA64, while the old-fashionedarchitecture is the AMD64? Yousuf Khan
Yeah right! I don't know that many programmers but the ones I do know
have been around a long time and none of them are even looking at IA64,
maybe it's just too hard to teach old dogs new tricks?

Ed

[Robert Myers]

On Tue, 06 Jan 2004 07:48:23 GMT, "Yousuf Khan"
<ABCbjsk90DEF@GHIhotmailJKL.com> wrote:

<snip>Well, isn't the architecture of the moment IA64, while the old-fashionedarchitecture is the AMD64?

I have nothing to rely on but my own instincts. I have no illusion
that my instincts are perfect, but, since I have to live with the
consequences of my guesses, I prefer to make my own guesses, rather
than to look around and to see what everyone else is doing.

There is a huge repository of x86 programming skills and AMD64 solves
one of the bigger problems of x86, which is that it is starved of
named registers. In the context of this thread, though, there is no
certainty, just at the moment, that AMD64 is an attractive target
architecture for development, and even the herd is waiting to see what
Intel will do.

In a larger context, I expect IA64 to be around for a while, and I
expect it to play a big role in the kind of computing I do, but, even
more important to me, I think it is a better jumping-off place for the
future of certain kinds of computing than is x86.

RM

[Dorothy Bradbury]

> That's just the way it goes. When AMD figures _that_ part out, I might invest in some AMD stock.

We need AMD (& Dual-G5s of this world) to kick Intel into moving faster,
so all those once 360ukp P4-3.06s end up faster at 60ukp or such like :-)

More seriously, AMD are correctly targetting niches around Intel:
o Off-Brand lower-cost laptop use re chipset/processor -- cost counts
o Off-Corp consumer use re games -- power & cost counts
o Corp appliance use re firewall/consumable brick -- SOHO cost counts

There is a growing market for very low power, but increasingly powerful
appliance boxes - Mini-iTX is slowly getting its finger out, Shuttle proved
people would buy boxes-to-fall-over-then-fall-for, P4M is going Mini-iTX.

The once niche segments are growing in revenue size, quietly but growing.

Same with Opteron, it's good for cheap horsepower - and Cray do plan
on tying that cheap horsepower with cheap roads to deploy it on. Intel is
not so much lacking in power, but in a hypertransport for Xeons to use.

Yes, I do expect Intel to soon beat the dual-G5 at mainstream level.
It has before, and before, and before, and before that right back to 386.

Intels job is to make you buy as many CPUs from A-B, making it A-B-C-D-E.
--
Dorothy Bradbury
www.stores.ebay.co.uk/panaflofan for fans, books & other items
http://homepage.ntlworld.com/dorothy.bradbury/panaflo.htm (Direct)

[Dorothy Bradbury]

> So I don't see why x86 can't become a mainfame processor, like Robert has said, it's simply a matter of adding hardware.

It comes down to viability - mainframe is "ECC HA throought the system",
so whilst the Opteron/Itanium may be getting there - is it economic to add the
required engineering time & productionisation to make it viable?

Put another way, what benefit would there be to having cheap processors
like Opteron/Itanium when that isn't necessarily the real cost in mainframes?

That said, I do think processors like Opteron/Itanium2 ironically give more
life to Sun/Cray because they give it cheap horsepower on a platform that is
cheap to develop with the their past techniques. The MIPS v x86 argument
is supplanted by time-to-mkt, cost, rather than purity of solution. It comes
down to what the economic buyer wants - which is price driven, either what
does it save me in cost or what does it create for me in revenue.

Delighted with Athlon64 & Opteron, since it hits Intel at the high-end on
single & dual-processor systems. For all the hype, I still find a P2-366 is
barely any slower than a P4-CEL-2200 (both as laptops I hastily add).
The gain of course is that the latter is 1/3-1/4 the price the former was :-)

Counting pipeline depth re pipeline miss cost is perhaps ignoring that there
may well be improvements in cache - admittedly not a huge solution, but
Intel is showing willingness to vastly undercut it's big-L3-cache Xeons by
very much cheaper uni-processor big-L3-cache P4 systems. I'd be more
interesting in what future transport technology is planned than pipeline depth.
--
Dorothy Bradbury

[Robert Myers]

On Tue, 6 Jan 2004 21:35:29 -0000, "Dorothy Bradbury"
<dorothy.bradbury@ntlworld.com> wrote:

<snip>Same with Opteron, it's good for cheap horsepower - and Cray do planon tying that cheap horsepower with cheap roads to deploy it on.

I was really disappointed in the "fabric" of Red Storm, and I don't
think either Cray or the DoE is ever going to tell the truth about how
easy it is to get a high linpack score on such a box and how hard it
is to get anything else that isn't embarrassingly parallel to run
right.

People wondered why the SGI Altix is so expensive. It's not because
of the Itania, it's because it has a _real_ interconnect fabric. See
figure 3 in

http://www.sgi.com/servers/altix/whitepapers/downloads/altix_shared_memory.pdf

We'll see lots of Optera and G5's in low-rent clusters that are good
for Top 500 rankings and not much else. Oh, I forgot to mention,
hard-to-program boxes justify huge IT staff at national labs.
Intel isnot so much lacking in power, but in a hypertransport for Xeons to use.
Intel got ever-so-close to allowing an open-standard interconnect to
its processors and then thought better of it. They are such control
freaks that they would rather lose performance and at least some
customers than run the risk of having some _unlicensed_,
_unauthorized_ party hooking silicon directly to one of their
processors.

The secret probably is that they don't want Xeons competing with
Itania. If you make it too easy to hook up Xeons efficiently, Xeon
will dog Itanium until Intel kills it dead. Solution: make it
incredibly annoying and inefficient to hook up lots of Xeons in
anything that looks like a true high-end box.

