• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

ClawHammer: The 64-bit Desktop

A

AGodspeed

Diamond Member
Link

Now there's the smell of change in the air. AMD is about to release the Hammer series of processors. It's the first real step ahead of the game that any of Intel's competitors have ever made. Oh, there have been other innovators; Transmeta with its Crusoe processors; NEC with its V20 and V30 all those years ago. But Intel always set the standard by which those processors were created.

AMD has come up with something new in the form of x86-64. It means, for the first time since the early 80s, Intel has found itself following instead of leading. It has spent so much time concentrating on the Itanium as their future that it could be its undoing.
 
The article has some nice points, but unfortunately there was nothing new.

" You can put money on it that a 64bit version of Doom III will run a hell of a lot faster than the 32bit version."
Just curious, what part of the Doom III code needs a 64 bit processor? I can't think of many reasons for a game to use it - could you help me out?

"The Itanium will die. Not straight away, but close to it."
I thought including the Itanium in this way was odd. Itanium was dead on arrival - straight away dead. The Itanium successor (is McKinley the name) might have a chance, but Itanium was dead long before it started. So why is Itanium mentioned as a comparison?
 


<< "The Itanium will die. Not straight away, but close to it."
I thought including the Itanium in this way was odd. Itanium was dead on arrival - straight away dead. The Itanium successor (is McKinley the name) might have a chance, but Itanium was dead long before it started. So why is Itanium mentioned as a comparison? >>

Lol, the Itanium isn't dead. Sure, the first Itanium was a flop (despite people attempting to convince me that the 1st Itanium was just a "test" CPU, I've read otherwise straight from Intel's PR). However, there's still a big CPU coming, namely McKinley (the successor to the first Itanium, Merced). Surely you've seen the specs on this mother of a CPU? It's true that specs aren't everything, and we'll have to wait for performance numbers. HP and CPQ are backing IA-64, which IMO signals that IA-64 must look mighty appealing if such huge tech companies are backing it (not to mention Dell and IBM will offer systems based on IA-64).

The article mentions IA-64 for a few reasons. One of these reasons are as follows; If Intel decides to stick with the Pentium 4 and not introduce Yamhill (Intel's CPU with x86-64 support) then they will lose the consumer market. But if Intel indeed decides to introduce Yamhill, then HP and CPQ (among other corporate partners) stand to lose ungodly amounts of money because IA-64 program would be greatly undercut by much cheaper Intel Yamhill 64-bit processors.
 


<< Sure, the first Itanium was a flop...However, there's still a big CPU coming, namely McKinley (the successor to the first Itanium, Merced). Surely you've seen the specs on this mother of a CPU? >>


Ok maybe I could have stated my point in a better way. Why compare AMD's future Hammer to Intel's Merced that was released about 1 year ago? Comparing processors released about 18 months apart is laughable. Wouldn't a better article compare AMD's next processor (Sledgehammer) to Intel's next processor (McKinley)? Those specs on that "mother of a CPU" seem impressive - the specs on Hammer seem impressive as well - and I'd like to see a good article comparing them for the high end market.

I honestly think the PR3400+ Hammer will cost and perform quite the same as the P4 at the end of 2002/beginning of 2003. It will be the same toss up that there is currently when comparing the >2GHz P4 to the >PR2000 XP. The PR4000+ Hammer will blow the P4 away if it is released as soon as some rumors have stated. But the P4 will keep ramping up to eventually match it. As of yet, it is far too early to say that "If Intel decides to stick with the Pentium 4 and not introduce Yamhill then they will lose the consumer market".

The interesting part comes in the high end market. How will the Sledgehammer compare to McKinley? I have no clue since they are still both just stats on paper. I have a feeling that McKinley might be rushed to market when Sledge comes around, meaning that its actual performance will be worse than the paper stats. But I'm just guessing at this point.
 
The PR4000+ Hammer will blow the P4 away if it is released as soon as some rumors have stated.

