Sign in with
Sign up | Sign in
Your question

Is the Hammer architecture too conservative?

Tags:
  • CPUs
  • Product
Last response: in CPUs
Share
July 29, 2002 2:10:42 AM

I’ve been reading up around the Hammer architecture this weekend and I’m must be turning geeky because the damn thing excites me. It seems such an elegant and simple evolution from x86 to x86-64.
Taking a look at the pdf from AMD’s site got me wondering about the gains that would be made moving from a 32 to 64 bit OS on a ClawHammer. I’m thinking here specifically about the gains that a desktop user can expect and not what a Workstation or Server will gain.

It seems as if many of the performance gains that the architectural changes will give will also be available even under 32 bit Windows. This is obviously to be applauded and is needed as 64 bit seems a long way off anyway. But then I started thinking well if most of the gains are already here and now where will be the incentive to move to 64 bit down the road? Am I missing something?

First off, I’m not an architectural know all but the purpose of this post is to learn more about how the shift to 64 bit on the Hammer architecture will improve performance.
I looked at 2 AMD pdf files and a few technical web site reviews and that’s made me come up with the following observations regarding the gains made moving to 64 bit on the hammer architecture:

1. 64 bit integers can be handled natively.
2. Addressable memory increases beyond 4 GB.
3. There are twice as many 64 bit General Purpose Registers than 32 bit.
4. Twice as many 128 bit SSE registers available under 64 bit operation.

1. As a programmer of multimedia apps I’ve never found the need to use 64 integers which are currently available under 32 bit Windows albeit with a performance hit. Improving floating point (FP) performance is generally more important which is why MMX never took off (integer only) and Intel introduced SSE which added FP support and improved FP support with SSE2. It’s great to see Hammer supporting SSE2. The FP/MMX registers remain 64 bit. No doubt some niche applications will greatly benefit from this feature but I’m curious to know if there are mainstream uses for this?
2. Even if you extrapolate in a linear fashion I can’t see that this limit is going to be reached much before the end of the decade. And at some point things aren’t going to keep on expanding in a linear fashion and will tail off. Although, it might be useful to be able to store a whole MPEG2 film in memory for processing J
3. This sounds more useful, can someone explain to me the impact that this will have on performance?
4. Again this sounds promising but I’m not sure how much this ties in with the move to SSE2 support? Anyone have a comparison with what the Pentium 4 offers?

I’m left with the conclusion that the two most important changes to impact performance might well come from items 3 and 4 which seem to be independent of the move to a 64 bit architecture. In other words AMD could have created a new 32 bit architecture that incorporated these changes instead. Not that I’m suggesting they do that, just an observation.
Am I missing something here or are there other factors that will increase performance when moving from 32 to 64 bit? Remember I’m not talking about architectural changes which are independent of the OS being 32 or 64 bit such as the on-chip memory controller and pipe-line changes.

You could even view the move from Athlon to Hammer as more of one to a better architecture that so happens to be 64 bit as well for future proofing!
Hell, I’m all for a move to an updated architecture at some point particularly as AMD seem to be offering a seamless upgrade path. But one thing worries me. Is this new architectural change radical enough?
Apple made a big jump when they moved from 6800x (if I remember that correctly) to PowerPC but the improvements were also big. Being first to market @64bit may hurt AMD if they aren’t aggressive enough. Imagine a scenario where Hammer is a great success @32bit but 64bit Windows doesn’t take off for say 5 years. This gives AMD a massive user base of 64bit compatible systems ready to make the move down the line. In the meantime AMD evolves the x86-64 performance in the same way that it has x86 but these improvements will need to remain x86-64 compatible of course which will limit their choices.
In the meantime Intel has all this time to take a more radical approach. Maybe they get their stuff together with Itanium and can turn that into a desktop processor? Or maybe they just take another route completely and not being tied to x86 they come up with something radical that blows x86 and x86-64 away.

The fear is that a Pentium 6 running Windows 32 could be as fast as a Clawhammer 3 running Windows 64. In other words there’d be no compulsion to jump ship to 64bit or in other words to AMD. Then Intel bring out Itanium 4 for the desktop @64bit and wham there’s a massive performance boost and a reason to finally move to 64 bit.