RM

[John Smith]

> What presentation was that?

This one: www.goatse.cx

[Yousuf Khan]

"Robert Myers" <rmyers@rustuck.com> wrote in message
news:l6ilvv48osb626egqe3c8ekfq4hltpqp3q@4ax.com... On Tue, 06 Jan 2004 07:48:23 GMT, "Yousuf Khan" <ABCbjsk90DEF@GHIhotmailJKL.com> wrote: <snip>Well, isn't the architecture of the moment IA64, while the old-fashionedarchitecture is the AMD64? In a larger context, I expect IA64 to be around for a while, and I expect it to play a big role in the kind of computing I do, but, even more important to me, I think it is a better jumping-off place for the future of certain kinds of computing than is x86.

Well, isn't that by definition what "architecture of the moment" means?

Yousuf Khan

[Robert Myers]

On Wed, 07 Jan 2004 00:53:53 GMT, "Yousuf Khan"
<ABCbjsk90DEF@GHIhotmailJKL.com> wrote:
"Robert Myers" <rmyers@rustuck.com> wrote in messagenews:l6ilvv48osb626egqe3c8ekfq4hltpqp3q@4ax.com... On Tue, 06 Jan 2004 07:48:23 GMT, "Yousuf Khan" <ABCbjsk90DEF@GHIhotmailJKL.com> wrote: <snip>Well, isn't the architecture of the moment IA64, while the old-fashionedarchitecture is the AMD64? In a larger context, I expect IA64 to be around for a while, and I expect it to play a big role in the kind of computing I do, but, even more important to me, I think it is a better jumping-off place for the future of certain kinds of computing than is x86.Well, isn't that by definition what "architecture of the moment" means?

Not in the way that I would use words. I don't want to call IA64 the
architecture of the future, because I don't think it is. There are
plenty of people working on IA64, but very few who own up to it on
Usenet, so calling it the "architecture of the moment" would draw more
guffaws than anything else.

I have some fairly idiosyncratic ideas about the future of computing.
I've aired most of them at one time or another on comp.arch. I don't
get many people saying, "Wow, that's a great idea," but no one has
persuaded me that they have any better ideas, so I continue to follow
my own instincts.

IA64, the compiler infrastructure, the research around it, and the
prospects for future use of the architecture all fit in with my own
instincts and needs. So that this post won't seem like an exercise in
solipsism, let me recap some of the ideas I've described elsewhere,
sometimes more than once:

1. Software is incredibly predictable in execution.

2. Not using the predictability of software in execution is silly.

3. Even though software is reasonably predictable in execution, it is
not perfectly predictable.

4. No matter how much incentive there is to do otherwise, people will
code in a linear imperative style for the rest of my lifetime.

The whole premise of IA64 fits in very nicely with those ideas. It
was built with the idea of exploiting predictability. The
infrastructure to do so is partly in place and is a subject of active
research.

IA64 was built with the idea of speculation and the ability to recover
from misspeculation.

It wasn't part of the original game plan, but the architecture and
infrastructure of IA64 make it an attractive platform for speculative
threading. If people won't write multi-threaded code, let the
compiler and the processor do it for them.

It is possible, even likely, in fact nearly a certainty, that these
ideas will influence x86 architecture and compilers. They have
already done so.

Andy Glew thinks he can do it all just as well in hardware. Linus
Torvalds thinks that OoO fixes everything. Tony Hill thinks that all
the same things that IA64 is trying to do are being done by everybody
else. They could all be right, and I could be just wasting my time.

There is really very little point in having the whole world pursue the
same well-worn path if you're trying to find better ways to do things.

RM

[Tony Hill]

On Wed, 7 Jan 2004 02:14:49 +0200, "John Smith"
<someone@microsoft.com> wrote: What presentation was that?This one: www.goatse.cx

Err, I think that one is a presentation of widening "pipelines", not
adding stages to them! :>

-------------
Tony Hill
hilla <underscore> 20 <at> yahoo <dot> ca

[George Macdonald]

On Mon, 05 Jan 2004 20:39:49 -0500, Robert Myers <rmyers@rustuck.com>
wrote:
On Mon, 05 Jan 2004 19:04:08 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:On Mon, 05 Jan 2004 10:00:10 -0500, Robert Myers <rmyers@rustuck.com>wrote:On Mon, 05 Jan 2004 02:54:43 -0500, George Macdonald<fammacd=!SPAM^nothanks@tellurian.com> wrote:<snip>>If AMD wants a "good" compiler, well matched to their software, how many>people do they need? How much money would it take? Hell there aren't that>many people around with the talent and experience for the job anyway ->typically a compiler team is quite small and if managed properly doesn't>cost that much. With open source software the options are even better.>They could do worse than shove a few $$ into OpenWatcom... and now, with>Opteron begging for success, would be a good time to do that. Time for>some action here I think.>ORC is a better bet. I've been told that a (proprietary, not underAMD control) backend for Opteron is in the works.Why? Because of its Itanium heritage? Watcom had very useable compilers;it would be a shame to see them disappear.I never looked at the internals of Watcom. ORC has a modern,hierarchical, sophisticated, and well-documented intermediaterepresentation that makes its easy to focus on different levels oforganization when optimizing. I don't know all the ancestry, butthere is a clear relationship between ORC and academic compilerefforts (SUIF, IMPACT). Among all the efforts, including the ChineseAcademy of Sciences, there is a huge body of research to draw on.

I don't know at what point in the "hierarchy" the target hardware must be
addressed but from what I read ORC is focussed on Itanium... where the
compiler and hardware are intimately related.
Maybe there is a corresponding level of effort around WATCOM, and I'mjust not aware of it.

The current WATCOM is the result of some rather sad transfers o