Few questions. How do we know what its going to blow the P4 away at? How much of the PR 4000+ is designed for 64 bit instructions. Are they even including the 64 bit instruction boost to the PR rating? How many programs are going to be out there that use 64 bit instructions and how many of them are we even going to use on a daily basis? How much is clawhammer going to cost? How can we even answer any of these questions now?
 


<< The PR4000+ Hammer will blow the P4 away if it is released as soon as some rumors have stated.

Few questions. How do we know what its going to blow the P4 away at? How much of the PR 4000+ is designed for 64 bit instructions. Are they even including the 64 bit instruction boost to the PR rating? How many programs are going to be out there that use 64 bit instructions and how many of them are we even going to use on a daily basis? How much is clawhammer going to cost? How can we even answer any of these questions now? >>

ClawHammer will sell at a premium at launch, that's a given. But it's replacing the Athlon guys, do you really think it's going to cost much more than Intel's competing processors a few weeks/months after launch? No.

I'm not sure what you mean by "64-bit instructions." As you know, ClawHammer will be compatible with all current and future 32-bit OS's and apps. ClawHammer's vastly improved TLB's and branch prediction, longer integer pipeline (for frequency headroom) and integrated memory controller are only some of the ClawHammer's features that will improve performance over the Athlon XP. If compiler support for x86-64 is good enough, then there's even more performance.
 


<< Few questions. >>


I've had this discussion many times over the last few months with AGodspeed. Here are my best guesses to those answers. Note these are my opinions, and may never be fact.
1) "How do we know what its going to blow the P4 away at?". We can estimate this. Most people would agree that the PR rating matches the equivalent Northwood quite well on average (give or take a few percent). Thus a PR3400+ would match a 3.4 GHz Northwood (give or take a few percent). However the 533 MHz fsb will give the Northwood roughly a 10% boost, so only a 3.1 GHz Northwood will be needed. The PR3400+ Hammer and 3.0 GHz Northwood are both rumored to be released within 1 or 2 months of each other. Thus initally the Hammer and P4 will be nearly tied in performance. The PR4000+ has been rumored to be released far sooner than the roughly equivalent 3.6 GHz P4 release rumors. So if the rumors are true, the PR4000+ will be the first Hammer to be a great success.
2) "How much of the PR 4000+ is designed for 64 bit instructions". I don't think many programs will benefit at all from 64 bit instructions. That is why I asked the Doom III quesiton above. Thus the rating should be roughly the same for 32 bit and 64 bit. Unless I have something wrong... It took years for the switch from 16 bit to 32 bit since few programs benefitted from 32 bit. It could be worse for the 32 bit - 64 bit transition.
3) "Are they even including the 64 bit instruction boost to the PR rating?" Probably it will be based on whichever is faster, but the speed difference will be quite minor for many programs and thus this question is neither here no there.
4) "How many programs are going to be out there that use 64 bit instructions" Lots of scientific programs will be available (I use some daily). But like the 16 bit to 32 bit transition, no normal user has any need for 64 bit - so there will be few if any programs using 64 bit instructions. This is especially true if AMD doesn't gain much marketshare. Why would a game producer make a game incompatable with the roughly 80% Intel computers out there?
5) "How much is clawhammer going to cost?" AMD itself said Hammer will carry a "price premium over the Athlon". What is this potential "premium"? AMD made a good profit in 2000, and lost money in 2001. Lets look at their pricing history (in 1000 lots at date of release and not street prices):
Feb 2000: $849
Mar 2000: $1299
Aug 2000: I lost this data (anyone know the 1100 MHz release price?)
Oct 2000: $612
Mar 2001: $350
June 2001: $253
Oct 2001: $252
Nov 2001: $269
Jan 2002: $339
Mar 2002: $420
I already see AMD ramping prices back up in an attempt to earn a profit again. From the trend since Oct 2001, I'd expect Hammer to be released at $500-$750 easily. Hammer is complex and will probably cost a lot to get production started (so I assume it will be closer to $750). However AMD does quite well in making the chip dimensions small so by the time the PR4000+ comes around, the price could be a lot lower than the initial PR3400+ Hammer's release price. Street prices will be lower.
6) "How can we even answer any of these questions now". Just like I did, using our best estimates. I could be dead wrong, or I could be really accurate. At this point these are my best answers I can give.
 


