Sign in with
Sign up | Sign in
Your question

Predictions

Tags:
Last response: in CPUs
Share
Anonymous
a b à CPUs
January 31, 2004 5:55:30 PM

A few months ago, I spent dozens of posts and countless hours on <A HREF="http://forumz.tomshardware.com/hardware/modules.php?nam..." target="_new">debating </A>the usefullness of 64 bit. I argued it was going to be useful soon -if not now- for a lot of people, and should definately be taken into consideration when buying a machine now when you expect it to last longer than 1 or 2 years. I also managed to <A HREF="http://forumz.tomshardware.com/hardware/modules.php?nam..." target="_new">convince </A> at least some regulars that it doesnt take 4 GB of RAM to require 64 bit addressing, and I <A HREF="http://forumz.tomshardware.com/hardware/modules.php?nam..." target="_new">predicted intel would not be able to ignore this for more than a year </A>, and would have no choice but to support it.

Well, few people believed me (any of it), precious few beleived intel would adopt the AMD standard, and lots of them continued arguing 64 bit was a non issue, and incorrectly relating it to >4 GB RAM (for those who still don't know, the real issue is 2 GB virtual memory per process, regardless of how little RAM you have).

This week, I linked to some very convincing rumours Intel was about to come clean on Yamhill, and would introduce 64 bit capable x86 cpu's. Again, most of the people that didnt believe me back then, would not believe it now, even if Otellini's words where really quite impossible to interprete any other way. A lot of them remained sceptical, some of them even now when every major newssite has reported this as a fact. Some other people are still unsure intel's Yamhill will be AMD64 compatible (it will).

Now I am going to make a few bold predictions, you may bookmark and throw it my face later if ever I'm proven wrong, but only if you took a position yourself in this thread. Its too easy to say 'you where wrong' when you didnt say a thing yourself. Here goes:

1) Intel will release 64 bit capable cpu's that are 100% compatible with AMD64. No new x86 extented ISA, no x86/IA64 hybrid.

2) These chips will not be available in the market until the very end of this year or more likely, somewhere in the first half of 2005. They may be announced this summer or so, but you won't be able to buy them before then, most likely when tejas ships, and not earlier. Even if windows for AMD64 ships much earlier. Intel is *not* waiting for windows like it will say, its desperately trying to fix their chips.

3) Intel fanboys that have been saying 'wait for prescott' since last summer, will soon start saying 'wait for the improved s775 prescott', and then, 'wait for tejas'. those same fanboys that ridiculed the 64 bit argument only a month ago, will very soon advice you to wait for 64 bit capable prescott or tejas chips, and if Intel adds some extra instructions to AMD64 (which is likely IMHO) use this as an argument against Athlon 64's (I hope you see the irony).

4) By the end of next year, 80-90% of us running 64 bit capable cpu's (intel or AMD) will run 64 bit windows (or Linux), and run/play several 64 bit games/apps. Those who just got a 32 bit only P4 will be slapping themselves, especially the one's that got themselves a >$1.000 P4EE.

5) But most of you will likely still not have more than 4 gigs of ram.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

More about : predictions

January 31, 2004 6:27:40 PM

Intel will release 64 bit capable cpu's that are 100% compatible with AMD64. No new x86 extented ISA, no x86/IA64 hybrid


Does it will be only for XEON or the full ligne from celeron to XEON MP.Yes XEON face a need for 64 bit computing.Itanium is still in heavy developement and there about the same number of itanium sold that opteron.2P-4P segment is not ready for itanium a tempory solution must be found.In a case there is a Xeon that support 64 bit mode it still can be emulate with Excution layer for itanium.Vanderpool technologie will be ready soon itanium will gain the most out of this.


To be AMD64 compatible AMD would have know it.A license is need as SSE2 and others been lincense by intel for AMD.They pay intel money for each CPU sold.So my point is AMD would have say that intel would back there technologie but no such thing have happen.


The pure need of 64 bit computing in desktop dont exist and wont improve performance SPEC cpu prove it the 64 bit mode is slower.So may bash SPEC but institution and enterprise will look at SPEC and respect it.Intel itself dont goes will on 64 bit computing even for there itanium.

I dont like french test
January 31, 2004 6:31:02 PM

Hm... This is speculation land, but OK for a gedanken experiment.
Quote:
precious few beleived intel would adopt the AMD standard

It is still unclear that intel will adopt x86-64 exactly in the same way as A64 will. It is much more likely that they will develop an extended version thereof and then say "this is our version of 64-bit extensions". There is nothing ironic in that, and there would be nothing wrong in recommending their approach to it if it turned out to be better or faster than AMD's equivalent. There's no "the original is better" in this game.

Therefore, Intel's 64-bit tech being compatible with AMD64 is a no brainer. But will it limit itself to that? Hell, noone here can know anything about that. Not even you, bob.

In any case, the two architectures (a64 and P4) are so different that, even if you were to implement the exact same x86-64 in P4s, it might bring some differences that I can't hope to predict. How will a high clock, low IPC architecture react, exactly? Will its high clock enable it to do lots of work (like with SSE2, for instance... it makes P4 quite strong), or will P4s basic low IPC structure cripple the 64-bit extensions?... Will optimising to x86-64 benefit A64 more than it will the future and eventual P4-64 (I know this is highly speculative, but we're at it, aren't we?)

But as for this one:
Quote:
2) These chips will not be available in the market until the very end of this year or more likely, somewhere in the first half of 2005. They may be announced this summer or so, but you won't be able to buy them before then, most likely when tejas ships, and not earlier. Even if windows for AMD64 ships much earlier. Intel is *not* waiting for windows like it will say, its desperately trying to fix their chips.

This one I can relate to. The problem for Intel right now is that they have said they'd introduce Prescott now (meaning, within 48 hours...). So if Prescott doesn't have the 64-bit tech now, because Intel didn't manage to get it to work, then that would be bad... it would be hard for Intel from a marketing point of view to launch a processor right after Scotty that uses 64-bit tech, because Scotty would be rendered obsolete. So I'm thinking maybe Intel will hold its 64-bit tech until the 2nd half of 2004. But it's hard to really predict this.

Unfortunately, though, I can't agree that most people will be using 64-bit software extensively by the end of 2004 at all. This is highly unlikely; it would mean that in 2004, most people upgraded to A64 or p4-64 tech, because in 2003, A64 tech carried a heavy price tag. But only enthusiasts actually change/upgrade their computers in under 12 months of them being new! And even that's a wealthy enthusiast. I have this computer for some 3 years now. Besides, Windows with extensive AMD64 support will take a while, for whatever reason that made MS delay it.

So, while I can agree that many people will be using 64 bit software and hardware by the end of the year, I can only say this to your prediction (90%): Hah! :lol: 

What I do predict is that Intel will not sit on their hands for the whole of 2004. The introduction of PCI-E and DDR-2 might signal that they're up to some sort of technology rush. Many other things are being "carried" by Intel: BTX standard, matrix-RAID functionality, new onboard graphics, new onboard sound... many new resources are being added to the Scotty lineup.

And my main prediction: <i>A new prescott will be launched with Alderwood and Grantsdale. It will probably be L775-based, and it will use 1066Mhz and clock at 13x266 = <b>3.46Ghz</b>. This will run synchronously with DDR2-533. You can hold me to this.</i>

Oh boy, it certainly will be interesting to visit this thread 6 months from now! :smile:

:evil:  <font color=red><b>M</b></font color=red>ephistopheles<P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 01/31/04 06:46 PM.</EM></FONT></P>
Related resources
January 31, 2004 6:34:59 PM

http://www.realworldtech.com/forums/index.cfm?action=de...

there is a discussion about IA32+

EX:



The success of Xeon in the server industry combined
with the lateness of IPF in gaining momentum just
about guaranteed that Intel would eventually extend
x86 to 64 bits to avoid leaving an opening in the
2-4P volume server segment. The timing of what is
widely expected to be an announcement of Intel's
x86-64 technology is a bit of a surprise IMO, its
about a year earlier than I expected.