This brings me back to my original question which could be rephrased as, “Any idea of the performance gain for Hammer when using a 64 bit OS?”.

Anyway I’m still looking forward to the release of Hammer which makes a change for a CPU as Intel’s original releases of a new CPU family tend to seriously under whelm.

:)  Crow

More about : hammer architecture conservative

July 29, 2002 2:58:25 AM

There not official compiler or annouce OS 64 bit for hammer but they have 1 to 2 year to put good OS upport and driver support and workstation aps for there 64 bit.They dont need 1 right they will be able to sold it like a 32 bit cpu.

SSE 2 coming more and moree mainstrean thank to intel so they dont need to do that job.

they biggest gain com from integrate MCH so it cut latency but still have 1/2 of bandwith compare to Intel presscot/springdale but cost less to be done.they 1 hick they will not be able to run faster memory like DDR-2 so intel will be the 1 on that.

Core little work but it dont take many time to devellope so it can allready start working and improve hammer and K9.

They have improve branch prediction so they extra pipeline add will be negate on IPC but have will be more scale able for to the P4 but have a higher IPC.

86-64 have moe marketing purpose that real performance increase as i read they have add 8 register 64 bit wide and standart register i dont know if they been improve to 64 bit wide so tiny compare to 256 128 bit wide register of a IA-64 but have way less transistor.AMd have 1 fab dresden so if they want to be able to the market they must keep small die size until 2004 when they will move 300 mm wafer.


So i dont call that conservative, they have do what they can.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
July 29, 2002 4:02:57 AM

Microsoft have announced support for Hammer with Windows at least. Not sure for which versions though, there are so damn many new ones on the horizon these days. Also a number of flavours of Linux have been confirmed. AMD have even shown Hammer in public running Linux 64 and Windows 32.

<< 86-64 have more marketing purpose that real performance increase>>

Yeah, it’s beginning to smell a bit like that from where I’m sitting.

All 16 of the general purpose registers are fully 64 bit but only half of them can be addressed in 32 bit mode.

<< So i dont call that conservative, they have do what they can.>>

Sure, I was just trying to look at it from the perspective of where the real gain lies with Hammer.

:)  Crow
Related resources
July 29, 2002 7:31:19 AM

The x86-64 component is sort of added there for the server/workstation market. Remember, the Hammer design isn't only for the desktop market but for the workstation/server market as well.
AMD doesn't have the resources to independently design 2 completely different core designs like Intel does so their approach is sort of a 1 size fits all kinda thing. Hammer will work well as a desktop 32-bit processor and as a server 64-bit processor.
July 29, 2002 7:01:23 PM

I don't know why but it would feel so warm and fuzzy inside if I had a 64-bit-capable chip inside my comp! Just the thought of having something the bastard workstation companies always get, is soothing! :smile:

--
The sound of determination is the echo of will...
July 29, 2002 8:29:27 PM

That's yer choice. Although personally I feel warm and fuzzy if there were a hot woman in my bed naked, but to each his own I guess.
July 29, 2002 10:55:01 PM

The big gain come from the better brach prediction integrade memory controleur SSE2.

To my own calulation hammer will

10% clock for clock for MCH
15% clock for clock for all other improve not more that palamino over thunder bird
-5% less vs tbred512 less L2 cache
X % due to clock speed

So intel need about 20% faster clock for clock vs nw to be equal 25 % made it faster that clawhammer 30% for beat it when on 64 bit.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
July 30, 2002 12:06:27 AM

I'm not exactly sure of this but I think AMD's official statement is to expect a 10-15% per clock gain in Hammer over the t-bred total. That is, accounting for the onboard memory controller and the scheduling enhancements. I'm not sure how much the cache will help as the design isn't very cache dependent anyway.
The rest of the 10-15% (with a total of 30% gain over the current Athlon) is said to be due to the x86-64 extensions. So for modern code, I wouldn't expect more than a 20% per clock gain. Now mind you that is no small gain. Combined with higher clockrates Hammer can be a significant increase over the K7 designs.
As for comparison with Intel products. Well, only time would tell. The P4 is scaling very well and will probably hit the neighborhood of 3.2ish GHz by the time Hammer is out. Assuming Hammer is able to gain a total of 40% improvement (accounting for the 2.0 GHz clockspeed) over the current 1.8 GHz t-bred, that still might not put it over the 3.2 GHz P4. Currently, the 1.8 GHz t-bred is about equivalent to a 2.4 GHz P4 (a little less on average). A 3.2 GHz P4 would provide a 33% linear improvement in all code than a 2.4. Assuming Hammer does live up to the 40% higher than a 1.8 t-bred assumption, that still wouldn't put it clearly past a 3.2 GHz P4. It may not even be past at all.
July 30, 2002 12:12:42 AM