<< Just curious, what part of the Doom III code needs a 64 bit processor? I can't think of many reasons for a game to use it - could you help me out? >>

the 64 bit instructions might be required to use the vastly improved FPU, which would really help out doom III. with 32 bit instructions you would get an increase vs ceteris paribus, but maybe 64 bit instructions really unlock it.


EDIT: i'd actually expect AMD to be very conservative with its performance rating, using some sort of standard benchmark to determine the ranking. they don't want people to buy one and say, "hey, it doesn't perform like i thought it would, you said 3400 and my friend's 3.2 p4 is faster. screw you amd."

and as for the prices, ASP is a much better indicator than the introductory price of processors, though it is illustrative.
 
christoph83, go here for Paul DeMone's performance estimate on ClawHammer and SlegeHammer. I'll summarize it here:

Clawhammer:

int: 20% MC + 5% FE + 5% x86-64 = 30%
FP: 5% MC + 0% FE + 10% x86-64 = 15%

Sledgehammer:

int: 25% MC + 5% FE + 5% x86-64 = 35%
FP: 40% MC + 0% FE + 10% x86-64 = 50%

Of course this is just a very rough estimate, Claw and Sledge could turn out to perform worse per clock or better per clock than DeMone's estimates indicate. We simply won't know until final samples start shipping to online reviewers (note: "int" stands for integer performance and "FP" stands for floating point performance).
 


<< and as for the prices, ASP is a much better indicator than the introductory price of processors, though it is illustrative. >>



The average selling price is better for AMD to keep track of - AMD needs to raise ASP to get a profit again in 2002 and beyond. However if someone asks what the next processor will cost on release, ASP tells us nothing (since typical ASP numbers include all processors even Durons). There is no possibility to get a single intro price out of an average. A history of intro prices is much better in extrapolating to the next intro price. All extrapolations can lead to great error, but they are the best tool we have (unless you have inside information). Street prices are often 1/2 of AMD's 1000 lot prices after they have a couple of weeks to settle down. Ex: the 2100+ XP sells as low as $243 shipped while AMD charges $420 in lots of 1000 (43% off). So if AMD releases the PR3400+ Hammer at $750, expect street prices in the $425 range.
 


<< The interesting part comes in the high end market. How will the Sledgehammer compare to McKinley? I have no clue since they are still both just stats on paper. I have a feeling that McKinley might be rushed to market when Sledge comes around, meaning that its actual performance will be worse than the paper stats. But I'm just guessing at this point. >>



There's just one little problem with that. McKinley systems will ship in a matter of a few months while SledgeHammer won't arrive until 2003 at the very earliest. But why anyone would compare Sledge with McKinley as competitors on the same market is beyond me. They are aimed at completely different markets. McKinley will go into high-end servers while SledgeHammer will be aimed at the Xeon market.



<< << Just curious, what part of the Doom III code needs a 64 bit processor? I can't think of many reasons for a game to use it - could you help me out? >>

the 64 bit instructions might be required to use the vastly improved FPU, which would really help out doom III. with 32 bit instructions you would get an increase vs ceteris paribus, but maybe 64 bit instructions really unlock it. >>



Doom III might run a little faster (read: ~10%) on x86-64 compared with x86-32 but I can't see why that would motivate a development of a 64-bit version of it. Hammer will support SSE2 in 32-bit mode and will be totally compatible with the P4 in that respect. And it's not like the 32-bit version will be slow on the Hammer. It will be their top of the line CPU at the time after all.
 