But from all that I can discern Intel is not backing
away from IPF in the least. Far from an indictment
of IPF's market success to date, I think the timing
of the announcement indicates that Intel thinks its
IPF development plans, associated OEM/ISV support,
and market segment acceptance have all reached a
level at which news of the extension of Xeon to 64
bits will not undermine IPF. At least among people
and institutions that specify, purchase, and use
the classes of machines IPF currently targets. The
usual suspects and chicken littles will be noisy
for a while, as we are certainly seeing in the last
few days, but it should die down as it becomes
apparent that IPF product introduction plans and
growth in revenue and MSS accelerate unabated.

Intel's plan for the near to mid term is quite clear.
It wants to level the server platform cost playing
field between Xeon and IPF. In fact some indications
are that future 64 bit Xeon and IA-64 products will
share the *same* infrastructure. Intel also indicates
that it thinks the architectural based implementation
advantages arising from IA-64 processor simplicity
and compactness will allow twice as many IA-64 CPU
cores per device as x86 based cores.

All process technology trends clearly favour cache
transistors over logic transistors and the advantage
IA-64 holds in this area will only grow over time.
The new Prescott core ostensibly requires over 70m
transistors. A future version of the I2 core with its now
redundant x86 logic removed would likely need less
than 15m transistors. What new ideas the former ADT
is bringing to bear on IA-64 in the upcoming Tukwila
core we can only speculate about. IMO ongoing long
term research in compilers/translators, both static
and dynamic, will also increasingly work to EPIC's
advantage.

I think we are on the verge of a grand experiment
in computer architecture. Intel plans to eliminate
platform related variables between Xeon and IPF.
If it also brings process technology utilization into
chronological alignment between IA-64 and x86
we could see a unique competition between two
very different schools of thought in instruction set
architecture design. The results will be obvious
to all in terms of relative product performance,
cost, price/performance, power/performance etc.
May the race be fair and the best architecture win. ;-)


I dont like french test
January 31, 2004 6:36:47 PM

Hey! You're right! Spec Scores for an Athlon FX-51 running Windows XP Professional are <b>higher</b> than for an Opteron 148 running SuSe Linux 9.0 (with 64-bit support)...
WTF? Is this representative?... Ouch.
Quote:
To be AMD64 compatible AMD would have know it.A license is need as SSE2 and others been lincense by intel for AMD.They pay intel money for each CPU sold.So my point is AMD would have say that intel would back there technologie but no such thing have happen.

I was under the impression that their cross-licensing implied that they didn't need to ask each other for those technologies... Was I wrong? :frown:



:evil:  <font color=red><b>M</b></font color=red>ephistopheles
January 31, 2004 6:52:48 PM

I was under the impression that their cross-licensing implied that they didn't need to ask each other for those technologies... Was I wrong

I dont know much either this deal was update 2 year ago and SSE2 was include and should be update soon in 1 or 2 year that might fit where intel wont to release a 64 bit pentium.It like RDRAM tale it cloudy all over and the same time rambus have more and more chance to win in court again infineon.

I dont like french test
January 31, 2004 6:57:00 PM

Its possible. I got tired of waiting last summer and bought my PIV 2.4c. I am glad I did. Everything might work out so that when I want a new system, 64bit will be an affordable reality...

----------------------
PIV 2.4c @ 2.89ghz
1gig PC3200 (512mbx2)
ASUS P4P800
GF3 Ti200 64mb (soon to be replaced)
WinXP Pro
3DMark2001SE: 6300
3DMark2003: 688
January 31, 2004 7:45:43 PM

I just wonder why the 64bit AMD dosn't smoke the 32bit P4EE?
Infact the are so close in preformance, that the idea to talk about a clear winner, is something only fanboys(on both sides)do...
Besides that is always "stupid" to but the top-of-the-line-CPU(moneywise) but then again it's peoples own money, so I guess it must be envy, over not self having suffcient funds to buy one, to ridicule these people?

Terra...

Don't pretend - BE!
Anonymous
a b à CPUs
January 31, 2004 8:23:36 PM

>Does it will be only for XEON or the full ligne from
>celeron to XEON MP.

I think CT (64 bit support) will be implemented in the same range of chips, and rolled out the same way as Hyperthreading. Xeon first, followed by high end P4s (maybe called P5 by then), and slowly moved down until everything except Celeron has it. I don't think we'll see 64 bit Celerons anytime soon.

>To be AMD64 compatible AMD would have know it.A license is
>need as SSE2 and others been lincense by intel for AMD

Intel and AMD have cross licenced their extentions to the x86 ISA. AMD could use MMX, SSE, SSE2 without even asking, but intel is also free to implement AMD64 and was free to use 3DNow if they had wanted.

>The pure need of 64 bit computing in desktop dont exist

I've already discussed this extensively.. let me just say I disagree, and refer to the discussion I linked above.

>SPEC cpu prove it the 64 bit mode is slower

You are comparing SPEC scores in 32 bit using Intels SPECial ICC compiler, and in 64 bit mode using GCC. Not 32 bit versus 64 bit performance. No, I never claimed 64 bit would give you a huge performance increase either, but being able to run large apps in a flat address space will surely become usefull sooner than you think.

BTW, Juin, is it so hard to use [ quote ] or ">" when you are quoting ? your posts are hard to read, and not even using the ">" makes you seem lazy and doesnt entice me to really give you an appropriate answer.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
January 31, 2004 8:31:59 PM

>It is still unclear that intel will adopt x86-64 exactly in
>the same way as A64 will. It is much more likely that they
>will develop an extended version thereof and then say "this
>is our version of 64-bit extensions".

Quite likely, in fact that's what I wrote.

>In any case, the two architectures (a64 and P4) are so
>different that, even if you were to implement the exact
>same x86-64 in P4s, it might bring some differences that I
>can't hope to predict.

Differences will be minimal. x86-64 is so close to x86 that I really don't think it will give weird results on a P4.. Unless... intel decides not to implement true 64 bit ALUs, but uses 2 32 bits to perform 64 int math. In that case, 64 bit INT performance will be very much worse than on AMD, but then again, 64 integer code is a very, very uncommon thing (mostly only encryption stuff uses this). Overall, P4 using AMD64 wil perform pretty much like a P4 running x86, with a similar small boost as AMD. That is my guess anyway.

> it would be hard for Intel from a marketing point of view
>to launch a processor right after Scotty that uses 64-bit
>tech, because Scotty would be rendered obsolete

Oh, but intel has a history of launching obsolete processors. Think 200 Mhz Pentiums without MMX, Willamette's, 3 GHz P4's without HT, etc.. No, that wouldnt stop them.

>Unfortunately, though, I can't agree that most people will
>be using 64-bit software extensively by the end of 2004 at
>all.

You misread me. end of NEXT year, so end of 2005. And as for using 64 bit extensively.. mainly the OS, and one or two games/apps (like divx encoding apps, and UT or HL2). I'm not suggesting you'd be running 64 bit office, antivirus or instant messenger software.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
January 31, 2004 8:35:31 PM

>Hey! You're right! Spec Scores for an Athlon FX-51 running
>Windows XP Professional are higher than for an Opteron 148
>running SuSe Linux 9.0 (with 64-bit support)...
>WTF? Is this representative?... Ouch.

Too lazy to look it up, but I'm fairly sure the XP scores are obtained using the (much faster on SPEC) intel compiler, and Suse scores using GCC. That explains everything. Once intel releases its 64 bit compiler, rest assured it will perform (slightly) faster than in 32 bit mode.

BTW, the main advantage of 64 bit is not better performance, its more freedom for developpers.

>I was under the impression that their cross-licensing
>implied that they didn't need to ask each other for those
>technologies... Was I wrong?