I recall AMD stating something to the effect that Hammer will be roughly 35% faster IPC in 32 bit and upwards of 50% faster when running 64 bit

Mark-

<font color=blue>When all else fails, throw your computer out the window!!!</font color=blue>
July 30, 2002 3:25:42 AM

That's what I also read. 10-15% seems rather small for an on-die mem controller and new core improvements.

"sigh", I wish AMD would post some news reassuring us about Hammer's performance. All it seems is that it will not be no K7, and in the end will feel like an AXP speed bump, a bit bigger than usual.

--
The sound of determination is the echo of will...
July 30, 2002 5:09:11 AM

Well, it may or may not be 30% just for the 32-bit enhancements. Although I doubt it. Who knows, maybe the scheduling enhancements and improvements in the caching system really does help that much, that plus the onboard memory controller really does improve per clock performance that much. Although I don't think so. The K7 (and ultimately K8) designs are not that cache or memory starved. I actually think the biggest improvement would come from the scheduling enhancements and that along may improve per clock performance a good 10%.
July 30, 2002 5:36:17 AM

A 20 to 25% improvement in 32 bit mode seems achievable from what I’ve gleamed from analysing various semi-official sources, wild guesses etc.
Nobody seems to be saying much about what the gains in moving to 64 bit will be although I guess that’s looking maybe too far down the road.

I wonder if the PR rating system for ClawHammer will be changed so that it reflects a relationship to the next generation Intel Prescott chip?
Because even if AMD do claim that their chips are rated in relationship to previous Athlon models we all know that the real reason for PR ratings is to show them in a truer light in relationship to P4s.
I think they’ve handled the PR ratings pretty well up until now by being conservative enough to stay ahead of Intel even with the release of the Northwood core.
I guess the big test is in the PR rating for ClawHammer. It’ll be released probably over 6 months ahead of Prescott (Pentium 5?) which might make it difficult for AMD to call.
If they’re too conservative, ClawHammer’s PR rating might not look attractive enough against a P4 3.0 even if in reality it licks its ass. If they’re too aggressive with the rating and Intel actually make Prescott a bigger architectural leap than they usually do then AMD might end up with egg on their face. If a Prescott 3.2 whipped a ClawHammer 3.2+ then PR will become to be known as Public Relations as in Public Relations Disaster.

All of which makes me wonder how AMD will price the ClawHammer? It’s easy enough to say, oh, somewhere between the priciest current Athlon (Barton) and the priciest Pentium 4 (3.0!) but that’s probably going to be a pretty wide spread to work with. Unless of course Intel pull a stunt like they’ve done before and release the 3.0 at a cut down price and squash the price range that they sell chips at for a while. I’m sure they’ve done this post Athlon although I forget which chip it was, it could have been a P3 even. Whatever it was it was released at half the usual price that Intel release new fastest chips at although I can’t remember how long they kept that strategy up?

With Intel aggressively ramping the Northwood rollout it could play into AMD’s hands in the sense that Intel will have very little to release between Northwood 3.0 and Prescott 3.x six months + later. Hell of an interesting duel developing here, maybe George Lucas should turn it into a trilogy, it couldn’t prove to be much more tedious than his current one!

Jeez, how’d want to be in the chip business man? Cut throat world.

:)  Crow
July 30, 2002 6:14:21 AM

I recall AMD stating something to the effect that Hammer will be roughly 35% faster IPC in 32 bit and upwards of 50% faster when running 64 bit

35% come from where ok i can found for the first 20% and a others 10% for 64 bit but the rest it overhype.Intel will release there presscott after clawhammer even if they release in 23000 they will clock for beat it or improve the core. For the (2Ghz) it still running at 800 mghz 0.13 micron is hard to get only intel have it on good scale.TSMC still on testing UMC far from it AMD no need to reminde you T-bred.