Thanks for the information Dullard. That was my only point is that these seem to be only estimates at the moment and we wont know exactly how this guy will perform till later this year. I hope it does live up to the performance rumors we've been seeing. I'm not keeping my expectations high though. Definatly going to be an interesting year in 2003. One thing not to forget when the 4000+ Clawhammer is release Prescott is right around the corner. 1mb of cache and hyperthreading on a 3.5ghz P4 on paper seems like it will perform very well, but thats just as estimate 😀
 
I know u always have someone posting something like this. but WOW. aren't processor speeds getting farther ahead of the industry than they've every been before? i could see where back in the 386/486 days, you could justify upgrading everyone, but upgrading from a 1.1 or in the case of the P4 2.2 to anything is it justifiable for over 80% of the population?

before you start flaming me, hear what i'm saying, i'm gonna upgrade myself as soon as i put together the money and the Claw hammers are available, but for 90% of the population, i think the TOP edge of performance is WAY beyond what they are going to need for at least 2 yrs or so. any thoughts?
 


<< before you start flaming me, hear what i'm saying, i'm gonna upgrade myself as soon as i put together the money and the Claw hammers are available, but for 90% of the population, i think the TOP edge of performance is WAY beyond what they are going to need for at least 2 yrs or so. any thoughts? >>

theres more people doing home video editing than you think. plus gamers. yeah, there are more people doing just internet and email. but then they buy a dv cam. oops, that p2 400 doesn't look so good anymore.
 


<< You can put money on it that a 64bit version of Doom III will run a hell of a lot faster than the 32bit version >>

Absolute rubbish. As this article is written by a "Technical Author and Business Analyst," I doubt he really understands the technology. It sounds like he thinks x86-64 recompilation will benefit from twice the bit-level parallelism (which is contrary to the truth). The fact is that AMD left x87 FP unchanged with x86-64 (it still uses the 8-register FP stack), so a recompiled DoomIII would receive little benefit from the addition 8 GPRs. x86-64 does add 8 more SSE registers, but as Carmack has said, 3D games do not auto-parallelize easily for SIMD, and thus receive little benefit. The most benefit of SIMD optimization is realizable in the video drivers, where the CPU spends much of its time while playing a 3D game. So only if you consider the few percent from microarchitectural optimizations available in x86-64 recompilation "a hell of a lot faster" would the author be correct. And I see little impetus for id to release a version of Doom III that will run on 10-15% of all PCs sold in 2003.



<< The Itanium will die. Not straight away, but close to it >>

As much as the Inquirer crowd would probably like to think so, the Itanium family is not going away. A lot of people don't seem to realize that single processor performance and multiprocessor scalability do not make an enterprise platform. Services, software, support, reliability (different than desktop reliability), reputation, and partners make the platform. And this is a market in which Intel has just started building a reputation for the last few years...AMD has none. Intel has spent years with its partners (IBM, Compaq, HP, SGI, Unisys) developing Itanium platforms; McKinley (like the EV7) first taped out in early 2001, and since then has undergone intense validation with its partners. Until Sledgehammer receives large design wins from the major first-tier enterprise systems manufacturers (such as HP's 64-way McKinley-based Superdome series), it is not going to compete against Itanium. The 2- to 8-way systems that, by AMD's own admission, Sledgehammer will be used in will put it in the low-end server market against the likes of Xeon. The mid- to high-end market is one in which x86 has zero penetration and reputation (except perhaps for clusters, but that's a different market)...it's going to take time for AMD to penetrate this market segment if they want to, and it won't happen with Sledgehammer.

Don't misinterpret my post; I have a lot of respect for AMD, and I believe that Hammer will be a success in the PC desktop and low-end server markets. But this talk of Hammer killing Itanium is ludicrous.
 


<< theres more people doing home video editing than you think. plus gamers. yeah, there are more people doing just internet and email. but then they buy a dv cam. oops, that p2 400 doesn't look so good anymore.
>>



ok, i'll grant you the P2 400 is a bit slow. but a ClawHammer by the end of next year? you really think software will have caught up to justify the purchase of the Clawhammer for 80 to 90% of people out there? even if they are doing video editing?
 


<< ok, i'll grant you the P2 400 is a bit slow. but a ClawHammer by the end of next year? you really think software will have caught up to justify the purchase of the Clawhammer for 80 to 90% of people out there? even if they are doing video editing? >>



Who said that Clawhammer would be marketed to 80 to 90% of the people? AMD will still be selling the Athlon XP
lines based on T-bred and Barton designs at that stage. I expect that Clawhammer will initially be targeted at the
workstation / power-user segment (typical anandtech member?). There won't be major impetus to adopt Hammer
for most situations until the industry starts to show greater support (when the Motherboard and Box makers start to
compete in providing Clawhammer based equipment).