No. Though their cross licence is limited to x86 extentions AFAIK. AMD is not allowed to use the P4 bus, intel can not just implement hypertransport either. But as I replied to June, intel CAN use AMD64, just like AMD could use SSEx.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
January 31, 2004 8:44:04 PM

>I just wonder why the 64bit AMD dosn't smoke the 32bit
>P4EE?

Because 64 bit isnt twice as fast as 32 bit (with the exception of 64 bit integer math like used in encryption). 64 bit is a few billion times the addressable memory of 32 bit though. That is the advantage.

A 32 bit cpu is practially limited to running 2 GB footprint apps, a 64 bit processor can allocate more terrabytes than you can image per app. And before you say: "who needs 2 gig ?", note that quite a few 2 or 3 year old games approach or exceed 1 gigabyte footprints (though they work fine with just 512 MB ram). I do expect next generation games to bump into those 32 bit limitations. "Out of virtual memory" error messages are upon us, and no ammount of RAM upgrades can solve it.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
January 31, 2004 8:46:06 PM

Quote:
1) Intel will release 64 bit capable cpu's that are 100% compatible with AMD64. No new x86 extented ISA, no x86/IA64 hybrid.

It's possible, but... Let's check history again:
- Intel brings us MMX.
- AMD brings us MMX, and adds 3DNow! At most 1% of people buy AMD for the 3DNow! capabilities. No sane developer actually uses it.
- Without much rumour, Intel brings us SSE. It's faster and has a superiour design. After a while developers start using it and many users benefit from it. AMD feels forced to adopt SSE support, but implements it badly.

I have studied AMD64's encoding format, and found it all to be a hack. A prefix byte here, extend some fields with some bits there and bingo: 64-bit support. Ok sure, it works, but that's where it ends. AMD seems to be incapable of totally reworking the core to improve the architecture, just like they failed with 3DNow!

Although I realize I could be totally wrong, this tells me Intel will do better than AMD once again. It might take an extra year, but that's a fair price for correcting AMD's mistakes.

AMD's core is still nearly the same as the Pentium Pro from '95. Even the 64-bit processor is little different. In the meantime, Intel designed the whole new Pentium 4 core, and the Itanium. AMD just doesn't have the R&D.

So here's my prediction:
- AMD thinks it's taking over the world but can't supply.
- Intel designs an entirely new architecture that makes AMD64 look silly.
- Microsoft uses Intel's design for Windows 64-bit versions.
- AMD adopts Intel's design but needs two years to get it right.
- In the meantime Intel uses 35 nm technology and runs dual-core quad-HT with integrated RAM.

Ok maybe not that last thing... Don't get me wrong, I wish AMD all the best. But I find it a bit hard to believe that the only thing Intel researchers have done in the past years is increasing FSB speed, enabling Hyper-Threading and adding a L3 cache for the P4EE.
Quote:
5) But most of you will likely still not have more than 4 gigs of ram.

You probaly already know this, but may I remind you that even having 8 GB or RAM on a 32-bit CPU is no problem at all? Like you've said yourself, every -process- can address 4 GB of memory. But there isn't a single application today that demands that amount of memory. And even if it does exist, it's not Office 2004 or Doom 3. And if any such application is developed in the next year, there's only one thing to say about it: bad programming! Developers don't look at the amount of addressable memory, they look at the average amount of memory the consumer has, and that number is between 256 and 512 MB. And an intelligent developer even knows that exceeding that limit is very bad for performance. I'll let you do the math: A Pentium 4 has 6.4 GB/s memory bandwidth, with a game that we'd like to run at 64 FPS, this gives us ... GB we can process per frame, maximum.

Intel knows what it's doing. Period.

<P ID="edit"><FONT SIZE=-1><EM>Edited by Whisper on 01/31/04 05:48 PM.</EM></FONT></P>
January 31, 2004 8:53:23 PM

quote

Intel and AMD have cross licenced their extentions to the x86 ISA. AMD could use MMX, SSE, SSE2 without even asking, but intel is also free to implement AMD64 and was free to use 3DNow if they had wanted


ME

No for SIMD they must license the instruction set.True for the SPEC result but dont jump to fast to think intel will have there compiler and tool ready for helping amd.

I dont like french test
January 31, 2004 9:02:54 PM

I have studied AMD64's encoding format, and found it all to be a hack. A prefix byte here, extend some fields with some bits there and bingo: 64-bit support. Ok sure, it works, but that's where it ends. AMD seems to be incapable of totally reworking the core to improve the architecture, just like they failed with 3DNow!

you forgot the extra bit for register adressing.Reg must be adresse like ram so having more need more bit so more RC delay in the front-end.

I dont like french test
Anonymous
a b à CPUs
January 31, 2004 9:24:11 PM

>Intel knows what it's doing. Period.

Hmmm. It seems to me this preconceived idea makes you think all the other things and is the basis of your arguments, instead of the other way around. Intel doesnt always seem to know what its doing, surely I don't need to remind you of all the RDRAM, 1.13 GHz and 820 fiasco's. People thought intel knew what it was doing when it ignored DDR too, but intel and people thought wrong. I wouldnt have blind faith in intel's judgements.

>AMD feels forced to adopt SSE support, but implements it
>badly.

There is nothing wrong with AMD's implementations of SSE or
SSE2. In fact, AMD's SSE2 is even slightly more capable
than intel's, but since AMD uses high IPC, low clock
designs rather than the other way around, SSE1/2
performance seems lower. Per clock, it is better (at least
SSE2).


>I have studied AMD64's encoding format, and found it all to
>be a hack. A prefix byte here, extend some fields with some
>bits there and bingo: 64-bit support.

LMAO. You just made a fool out of yourselve. Every (and I mean *every*) person that is knowledgeable in this field that I have heard, all agree: with AMD64, AMD did the best they could with the turtle the x86 ISA is. They fixed its biggest problems (introduced by Intel btw), and the extention to 64 bits is both as elegant and powerful as it gets without breaking compatibility. OTOH, I've heard very different opinions on IA64 even when intel had the unique opportunity to start from scratch..

>AMD's core is still nearly the same as the Pentium Pro from
>'95. Even the 64-bit processor is little different. In the
>meantime, Intel designed the whole new Pentium 4 core,

Another useless BS statement.

> Intel designs an entirely new architecture that makes
> AMD64 look silly

They tried that with IA64 and mostly managed to make themselves look silly. You think they are going to try that AGAIN ? Seriously ??

>- Microsoft uses Intel's design for Windows 64-bit
>versions.

Sure.. quite likely. they sold 15 copies of windows for IA64, now I assume they will ditch their nearly finished copy of windows for AMD64 and start all over again for the next intel fiasco. I think that is credible. Want to bet ? I'll take 50-1 odds.

>But I find it a bit hard to believe that the only thing
<Intel researchers have done in the past years is increasing
>FSB speed, enabling Hyper-Threading and adding a L3 cache
>for the P4EE.

WHy is it so hard ? its reality. You could imagine it, yet it is exactly what has happened ! Oh, they did make a nice chip based around that ancient old Pentium Pro core from '95 you so despise though.. IMHO the nicest chip on the market: the Pentium M.

>You probaly already know this, but may I remind you that
>even having 8 GB or RAM on a 32-bit CPU is no problem at
>all?

It may not be a problem now that 4 GB dimms have hit the market, but its not very usefull if you can't access that memory without ugly slow and complicated tricks a la PAE.

> Like you've said yourself, every -process- can address 4
>GB of memory. But there isn't a single application today
>that demands that amount of memory

2 Gigabyte. Every 32 bit OS reserves 2 GB for the kernel and VM, and 2 for the apps. With tweaks and hacks this can be turned into 3 GB in theory, 2.5 max in reality (requires recompiled apps though). If you think 2 GB is outrageous, I suggest you plot a line trough the datapoints of memory useage over the last ten years. Anyway, I discussed this so often, Im not gonna restart it. Lets just agree to disagree, and see who is right in 2 years.