Most will consider that like trolling.
hammer still feature cheat memory controleur hub compare to intel also a cheat southbridge compare to ICH5.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
July 30, 2002 7:17:30 AM

((((1. As a programmer of multimedia apps I’ve never found the need to use 64 integers which are currently available under 32 bit Windows albeit with a performance hit. Improving floating point (FP) performance is generally more important which is why MMX never took off (integer only) and Intel introduced SSE which added FP support and improved FP support with SSE2. It’s great to see Hammer supporting SSE2. The FP/MMX registers remain 64 bit. No doubt some niche applications will greatly benefit from this feature but I’m curious to know if there are mainstream uses for this?)))

You know That sounds like the People in the Early 80s saying why do I need 32 bit? Can I ask you a Question now. Can you live running 16 Bit After use useing 32 bit for so long? My answer is this NO. And Give it 3 to four years. You will be going I can t live with out it 64 Bit.
July 30, 2002 8:06:55 AM

<< You know That sounds like the People in the Early 80s saying why do I need 32 bit? Can I ask you a Question now. Can you live running 16 Bit After use useing 32 bit for so long? My answer is this NO. And Give it 3 to four years. You will be going I can t live with out it 64 Bit.>>

For me the move from Windows 3.11 to NT 3.51 was a blast as I gained a 32 bit OS and freed myself of the DOS/16 bit Windows quagmire in one fell swoop. So happy memories there.

Hell no, I don’t have enough information to have an opinion either way on whether the move to 64 bit is relevant for desktop users which is why I posted this in the first place. The whole thrust of this post is that most of the gains that the ClawHammer seems to offer are independent of the move to 32 bit. Of course it makes sense that if you’re building a new architecture at this point in time that you make it 64 bit. What I’m getting at is if AMD had say built a new architecture that was exactly the same as ClawHammer but purely 32 bit (i.e. it still had twice as many general purpose registers and double the SSE registers etc) would it be significantly slower?

I’m questioning the need for 64 bit integers as I’m ignorant to how important they are. I want to know! I’m not questioning the move to 64 bit although from my current level of knowledge on the issue it doesn’t seem a radical step. It’s certainly not in the same league as the move from 16 bit to 32.

Not everything scales in a linear way. In fact most systems are probably non-linear. Eventually you reach a point on the curve when there’s little to be gained from increasing a particular parameter, in this case we’re talking bit size.

As an analogy if you move from 8 bit audio to 16 bit the sound quality improves dramatically. Whereas the move from 16 bit to 24 bit audio is not remotely as noticeable to the ear although it’s certainly measurable. And that’s only a 50% increase in bit depth. The difference between 16 and 32 bit audio would I guess be even less noticeable relative to the percentage increase in bit depth.

3 to 4 years you say! I wonder how many people will even have started using 64 bit Operating Systems by then? It’ll be interesting to see how it pans out.

:)  Crow
July 30, 2002 9:40:14 AM

One thing is this If amd dont do it When would of Intel do it? I think the 64 bit would be cool. Let see what it brings
July 30, 2002 3:14:44 PM

Intel are sure going to want to join this party at some point. If they get the timing right then they’ll turn up just when the party starts to take off. It’s a bit of a drag getting to a party too early before the real action starts. Let’s hope for AMD’s sake that Intel don’t gatecrash this one, busting in with a case of Dom Perignon to AMD’s sparkling white!

Crow
July 30, 2002 6:03:57 PM

?4 offer almost nothing over 32 except 64 bit adressing.A Xeon can use up to 64 gb of memory with AEW adressing but it slow prossces.So what will happen in 4 year when standart app like photoshop.Pro will want more that 4 gb and not intel slow AEW so they have 2 option Ia-64 vs X86-64.So who will get the market.I wish Ia-64 but i know that many will not buy new software even if there a IA-64 version.Intel have few option made a deal out of amd to concede victory move to the X86-64, 2 made a deal out with amd licence IA-64 to all others player for free and little extra to amd, 3 push as hard as you can on Ia-64, 4 Put tejas project on top priority.
So there many option and intel can wait others 2 or 3 year before made a decission.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
July 31, 2002 3:46:00 AM

Those trying to compare the move from 16-bit to 32-bit to the move from 32-bit to 64-bit really need to learn exactly what the bitness is about. The effective usefulness of increasing the "bitness" of a CPU is not linear. That is, the move from 16-bit to 32-bit didn't just double performance in applications that used 32-bit.