I think that saying the average user does not need the kind of performance that newer processors offer is overlooking
what the average user does with their system. No two users are alike in what they start using a computer for, and as
they learn more what they can do, they start stressing the system in different ways. The better performing a general
purpose CPU is, the better it can provide a baseline for performance in any situation that the user may encounter.
Once the user gets to that baseline, they typically start adding "features" on to the system that slow it down or
use up resources, which makes it more important that a fast processor can handle all the background tasks and still
keep the rest of the system responsive to the user's requests.

As long as we keep going along with technology that is just "good enough", we will never open up the possibility of what
can happen when users start to think they have real power under the hood on their systems.





 
Even though I think this discussion is interesting, I am so sick and tired of hearing processors run at these extreme speeds. We all know that the bottleneck of every computer is still the hard drive. Most people can do all they need to do with a 1GHz processor, and here Intel and AMD are beating each other's brains out to have to fastest processor? WHAT FOR?! It doesn't benefit the consumers in any way, just makes us spend money on things we really don't need. What we need is better technology from our storage devices. So until we get better performance from our hard drives, all this talk mean squat.
 


<< Even though I think this discussion is interesting, I am so sick and tired of hearing processors run at these extreme speeds. We all know that the bottleneck of every computer is still the hard drive. Most people can do all they need to do with a 1GHz processor, and here Intel and AMD are beating each other's brains out to have to fastest processor? WHAT FOR?! It doesn't benefit the consumers in any way, just makes us spend money on things we really don't need. What we need is better technology from our storage devices. So until we get better performance from our hard drives, all this talk mean squat. >>



You have a decent point, but it's not entrely true. There's been a lot of talk about Doom III for example. I'm pretty sure that most of the data required by Doom III will fit in memory, so the only thing holding back the performace of the game will be raw number crunching. For things like running the OS and browsing the web, to video and photo editing, faster I/O systems are needed, but there is plenty of benefit to a higher clock. Plus you can crunch more THINK units with a faster chip😀

Kramer
 
Sorry, but until I can encode a full-length MPEG2 feature in realtime (VBR) 5-pass and burn the images in realtime, I'll never be satisfied and always craving more 😀

And a few extra FPS can't hurt either 😀 🙂
 
what mhz are this supposed 3400+ and 4000+ going to be running at???

I agree with above...way to early to speculate on this....p4 sse2 encoding is still not up to par and widely blown out of context by intel marketing as this great boost....p4 has been claiming it for a year now and it hasn't really materialized to make up for the neutered fpu...when it has been optimized it has been reviewed in websites as maybe making a 10-15 percent increase.


I agree with pabster as well
🙂
 
It's true that for today's and even the next year's apps, 95% of the population doesn't need a 2.4Ghz CPU.
But I think that is entirely missing the point.
Ever hear the phrase about Intel giving and Microsoft taking away?
How about "640K ought to be enough for anyone" At the time Bill said it, it was true.

The fact is, we have no idea what possibilities exist for the future of computing. I guarantee that in 5 years, people will be laughing about the old days when they only had a 3Ghz processor.

As processors and other hardware continue to increase in performance and speed, software developers will try to keep up and will find ways to take advantage of all that power. That means better and better software.

I'm not a huge microsoft fan, but look at the difference between Windows XP and Windows 3.1
 
people also expect their computers to last for several years. the average user isn't an anandtech guy that thinks changing out the mobo, processor, ram, and gfx card each year is a normal thing. i talk to people all the time that say, "well if i'm just gonna get a new computer in a year then why bother upgrading the memory on this machine?"
 
64bit Doom3 would be faster than 32bit one. Why? Simple: compiling it for x86-64 would expose those extra registers in Hammer, which would increase the performance of the CPU.
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top