= The views stated herein are my personal views, and not necessarily the views of my wife. =<P ID="edit"><FONT SIZE=-1><EM>Edited by bbaeyens on 01/31/04 06:25 PM.</EM></FONT></P>
Anonymous
a b à CPUs
January 31, 2004 9:35:30 PM

>No for SIMD they must license the instruction set

Got any link ? I've read the opposite on so many places, I won't just take your word for it. They must respect each others copywrite (eg AMD can not claim the rights to the SSE name, and may have to add "SSE is a trademark of Intel Corp"), but they can implement it. VIA and transmeta have to licence it,just like transmeta had to licence AMD64 (and did). Not intel. This is the result from a lawsuit in which intel tried to revoke AMD's licence to build x86 chips. It got settled, and AMD has an eternal licence to x86 and its extentions (cross licence), but as a result of that settlement, lost the rights to use intel's sockets (hence their need to licence EV6 bus instead of using the P4 bus).

>True for the SPEC result but dont jump to fast to think
>intel will have there compiler and tool ready for helping
>amd

Well.. they may not *want* to help, but they are going to need a compiler for their own CT chips. the only way around this I see, is that CT would be some sort of "AMD64+" and ICC would only support this AMD64+, and not AMD's AMD64. Its possible, but not likely. So far, intel has always played it fair in this regard, and ICC does not penalize AMD's chips on purpose. The proof is AMD's SPEC scores. Since intel wants developpers to use ICC, I don't think this will change. If they do, its simple, ICC will rarely be used to compile 64 bit binaries, and that just isnt in intel best interest either.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
January 31, 2004 10:59:08 PM

Quote:
It seems to me this preconceived idea makes you think all the other things and is the basis of your arguments, instead of the other way around.

No it isn't. All I'm saying is that Intel has worked hard on Yamhill the last few years. So it seems really unlikely to me that Intel would suddenly copy all of AMD64 specifications. They have the R&D to make a better 64-bit design. I'm not blind or anything, AMD might be the big winner after all, but you seem to underestimate Intel's capabilities.
Quote:
There is nothing wrong with AMD's implementations of SSE or
SSE2. In fact, AMD's SSE2 is even slightly more capable
than intel's, but since AMD uses high IPC, low clock
designs rather than the other way around, SSE1/2
performance seems lower. Per clock, it is better (at least
SSE2).

The Athlon had higher latencies on nearly all MMX/SSE instructions than the Pentium III. Their higher IPC compared to the Pentium 4 also isn't worth anything when multiplied by the clock frequency. In your metric the 486 was probably the best chip ever?
Quote:
Every (and I mean *every*) person that is knowledgeable in this field that I have heard, all agree: with AMD64, AMD did the best they could with the turtle the x86 ISA is. They fixed its biggest problems (introduced by Intel btw), and the extention to 64 bits is both as elegant and powerful as it gets without breaking compatibility.

They did not do the best possible. They did the best possible with the least effort, that part is true. In the end it's all just a hack. The introduction of new modes brings lots of possibilities to clean up the instruction set, but they didn't do that. And what are these "biggest problems" you're talking about?
Quote:
Another useless BS statement.

Ah, the easy answer. Allow me to respond with the hard question: show me wrong.
Quote:
They tried that with IA64 and mostly managed to make themselves look silly. You think they are going to try that AGAIN ? Seriously ??

IA64 is a fixed architecture always targetted at the server market, where it performs very well. Besides, I wasn't talking about a whole new architecture.
Quote:
Want to bet ?

No. The smartest choises are not always the ones taken. I don't want to bet on somebody else's stupidity.
Quote:
Oh, they did make a nice chip based around that ancient old Pentium Pro core from '95 you so despise though.. IMHO the nicest chip on the market: the Pentium M.

I don't despise the Pentium Pro. I'm just saying newer architectures like the Pentium 4 have a lot more future. The Pentium M is indeed a very nice chip but it isn't the fastest.
Quote:
It may not be a problem now that 4 GB dimms have hit the market, but its not very usefull if you can't access that memory without ugly slow and complicated tricks a la PAE.

I was talking about multiprogramming. Workstations with 8 GB of RAM don't need it to run one program, they need it to run many. Even servers with 64 GB of RAM only need it because thousands of processes each take a little bit.
Quote:
If you think 2 GB is outrageous, I suggest you plot a line trough the datapoints of memory useage over the last ten years.

I never said it's outrageous. It just isn't wise to use 2 GB of memory for a single application. And when I plot that line of memory usage I see Intel still has several years to perfect its 64-bit design. They just have to rush it now because of the AMD64 hype. Please give me an example of an application that would benefit immensely from 64-bit addressing, within the next twelve months. Even a game of 3 CDs would still fit entirely in 2 GB addressable memory.

But again, I have totally nothing against AMD. I'm just trying to advice everyone to wait for Intel's 64-bit design and then make a wise decision. If AMD happens to be the best, I'll buy AMD as well. Just don't count Intel out that fast...
February 1, 2004 12:25:03 AM

Bob Baeyens, I appreciate your opinion, but I'm afraid I can certainly relate to Whisper's. I just want to remind you that while your suspicions are natural, it is also somewhat likely that the truth lies somewhere in between the two perspectives. If Intel really is as bad as your post wants to show it to be, then they'd be out of business. They make good processors, but like any other company, their policies sometimes aren't the best. This is not to justify them, but rather to encourage a more balanced attitude. This "Intel caused the big problems in x86", "Intel did this and that fiasco" attitude is exaggerated. One could find problems in AMD's processors and policies as well. Remember the old Palominos, with no thermal protection whatsoever?

But reminding us of those events is pointless. It suggests that Intel and AMD are perfectly integrated entities that follow basic guidelines... Well, be careful not to oversimplify things. It's possible to quote events from Intel that would suggest it's a respectable and good company, but the reverse is also true. Quoting good things will make you have faith in them (and there are good things), and quoting bad things will make you suspicious... But you can't choose your pick: we'd be talking about the same old Intel, in both cases.

Therefore, I wouldn't suggest blind faith in anything nor anyone, and I wouldn't suggest hard skepticism either... Just pay attention and don't lose perspective in predictions of the future... But hey, that's just me. Sorry, bbaeyens, I wasn't picking on you, I was just expressing my thoughts here... :smile:

BTW, you're right, I must have misread you on that 90% 64-bit software projection. If it's by the end of 2005, then I might agree with you. This sounds reasonable. It's still optimistic, I think, but quite reasonable.

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
February 1, 2004 3:11:51 AM

<There is nothing wrong with AMD's implementations of SSE <or
<SSE2. In fact, AMD's SSE2 is even slightly more capable
<than intel's, but since AMD uses high IPC, low clock
<designs rather than the other way around, SSE1/2
<performance seems lower. Per clock, it is better (at <least
<SSE2).

What you talking about SSE2 is SSE2 there no AMD or intel implemtation and i can guess that instruction lantency where fix so that a pure question of clock.Many benchmark show that.


<LMAO. You just made a fool out of yourselve. Every (and I <mean *every*) person that is knowledgeable in this field <that I have heard, all agree: with AMD64, AMD did the <best they could with the turtle the x86 ISA is. They <fixed its biggest problems (introduced by Intel btw), and <the extention to 64 bits is both as elegant and powerful <as it gets without breaking compatibility. OTOH, I've <heard very different opinions on IA64 even when intel had <the unique opportunity to start from scratch..

So explaime me this they change nothing they add 1 bit for reg and 1 bit for the long mode.


>AMD's core is still nearly the same as the Pentium Pro from
>'95. Even the 64-bit processor is little different. In the
>meantime, Intel designed the whole new Pentium 4 core,

>Another useless BS statement


K7 core was a over boost P6 architecture K8 have the same base the same load/store architecture same decoder same excution units same cache systeme are the same OoO are very similar to P6 and K7 if not identical.