The reason software needed 32-bit processors is because 16-bit integers were far too small. 32,000 is way too small. 2.7 million is much better and most modern applications doesn't use anything beyond that. Will software in 3 years use it? Maybe. But the move has been towards more FP intensive applications, not integers, so expanding the integer data type isn't exactly gonna be a huge jump.
July 31, 2002 4:07:16 AM

I speak for 64 bit memory adressing.

and what mean 64 bit cpu even itanium is not a 64 bit cpu.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
July 31, 2002 5:57:49 AM

Itanium is indeed a 64-bit CPU just like Hammer. They both have 64-bit general purpose registers, which is what defines the CPU's "bitness". This means it is possible to use 64-bit memory address (although it is not always the case that they are used) and use any other type of 64-bit data, be it integer or FP.
July 31, 2002 6:41:32 AM

itanium use 41 bit adressing and 54 bit virtual adressing
128 bit wide register static or general purpose.

This work for 41 bit instruction.itanium allwayse work with bundle 3*41 bit +5 template for 128 wide instruction.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
July 31, 2002 11:01:24 AM

I don't actually think so. <A HREF="http://www.extremetech.com/article2/0,3973,306,00.asp" target="_new">This article at extremetech</A> along with many others I've read seem to indicate IA-64's general purpose registers are indeed 64-bits (plus 1 bit) wide.

You're thinking of the instruction length, which is 41 bits wide and, due to Itanium's VLIW-style, is processed in bundles. Unlike x86, IA-64 is a fixed instruction-length architecture so instruction size can be specified. That still doesn't change the fact that the addressable data types in the general purpose registers are 64-bits wide, making it, by definition a "true" 64-bit processor.
August 1, 2002 9:08:36 PM

There’s quite a mish mash of bit sizes in modern PC architectures which makes it difficult for Joe Public to determine what performance they’re getting for their cash.
It’s certainly more confusing than the megahertz race as AMD helpfully compares Athlons to P4s with their PR rating. I think they were right to do that as long as they keep the ratings conservative enough so that they don’t overstate their case.
But it’s not surprising to see some vendors advertising P4s as being genuine 2 GHz chips unlike chips from other companies. Marketing folks don’t seem to think twice about playing on the ignorance of the public and perpetuating said ignorance.

As architectures change I suppose PR ratings can still be used to compare say Win-64 on ClawHammer versus Win-64 on an alternate Intel 64 bit architecture be it IA-64 or a separate Intel x86-64 initiative. But hold on, this whole PR thing would become a quagmire if Intel introduced IA-64 for the desktop as it would probably run at clock speeds less than ClawHammer. Can you imagine AMD releasing a 3 GHz ClawHammer with a PR rating of 2,000+ because they were comparing it to Itanic? Ha ha, the whole shebang would fall apart. I suppose Intel could release their own PR rating which compared Itanium with Prescott and then AMD could release another PR rating comparing Clawhammer with the Itanium PR rating. Is that clear??? Jeez, what a mess.

As for the definitive definition of a 64 bit architecture ultimately who cares? It’s performance that counts after all surely as well as backward compatibility and future scalability? Personally I don’t think that Itanium’s 50 bit physical addressing and 64 bit virtual addressing (taken from the Itanium 2 datasheet) should preclude it from being described as a 64 bit architecture. Although a purist may disagree that just seems pedantic to me and I’m a big fan of pragmatism. I’m not sure if they saved much in resources by restricting physical addressing to 50 bit but that’s still one hell of an addressable range, ~ 1,000 TB.

Going back to the mish mash of bit widths in current 32 bit PC architectures in a P4 system for example we can have anything from a 16 bit memory interface with RDRAM to the 256 bit interface of the L1 cache (and possibly also L2, I’m not clear on that).

Itanium is radical not because it’s 64 bit but because of it new instruction set. A 32 bit Itanium would be almost as radical although a wasted opportunity because of the market it’s in.

Bit width is just one factor to use in analysing the raw performance of a system. Obviously you need to look at the frequency of a particular buss and then you have to consider whether it’s dual or quad pumped a la DDR or the P4 buss. Then you have multiple interfaces to the same buss a la RDRAM and some DDR solutions.