I dont like french test<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 02/01/04 02:22 AM.</EM></FONT></P>
February 1, 2004 3:18:08 AM

True intel are fair with ICC but that dont mean they will make it as there compiler is not use in overall software.

It got settle yeah right more the FTC as settle it for them as intel was in right they have made X86-32 bit and no i dont get a link i remember a article from the inquirer they were giving link to the settlement.They have to update cross license over time and new point got debate.
In any case that intel come to have 50% of the market share they can sue AMD for patent use.

I dont like french test
February 1, 2004 3:44:28 AM

Only a fool would suggest that the pentium pro architecture is capable of 2.2 gig speed.
While the basic layout of the K8 is similar to that of the K7, the components bare less similarity than do the volkswagon and porshe engines. If you think intel can build there vehicle better by putting the wheels on top, think again. The placement of components on die is the reason that Amd has been able to reach the speeds they have with thier silicon. Do you think intel is happy with thier own chip layout?
February 1, 2004 5:24:20 AM

Sorry but i do not reply to this

I dont like french test<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 02/01/04 02:25 AM.</EM></FONT></P>
February 1, 2004 5:31:05 AM

AS you make prediction and that risk.I made my on prescott long ago about prescott.I was saying that it will be 10 to 15% faster clock for clock compare to northwood i will see if i am still right.

I dont like french test
February 1, 2004 5:31:28 AM

Yes, I see...
Anonymous
a b à CPUs
February 1, 2004 6:37:36 AM

>No it isn't. All I'm saying is that Intel has worked hard
>on Yamhill the last few years. So it seems really unlikely
>to me that Intel would suddenly copy all of AMD64
>specifications.

Its not unlikely if you factor in Microsoft. Intel started working on Yamhill well before AMD released their specs of AMD64. I simply don't believe MS will support two competing x86 extentions, because it is not in their interest to fragment the market; and since much to everyone's surprise, they embraced AMD's solution, I think there is no alternative for intel than either ignore AMD64 and bet on IA64, or go back to the drawing table.

>The introduction of new modes brings lots of possibilities
>to clean up the instruction set, but they didn't do that.
>And what are these "biggest problems" you're talking about?

Extending the register set, and dropping virtual 8086 mode to name the most important. BTW, there are not 300 ways to extend a ISA from 32 to 64 bit. What AMD did is completely in line with any other 64 bit ISA extention (with the exception of what I noted above) or even Intel extentin to IA32, and probaly intel's solution wouldnt have been all that different.

>Ah, the easy answer. Allow me to respond with the hard
>question: show me wrong.

Show you wrong about what ? You stated nothing usefull. AMD would still be "like the PPRo", while intel has this all new shiny P4 core. IMHO the P4 isnt nearly as new as the K8. It has a looooong pipeline; big deal. And a trace cache. Nice. But no integrated MC, no glueless SMP, no 64 bit extentions, etc, etc. In the end, with its "all new core", its still the inferior product, especially in SMP configurations.

>IA64 is a fixed architecture always targetted at the server
>market, where it performs very well.

Itanium performs rather well though x86 in all its uglyness is within armsreach (actually beats it on INT, and not all that far on FP), in spite of the fact it only has a fraction of the cache, a fraction of the die size and transistor count, a significantly smaller power consumption, and 15 years of legacy to be compatible with instead of being able to start with a clean sheet. Where exactly is this huge advantage of this *ISA* ? If you compare Itaniums with similar ammounts of cache and die size with x86 cores, it gets spanked. Not too mention the fact the ISA so overly complicated it is completely impossible for a human to understand and debug the compiled code. Time may prove me wrong, but for now, I see IA64 as two steps back from the ISA's it was supposed to replace, like Alpha.

>No. The smartest choises are not always the ones taken. I
>don't want to bet on somebody else's stupidity.

I'm not even suggesting intel is making stupid choices. They betted on Itanium, which wasnt a stupid idea at all. In fact, if it werent for AMD and Microsoft working together, they would probably have pulled it off, and in a few years we'd all be running IA64 boxes. It seems they just overestimated Itaniums succes, and underestimated AMD.

> I'm just saying newer architectures like the Pentium 4
>have a lot more future. The Pentium M is indeed a very nice
>chip but it isn't the fastest.

I think just the opposite. With its stellar clock, but less than stellar performance, power consumption goes through the roof and limits scaling. Prescott should give you a hint of this; I think Banias/Dothan has more future than Netburst. Its not for no reason that intel has gone a different route with with their new server chip (Itanium) and their latest mobile chip (banias/dothan). Both are low clock high IPC braniacs, and not speed demons like netburst. Coincidence ? perhaps, but I think not.

>Workstations with 8 GB of RAM don't need it to run one
>program

Hu ? Typical workstations are purchased to basically run just one app (that usually costs far more than hardware).

> Even servers with 64 GB of RAM only need it because
>thousands of processes each take a little bit

If that is true, why has every other server ISA migrated to 64 bit nearly a decade ago then ? Do you still see any 32 bit Power, SPARC, or PA Risc systems being sold ?

>Please give me an example of an application that would
>benefit immensely from 64-bit addressing

Any app that can use more than 2 GB addressing, wether that is a game, engineering tool, techincal simulation, database, sound engineering, video authoring, DTP orimage processing. While we are at it, can you give me one example of an app that "benefits immensily" from SSE or hyperthreading ? 64 bit isnt more expensive in transistor count or application development as those technologies.

> Even a game of 3 CDs would still fit entirely in 2 GB
>addressable memory.

LOL.. for one thing, we have DVD's these days, and secondly, its easy to write a 500Mb program that uses for more than 2 GB (in fact, I could do it in less than 500 Kb). At work I use OLAP tools to generate multidimensional datacubes. Starting with a ~100MB product/sales database, I end up with cubes that are roughly 20 GB. Fully processed and precalculated, they might be a terrabyte or more. Just giving an example, hope you get the point.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 1, 2004 6:46:37 AM

Extending the register set, and dropping virtual 8086 mode to name the most important. BTW, there are not 300 ways to extend a ISA from 32 to 64 bit. What AMD did is completely in line with any other 64 bit ISA extention (with the exception of what I noted above) or even Intel extentin to IA32, and probaly intel's solution wouldnt have been all that different.

All OS (windows i am sure) boot in virtual 8086 mode.nb Reg have nothing to do with the ISA

I dont like french test
Anonymous
a b à CPUs
February 1, 2004 6:57:14 AM

>If Intel really is as bad as your post wants to show it to
>be, then they'd be out of business.

LOL.. read carefully, I'm not saying "they are that bad". And I am *definately* not predicting they will go out of business, or even get into any sort of trouble financially any time soon. I'm just getting fed up with people that blindly take intel's word for granted, and thnk because intel does something in a certain way, then surely it is the right way. Well, wrong, intel can blunder just as well, and with the 64 bit migration they are about to fall flat on their face (killing Itanium chances at anything but HPC and high end server niches, and giving the x86 ISA out of hands to AMD). It will cost them a lot of money IMHO (like not being able to recover those billions invested in IA64), but a few years from here, everyone will have forgotten it, just like we forgot fiasco's like RDRAM.

>They make good processors,

Yes. Even some exceptionally good one's like Banias.

>but like any other company, their policies sometimes aren't
>the best. This is not to justify them, but rather to
>encourage a more balanced attitude.

No need to, I can see things in perspective.

>This "Intel caused the big problems in x86", "Intel did
>this and that fiasco" attitude is exaggerated.

No its not; not in the context where someone bashes AMD's effort to move the ISA to 64 bits. Obviously AMD64 isnt the greatest ISA ever conceived, but saying AMD can't design a proper iSA, while intel surely could come up with one that would ridiculise it, is, well.. not quite objective. All the flaws AMD had to work with or around, are indeed introduced by intel after all.