The marketing guys are going to be creaming it over the number of opportunities they’ll have to horse-crap with 64 bit this and that. If you ever get the chance to hear a Bill Hicks rant against marketing folk then I can heartily recommend it.

The reason I posted the original question; ‘Is the Hammer architecture too conservative?’ is because although it looks an interesting x86 32 bit processor and also from what I can tell shows great potential as a 64 bit x86-64 SMP solution, as a new platform for the 64 bit desktop I wonder if it’s radical enough?

It has 16 64 bit general purpose registers which is double that of the Athlon but IA-64 has 128 registers? It still has I think 64 bit data paths when Itanium and even the Alpha used 128 bit. Maybe AMD will end up winning support in the SMP marketplace but losing big time on the desktop which will leave them in the same position that the Alpha was at a number of years ago. Which wasn’t good enough for its owners by then, Compaq, to get (pay!) MS to support it under Windows 2000. Bye bye Alpha.

In case some people are reading all this and determine that I’m an Intel fanboy let me be clear that I’m a hardware agnostic. On the PC the only long term relationship I have is with Windows itself but that’s more of a habituation than anything else. We aren’t exactly married but parting at this point would be like splitting with a partner that you’ve lived with for 8 years. Not quite a divorce but not much easier. Once Linux dons a blonde wig and better makeup I might be much more tempted to jump ship.

:)  Crow

I thank God everyday for making me crazy, otherwise I’d be insane like most of the rest of ya.
August 2, 2002 1:54:14 AM

Quote:
As for the definitive definition of a 64 bit architecture ultimately who cares? It’s performance that counts after all surely as well as backward compatibility and future scalability? Personally I don’t think that Itanium’s 50 bit physical addressing and 64 bit virtual addressing (taken from the Itanium 2 datasheet) should preclude it from being described as a 64 bit architecture. Although a purist may disagree that just seems pedantic to me and I’m a big fan of pragmatism. I’m not sure if they saved much in resources by restricting physical addressing to 50 bit but that’s still one hell of an addressable range, ~ 1,000 TB.

It really matters when you are trying to claim you are a "64-bit processor". Just as you can't claim a 2.0 GHz processor is really 5 GHz.

Quote:
Going back to the mish mash of bit widths in current 32 bit PC architectures in a P4 system for example we can have anything from a 16 bit memory interface with RDRAM to the 256 bit interface of the L1 cache (and possibly also L2, I’m not clear on that).

The memory bitwidth is far different than the processor's addressing capability.

Quote:
Bit width is just one factor to use in analysing the raw performance of a system. Obviously you need to look at the frequency of a particular buss and then you have to consider whether it’s dual or quad pumped a la DDR or the P4 buss. Then you have multiple interfaces to the same buss a la RDRAM and some DDR solutions.

In processors, "bitness" is a very precise thing and its affects on performance varies significantly with code. So really, it's not even a factor.

Quote:
The reason I posted the original question; ‘Is the Hammer architecture too conservative?’ is because although it looks an interesting x86 32 bit processor and also from what I can tell shows great potential as a 64 bit x86-64 SMP solution, as a new platform for the 64 bit desktop I wonder if it’s radical enough?

Radical "enough"? How much is enough? It is a pretty big change to say the least but it is some improvements built upon other existing architecture. That's not radical enough? Furthermore, why is there a need to be "radical"?
August 2, 2002 3:17:52 AM

Imgod2u:

I’m confused, first you say:

<<They both have 64-bit general purpose registers, which is what defines the CPU's ‘bitness’>>

and I agree and follow on by stating the precise addressing parameters of Itanium 2 and then you say:

<<The memory bitwidth is far different than the processor's addressing capability>>

Of course it is, I was using this comparison to show how difficult it is for Joe Public to understand where the REAL gains are to be made as computer architectures evolve. Sorry if I didn’t make my statement Joe Public friendly enough, I thought I was talking to a different audience! Sure I have extended the discussion beyond the realms of just the processor design but this discussion has always been about 64 v 32 bit architectures which since they will be running different OSs has implicitly at least included the whole system architecture.