>One could find problems in AMD's processors and policies as
>well. Remember the old Palominos, with no thermal
>protection whatsoever?

Oh, sure, and plenty more ! Good grief, I'm not claiming everything AMD toches turns gold, but I do give them credit for making a few important and smart decissions lately.

With K8 they did pretty much everything right within their budget limits. Integrated MC, hypertransport, AMD64 and SOI. We've seen what performance the MC gives you, and wait for intel to do the same. Hypertransport is just beatiful, and really for the first time gives them a serious edge in the SMP market. Its the brightest idea i've seen in a long time. SOI is likely to start paying off as well, as intel seems to struggle to clock prescott higher without hitting a thermal brick wall (did you notice they couldnt decrease Vcore going from 130 nm to 90nm ? Did you notice Precott is getting near 100W/cm², 50% more than Northwood and 100% more than K8 ! ).And AMD64 is about to be embraced by intel, so it seems they did at least a few things damn well.

= The views stated herein are my personal views, and not necessarily the views of my wife. =<P ID="edit"><FONT SIZE=-1><EM>Edited by bbaeyens on 02/01/04 05:33 AM.</EM></FONT></P>
Anonymous
a b à CPUs
February 1, 2004 7:00:53 AM

>All OS (windows i am sure) boot in virtual 8086 mode

No, BIOS perhaps boots the cpu in 8086 mode, not the OS. Once you put the K8 in long mode, there is no more virtual 8086 mode. Of course AMD still supports it in legacy mode, and you could perfectly run Dos 3.11 on it if you liked, but unlike in IA32 protected mode, AMD dropped V8086 support in long (64 bit) mode.

>Reg have nothing to do with the ISA

Excuse me ???? Register set is one of the corner stones of a ISA. ever seen assembler code ? its nothing *but* register names.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
February 1, 2004 7:48:27 AM

Oh, and just one more argument in favour of AMD64 (courtesy of Brian Neal at aces):

video card memory is mapped in the upper 2 GB region (reserved for the OS), and in windows, is actually mapped twice. Image using a 512 MB videocard, it would aleady eat 1GB of your apps address space on its own, maybe limiting it to 1.5 GB or so. Now imagine using *two* such videocards (possible with latest AGP specs and PCI express). 256 MB cards are the norm in the high end today (actually, 512+ cards exist for CAD), ironically, 512 MB or more videocards may require a 64 bit cpu and OS.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 1, 2004 8:02:46 AM

Quote:
4) By the end of next year, 80-90% of us running 64 bit capable cpu's (intel or AMD) will run 64 bit windows (or Linux), and run/play several 64 bit games/apps. Those who just got a 32 bit only P4 will be slapping themselves, especially the one's that got themselves a >$1.000 P4EE.


By the end of 2005? Now *that* is a bold prediction. Even assuming Intel does release an x86-64 chip, even assuming Microsoft does get an x86-64 OS out before then, even assuming that Intel *heavily* pushes this, to gain 80-90% market penetration with a new chip line has....never happened before. A year after the P4's release, it was no where near 80-90% of the market penetration.
If you're only speaking of enthusiastes, how many of us run P4C's or better? I'll bet not 80-90%.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
Anonymous
a b à CPUs
February 1, 2004 8:29:03 AM

Euh.. you misread me. I claimed 80-90% of those that have a 64 bit capable chip, would use its capabilities by running a 64 bit OS and apps.

I am not claiming anywhere near 90% of us would have one. Most optimistic estimate would put AMD's product offering at 80% 64 bit capabale by then, and intel maybe 50%, so not even near 80% of new computers would be 64 bit, let alone installed user base. My predictions arent *that* bold (or rather, insane) :) 

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 1, 2004 8:47:37 AM

ROFL

I didnt know bbaeyens was head master of the special olypics this year, man some stiff competition here this year. Do these people believe the crap you spew? You predicted nothing (like you are a prophet or something). I posted here 1 year and 6 months ago the fact that Intel holds the x86-64 bit card and AMD was holding their package. It still stands today. Microsoft made a deal with Intel not to release a x86-64 bit OS until Intel was ready. Not the other crap you been spewing with your fanboy savior angle you been spinning on it.

Go to microsoft.com
search for yamhill in the search box on that page
click the only link returned
search that page for yamhill

Now read, yes it requires some effort.

Bill Gates on the stand telling the grand jury what I just told you above. Not your poorly painted picture.

Now that we all know who hold the cards, you better pray that Intel makes a something compatable or AMD will have some reworking of the instruction set to match Intel.

I am going to need you to step off the soap box please.

<b>"You haven't proven anything that once 64-bit support comes out, it will perform even better." -EDEN</b>
February 1, 2004 9:06:56 AM

I won't buy A64 because I see no use of 64bit in Windows gaming at the moment. The only advantage is the built in memory controller, but unless you have an Athlon FX, you won't enjoy having a dual channel capability.
For most people, the choice of 2004 is either P4-2.6/P4-2.8 with 800FSB and Dual Chanel DDR or Athlon XP 2500+/2800+ with nforce400ultra mb O/Ced to whatever you can. (Or anything faster that dropped to the above price levels)

Anyway we average users have to thank the "pioneers" for testing out defects for us. Thanks!

A fine day!
February 1, 2004 10:45:29 AM

Quote:
Intel started working on Yamhill well before AMD released their specs of AMD64.

Doesn't that open your eyes that AMD64 is a quick hack and Yamhill is most likely superiour?
Quote:
I simply don't believe MS will support two competing x86 extentions, because it is not in their interest to fragment the market; and since much to everyone's surprise, they embraced AMD's solution, I think there is no alternative for intel than either ignore AMD64 and bet on IA64, or go back to the drawing table.

Why would Microsoft be interested only in the technology of AMD? Microsoft has very close connections with Intel and wouldn't just choose for AMD if they haven't looked at Intel's 64-bit design yet. If they release Windows for AMD64, only a marginal fraction of people will buy it. Microsoft has no profit in that. AMD can't supply the whole world anyway. So it's best to wait for Intel and if this means a recompilation they won't cry about it. The only one trying to fragment the market is AMD.
Quote:
Extending the register set, and dropping virtual 8086 mode to name the most important.

Extending the register set isn't suddenly revolutionizing x86. And why would virtual 8086 mode ever have been a problem?
Quote:
IMHO the P4 isnt nearly as new as the K8. It has a looooong pipeline; big deal. And a trace cache. Nice. But no integrated MC, no glueless SMP, no 64 bit extentions, etc, etc. In the end, with its "all new core", its still the inferior product, especially in SMP configurations.

Integrated memory controller is just cut-paste from the northbride. Glueless SMP is just adding more pins. And 64-bit is again just making the registers longer. Never did they truely change the core. And I'd really like to know what you mean with "etc, etc". NetBurst, Hyper-Threading and even the trace cache are bigger changes than AMD ever did for a whole generation.
Quote:
They betted on Itanium, which wasnt a stupid idea at all. In fact, if it werent for AMD and Microsoft working together, they would probably have pulled it off, and in a few years we'd all be running IA64 boxes. It seems they just overestimated Itaniums succes, and underestimated AMD.

It's a shame AMD gets in the way. I mean, AMD is very good at semiconductor engineering and should try to beat Intel at pure performance, not design. They simply can't take over the world with a few hacks. Once they beat Intel at performance for several years and have gained equal marketshare and enough fabs to supply, they can give it a try. All attempt now is just doomed for failure. AMD only recently started making some profit. Do you seriously thing Intel will just bend now?

This discussion isn't about Itanium but anyway: Performance still rises every few months thanks to compiler optimizations and software design that makes better use of the architecture's capabilities.
Quote:
Its not for no reason that intel has gone a different route with with their new server chip (Itanium) and their latest mobile chip (banias/dothan). Both are low clock high IPC braniacs, and not speed demons like netburst.

You can't increase performance of a high IPC chip much if it is limited in clock speed by a short pipeline. A low IPC chip on the other hand just needs more execution units and extended Hyper-Threading to increase it. Then the clock speed is all that matters. Besides, since AMD isn't capable of redesigning the core, all that has kept them alive for the past years is clock increases.
Quote:
Typical workstations are purchased to basically run just one app (that usually costs far more than hardware).

I wasn't talking about what you use a workstation for. They do get sold with 4, 8 and even 16 GB of RAM. Besides, even one application can consist of many processes.
Quote:
If that is true, why has every other server ISA migrated to 64 bit nearly a decade ago then ? Do you still see any 32 bit Power, SPARC, or PA Risc systems being sold ?

That's mostly because they need 64-bit arithmetic, not 64-bit addressing. If you have a database of several terabyte the indexes (not necessarily pointers) could easily exceed 32-bit range. There would actually be little problem doing it on a 32-bit CPU. For desktops, neither 64-bit arithmetic or addressing is really critical yet. That's why AMD64 is just a hype and Intel has plenty of time to finish a brand new 64-bit design.
Quote:
While we are at it, can you give me one example of an app that "benefits immensily" from SSE or hyperthreading ?

MP3 and DivX playback and processing both use MMX extensively. SSE is used in many scientific applications including Matlab. And both are also used by DirectX and graphics drivers. By the way what's your point with this question?
Quote:
At work I use OLAP tools to generate multidimensional datacubes. Starting with a ~100MB product/sales database, I end up with cubes that are roughly 20 GB. Fully processed and precalculated, they might be a terrabyte or more. Just giving an example, hope you get the point.

At work, indeed. We're still talking about desktop CPUs here. Besides, precalculating everything is often not the fastest solution since you have to go to disk so recomputing things is often faster, except when you have 20 GB of RAM but no desktop user has that. And I already showed that real-time applications should not process more than 100 MB per frame, because of bandwidth limitations. 64-bit processing really doesn't mean a lot for desktops right now. I'm still waiting for that example of yours where 64-bit addresses is almost a necessity...

Again, don't get me wrong. Eventually 64-bit will be needed. I just thing AMD rushed it -only- to gain market share with the hype.

<P ID="edit"><FONT SIZE=-1><EM>Edited by Whisper on 02/01/04 08:12 AM.</EM></FONT></P>
February 1, 2004 11:01:50 AM

Quote:
That's mostly because they need 64-bit arithmetic, not 64-bit addressing.

I can confirm that. I work with scientists, and they couldn't care less about their CPUs doing 64-bit integer work; what they want is 64-bit arithmetic (FP) work, and fast one at that. But... current 32-bit processors can use 80-bit FP... So this is not really an advantage anymore. And it was the biggest thing that made everyone go 64-bit... it was not the superior adressing capabilities at all. But there are still some people interested in those, mind you.

Anyway, I do think that once these 64-bit extensions become widely available, people will use them more, but it will take quite a while.

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
a b à CPUs
February 1, 2004 11:41:51 AM

I was on the Windows Update site looking arround and found an Update for XP 64bit. Not AMD64XP or Intel64XP, but just plain XP64.

So it appears both will have 64bit processors and the Alpha or Beta version is already out to the testers.

I aint signing nothing!!!
Anonymous
a b à CPUs
February 1, 2004 11:42:58 AM

> You predicted nothing

Ahm.. I proved some clicky links for ya.

> I posted here 1 year and 6 months ago the fact that Intel
>holds the x86-64 bit card and AMD was holding their
package.

And what is that supposed to mean ?

>Microsoft made a deal with Intel not to release a x86-64
>bit OS until Intel was ready.

Well, if true (big if) what exactly would this mean according to you ?

>Now read, yes it requires some effort.

Thats a mighty long page, I have no interest in reading it. Just quote whetever you think is relevant. I see this, but fail to see what is your point:

Quote:
you told him that Intel was going to try -- that you
22 feared Intel was going to try to squash your achievement in
23 Hammer by announcing that Intel 2 was going to develop a 64-bit
24 extension code name Yamhill. Y-a-m-h-i-l-l I believe, correct?
25 Is that correct?

3762
1 A. Yes.
2 Q. And then you told Mr. Gates in this same call that you were
3 not interested in having Microsoft announce support of Intel's
4 Yamhill before Microsoft announced support of AMD's Hammer;
5 correct?
6 A. Yes. There was no Yamhill.
7 Q. But you told Mr. Gates in this call that you were not
8 interested in having Microsoft announce support for Intel
9 before Microsoft announced support of AMD's Hammer, did you
10 not, sir?
11 A. I asked Mr. Gates to hold Intel to the same standard he
12 held us and not announce public support until he had seen a
13 working platform with real silicon executing code. So that
14 would imply he could not announce it before us because Intel is
15 years behind us.


This only seems to enhance my point that yamhill was supposed to be different from AMD64, while intel is now backpedalling to make it AMD64 compatible since MS chose to support it, and not Intel's standard (or rather, intel initially didnt want to release Yamhilla at all, betting on ITanium, hence MS went the AMD route).. Again, what is your point ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
February 1, 2004 12:24:09 PM

>Doesn't that open your eyes that AMD64 is a quick hack and
>Yamhill is most likely superiour?

Ahem.. no. Did I guess when AMD started working on AMD64 ? Could have been 1980 for all I know. Of course you cant publish a spec until it is completely done. Think mate, think...

>Why would Microsoft be interested only in the technology of
>AMD?

Simple. Intel may have had Yamhill in its labs, but clearly didnt want to release it until Itanium sales took off.. or rather, not *at all* as not to jeopardize Itanium. Microsoft likely just hedged their bet with AMD64, in case Itanium didnt take off, and would never make it to the desktop. Nothing more, nothing less.

>Integrated memory controller is just cut-paste from the
>northbride.

Sure...

>Glueless SMP is just adding more pins.

Sure ! A few pins !! Well, S775 then surely has enough pins to support this, wouldnt you think ?

>And 64-bit is again just making the registers longer.

sort a.

> NetBurst, Hyper-Threading and even the trace cache are
>bigger changes than AMD ever did for a whole generation.

Using your level of technical genius, I'd say its just a few more transistors.. and a marketing name for a longer pipeline.

>You can't increase performance of a high IPC chip much if
>it is limited in clock speed by a short pipeline. A low IPC
>chip on the other hand just needs more execution units and
>extended Hyper-Threading to increase it. Then the clock
>speed is all that matters. Besides, since AMD isn't capable
>of redesigning the core, all that has kept them alive for
>the past years is clock increases.

Sheesh.. you should apply with Intel, maybe you can design their next generation chip ? This is so utterly nonsensical I will not even bother trying to correct you. In fact, I think this is where I end this CPU design discussion, it goes nowhere. I'll give you one hint though.. have a look at the architecture of the fastest cpu's available.. Alpha, Power, Itanium, Opteron.. all of them are braniacs. Coincindence ?

>I wasn't talking about what you use a workstation for. They
>do get sold with 4, 8 and even 16 GB of RAM. Besides, even
>one application can consist of many processes

Yeah, and 99% of these workstations (with more 4 gigs) are 64 bit machines (Sun, Power/RS6000, PA Risc, even Itanium..).

>That's mostly because they need 64-bit arithmetic, not
>64-bit addressing. If you have a database of several
>terabyte the indexes (not necessarily pointers) could
>easily exceed 32-bit range. There would actually be little
>problem doing it on a 32-bit CPU

I have no idea what you are trying to say here (and I doubt you have), but if you are claiming you could address more than 4 GB without 64 bit addressing, but with 64 bit ALU, you are desperatly wrong. And if you think more than 1% of desktop or server apps have any use for 64 bit integer math, you are even more wrong.

>By the way what's your point with this question?