Compare let’s say 64 bit Windows on a P4 with 64 bit SDRAM against 32 bit Windows on a P4 with dual channel 16 bit RDRAM with all other variables being equal. Which would you rather use? Joe Public could easily be convinced that a slower architecture is better just because certain numbers are higher.

Why do Nvidia bother with a 64 bit DDR versus 128 bit SDR memory interface on identical GPUs? It’s purely down to marketing. Joe Public doesn’t know any better and manufacturers play on that. Isn’t the same thing going to happen when 64 bit OSs get released?

<<In processors, "bitness" is a very precise thing and its affects on performance varies significantly with code. So really, it's not even a factor.>>

I have no absolutely no idea what you are trying to convey here in context to what I said. I feel as if you’re throwing 64 bit code at me and I’m an 8 bit processor!

I have to pick up on this though I was very clearly talking about system performance rather than just CPU architecture in the quote of mine that you gave before making the quote of yours I included above. “Bitness” as you put it is of course always a precise thing wherever you find it. I can’t think of an example where it’s vague??? The only vagueness I can see is in your quote.

My point was that bit size or any other system attribute is not always a clear indication of overall system performance which is why you get manufacturers emphasising whatever they think the public perceives is the performance indicator of the month even though they know it’s bogus.

<<Radical "enough"? How much is enough? It is a pretty big change to say the least but it is some improvements built upon other existing architecture. That's not radical enough? Furthermore, why is there a need to be "radical"?>>

I’d hoped I’d made that point clear. If AMD tie themselves now to an architecture that turns out to be too conservative before anyone gives a hoot about 64 bit then 3 to 5 years down the line Intel can bring a more radical architecture to market which kicks x86-64 into touch.

In horse racing terms, the odds you get on a particular horse for a race that is way in the future are much higher than if you bet on it trackside on the day of the race. This makes sense because there are so many unknown factors between now and then. In other words between now and 64 bit being accepted as a desktop necessity there is a hell of a lot of opportunity for curve balls to be thrown. Personally I have no beef with either company so I look forward to see how this battle pans out with equanimity my only hope being that both companies survive so that we have continued competition.

Where’s the beef dude?

:)  Crow


I thank God everyday for making me crazy, otherwise I’d be insane like most of the rest of ya.
August 2, 2002 4:03:00 AM

i guess you right i must read again about IA-64 i just finish MAJC sometime i just read to much.

The day i meet a goth queen that tell me Intel suck.I turn in a lemming to fill is need in hardware.
September 13, 2002 7:32:05 AM

Nobody argued about the memory addressing capabilities. However, saying a processing isn't "whatever-bit" based on memory addressing capabilities is not accurate. The processor's bit-ness is based on the width of the GPR's period.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
September 14, 2002 12:28:44 AM

Microsoft has announced nothing for the Hammer.

Check MS website, you will find nothing. Windows XP in 32bit does run but no one was allowed to touch the demo I saw nor was the Linux actually doing anything.

Sucks it is pushed back, was looking forward to seeing benchmarks this year even if it was in 32bit mode.

You are limited to what your mind can perceive.
September 14, 2002 8:27:55 PM

you could be right fugger..wouldn't suprise me if microsoft pull the plug.

<font color=purple>Ladies and Gentlemen, its...Hammer Time !</font color=purple>
September 14, 2002 9:28:11 PM

Hmm that might be another reason to why Claw was delayed!
Perhaps they wanted to launch and advertise the product having 64-bit, but MS had not finished the OS yet, so they are waiting to launch in sync?
It would not surprise me, since the capability NEEDS a new OS, and for marketting to work without any suing, you need the OS that runs it.

Add that to the list of possible reasons!

--
Where did your THGC username come from and why did you choose it? <A HREF="http://forumz.tomshardware.com/community/modules.php?na..." target="_new">Tell here!</A>
September 14, 2002 10:17:47 PM

make the compiler is the start for 64 bit software

At the end i have speak with a horny lady
September 14, 2002 11:17:42 PM

"make the compiler is the start for 64 bit software"

That will take time. You can't just snap your fingers and say "you will be a 64 bit compiler". Compilers are written in assembly language. You need to use the instructions that the processor(s) support. It will especially take time for 64 bit software for the IA-64 platform. Why? Because it's not an x86 processor.

Knowledge is the key to understanding
!