No one doubted the usefullness of MMX or SSE, while everyone whines about 64 bit being unecessary, that is the point. MMX and SSE help speed up certain types of code, even though nothing a faster cpu couldnt do, while 64 bit support allows you to work with apps or datasets you simply will never be able to run/use on a 32 bit cpu. And in the case of AMD64, it speeds up certain types of code. Supporting 64 bit is easier than SSE (with the exception of the OS), and it hardly costs more to implement in silicon either. So what the f*ck is wrong with that ?

>At work, indeed. We're still talking about desktop CPUs
>here.

YOu really need me to give you a link so you can download a 100 meg game demo that eats up 500 Mb RAM when you run it ?

>And I already showed that real-time applications should not
>process more than 100 MB per frame, because of bandwidth
>limitations

You showed nothing of that sorts. Your calculation was pretty meaningless as you forgot the GPU is doing most of the rendering, and has its own fast memory to buffer geometry and textures.

>Again, don't get me wrong. Eventually 64-bit will be
>needed. I just thing AMD rushed it -only- to gain market
>share with the hype.

You need chips out in the market to get the software coming. If AMD and intel had waited until 2007 like would have according to their comments, it would take another 3 years, so until 2010 for apps to come on the market, and we'd be banging our heads against the wall with these tiny 2 GB address spaces while we'd have 16 GB Ram. It would EMS all over again. In fact it is already now. BTW, you do realize intel extended physical addressing of the Pentium Pro to 36 bits back in 1995 ? Almost TEN years ago ? Why do you think they did that, cause no one would ever need for another decade and a half ? 64 bit is overdue if anything, definately not premature.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 1, 2004 12:26:52 PM

Exactly...

This is all very interesting, but it doesn't tell you if a dual Opteron system is significantly better value than a (cheaper) dual Xeon system.

If 64 bit versions of MAX, Maya, AutoCAD etc. are not going to be with us for XX months, then why should we be rushing out to buy 64 bit CPUs (which will be cheaper in XX months)?

"Some mice have two buttons. Macintosh has one. So it's extremely difficult to push the wrong button." - Apple ad. circa 1984.
February 1, 2004 12:34:38 PM

Quote:
If 64 bit versions of MAX, Maya, AutoCAD etc. are not going to be with us for XX months, then why should we be rushing out to buy 64 bit CPUs (which will be cheaper in XX months)?

Why should we indeed? I know I'm not. Clever thinking of yours. I do not need 64 bit in my desktop <i>right now</i>, nor do I have money lying around for an upgrade.

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
Anonymous
a b à CPUs
February 1, 2004 12:36:54 PM

Carefull there, if I'm not mistaken "XP 64 bit" is actually the Itanium version. The AMD64 version is called 'for extended systems' or something. But sure, there will be only one 64 bit x86 OS from microsoft, only a fool would believe otherwise. And it is in beta btw, due to be released second half of this year. No secrets there.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
February 1, 2004 12:39:40 PM

>his is all very interesting, but it doesn't tell you if a
>dual Opteron system is significantly better value than a
>(cheaper) dual Xeon system.

The benchmarks are all over the web; just pick the app you will be using, and buy whatever suits you.

>If 64 bit versions of MAX, Maya, AutoCAD etc. are not going
>to be with us for XX months, then why should we be rushing
<out to buy 64 bit CPUs (which will be cheaper in XX months)

You shouldnt. And I compeletely agree with anyone buying cheap 2500+ or P4 2.4C systems. But if you are going to chose between a >$500 32 bit chip and a >$500 64 bit chip that perform roughly on par on 32 bit code, I'd know what to chose.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 1, 2004 12:43:45 PM

You're right, they will only sell one Windows XP with 64 bit extensions. However, like with HT, the installer will determine exactly what kernel you will use - HT-enabled computers will typically require a slightly different install.

In that sense, MS could sell XP64 for Intel and AMD both... If they've got two different kernels for each. Both of these kernels would be in the installation CD...

Just a suggestion.

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
February 1, 2004 12:45:27 PM

>$500 on a processor? Who the hell has this kind of money?... :eek: 

Not me, for sure. :frown:

:evil:  <font color=red><b>M</b></font color=red>ephistopheles
February 1, 2004 1:14:38 PM

Im not sure I make 5k a month. F@ck I dont know if Ill be able to buy one. Better start bringing in my beer bottles.

-taitertot

If this post has attitude, seems to be overly aggressive, rude, distasteful to 99% of the users here, and shows a zealous defense of Intel... It’s probably Spud.
Anonymous
a b à CPUs
February 1, 2004 1:55:59 PM

I guess I can't rule that out completely, but they would have to be binary compatible on the application level, which qualifies as "compatible" in my book.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 2, 2004 2:42:53 AM

Quote:
Doesn't that open your eyes that AMD64 is a quick hack and Yamhill is most likely superiour?

Trouble with your logic is that you neglect the political forces at work within Intel. Remember, the IA64 and Yamhill teams got separated, and Yamhill shuffled off into obscurity, so that Intel could devote most of its resources to IA64.

Then IA64 laid an egg.

Now Intel has to pull out Yamhill, dust it off, <i>and</i> rework it to make it AMD64-compatible (per MSFT's ultimatum).

That may help you understand why Yamhill took (or is taking) so long. It's not because Intel "took the time and did it right," it's because Intel wasn't planning on ever doing it at all.

As for AMD64 being a hack, remember that the x86 ISA has been a hack since its inception. Intel threw the crappy thing (8086) together in three weeks after their first proposal got shot down. Then Intel had to strip the turd down to half-bandwidth (a la 8088) to make it cheap enough to achieve market penetration. The rest is history.

Now we're kind of stuck with it, and thrice-extending it isn't really meant to make it cleaner, no matter who does it.

AMD64 does add a few critical features. 64-bit addressing is critical for databases and massive thirty-levels-deep calculation apps. Extra GPRs are something x86 has sorely needed since the beginning. Execute permissions on memory are something vastly more critical than you might think. Most of the other messiness of x86 is so out-of-sight anyways that practical-minded people largely just don't care.

Quote:
Why would Microsoft be interested only in the technology of AMD? Microsoft has very close connections with Intel and wouldn't just choose for AMD if they haven't looked at Intel's 64-bit design yet.

Remember, Microsoft DID look at Intel's 64-bit design. Intel's 64-bit design was IA64. IA64 laid an egg. MSFT got dreadfully insufficient ROI on Windows-IA64 and was understandably annoyed.

Quote:
Extending the register set isn't suddenly revolutionizing x86. And why would virtual 8086 mode ever have been a problem?

It's meant to be evolutionary, not revolutionary. It's just that the jump from 32 bits to 64 bits is a big deal for a lot of professionals. Being able to make the jump without paying more than your basic x86 price or making a huge ISA transition--now that's what kicks ass.

Quote:
Never did they truely change the core.

What part of the core really needed changing? K7 was a pretty good core architecture to begin with. All that was really needed was to iron out its major weak spots, which AMD did.

Quote:
NetBurst, Hyper-Threading and even the trace cache are bigger changes than AMD ever did for a whole generation.

There's something to be said, though, when all that extra work isn't quite enough to keep the P4 in a performance-leading position. That something is "Keep it simple, stupid!" :wink:

<i><Lionel Hutz> I'll be defending...The SCO Group!!!??? Even if I lose, I'll be famous!</i>
February 2, 2004 3:25:09 AM

No, BIOS perhaps boots the cpu in 8086 mode, not the OS. Once you put the K8 in long mode, there is no more virtual 8086 mode. Of course AMD still supports it in legacy mode, and you could perfectly run Dos 3.11 on it if you liked, but unlike in IA32 protected mode, AMD dropped V8086 support in long (64 bit) mode.

No OS boot in virtuakl 8086 and swith to 32 bit mode at the start but.Anyway there 8086 mode and vitual 8086 mode is not the same thing by the way.

I dont like french test
!