AMD ClawHammer Benchmarks!?!?!?!

[MadRat]

I hope AMD debuts in the mid-$1000 range to up their stock value. Lets face it, if ClawHammer works to the bill then it outclasses every P4-family derivative to date. AMD does not need to sell too low just to sell out of their production until they hit stride with the SOI process.

I am personally hoping that they make Thoroughbred into a $200-400 processor, up from the Palomino's $100-300 price range. There is no reason to believe a die shrink, move to a true 2GHz top end, and an L2 cache jump to 512k couldn't fetch $400.

 

[BrunoPuntzJones]

<< I'm told these latest figures are compiled from .13 micron cores with SOI, running at a clock of 2.0GHz. Indeed, the future looks very, very bright. >>

wow, is that for the 3400+???

 

[ST4RCUTTER]

I honestly don't think the initial Hammer release will sell many processors. I could be very wrong. But how many home consumers have a 64 bit version of Windows? (The unix community isn't large enough to make a significant effect on sales). How much does an upgrade to the 64 bit Windows version cost? How many home users have programs that will take advantage of the 64 bit processing? I don't know, but the fewer people that are estimated to buy it, the higher I think AMD's initial price will be.


Who cares about 64-bit Windows? The Hammer is 100% X86 compatible (32-bit). Not only is it compatible, but it will likely be the fastest X86 based chip ever constructed. This should make things interesting towards the end of 2002. The Northwood will have a tough time competing against Hammer. The Itanium doesn't support 32-bit apps in hardware. AMD said in their conference call that Clawhammer (desktop chip) will be the first to ship with the Sledgehammer to follow. I wouldn't be surprised if Clawhammer weighed in around $350-$450 either.

 

[BlvdKing]

I wonder what type of motherboard these benchmarks were run on since the hammer will require a new chipset. I am not sure but I get the impression the benchmarks are simulated.

 

[WilsonTung]

I remain highly skeptical of this chart posted on Xbitlabs. This shows the Clawhammer beating out IBM's POWER4 and the nearly defunct DEC/Compaq EV7.

x86 chips are notoriously complex logicwise. Enthusiasts are billing it as a tiny, all-in-wonder with integrated this and that, SSE2 support, integrated memory controller, etc.

As I have no access to AMD info, I can't say whether this is for real or not. Perhaps they picked up a genius design team somewhere that can beat the eccentricities of the x86 ISA.

But these expectations defy common sense. I would be willing to bet though that this chip will run hot-as-hell.

 

[Pabster]

Sure easy to spot the zealots, eh?

Hammer is an architectual wonder. There's no denying that. It will be, unquestionably, the fastest x86 core on the market. IA64 is heading nowhere right now, and unless something changes (in a hurry) ... The beauty of x86-64 is the backwards compatibility. Unlike IA64, where 32-bit code has to be "parsed", turning a 700MHz in to 66MHz ... Hammer won't reach the absolute pinnacle of performance with 64-bit code, but the backwards compatibility is a major plus. It's gonna be awful hard to market 64-bit stuff to consumers in the first place; telling them all their "old" software is useless just isn't gonna cut it. And to those who honestly believe Intel will stand by and watch AMD dominate ... You can bet they have a design in the works which is 100% x86-64 instruction set compatible. Their sales thus far of Itanium have been a disaster, and with the weak economy as well...

This is gonna be a great year, for sure!

 

[Pabster]

WilsonTung wrote:

"But these expectations defy common sense. I would be willing to bet though that this chip will run hot-as-hell."

Why? You've got to put things in perspective. Hammer is nothing like the current Athlon we love so well. It's a radically different design. If you'll recall, it was rumored about 6 months or so ago that a 2GHz Hammer core would offer the same performance as a 3GHz Thunderbird core. And that figure was on a .18u process. Factoring in the die shrink (0.13u) as well as SOI, I think it is very practical.

As for "running hot-as-hell", I wouldn't place any bets.

 

[Bluga]

<< As for "running hot-as-hell", I wouldn't place any bets. >>

 

[AGodspeed]

As for "running hot-as-hell", I wouldn't place any bets

If the Northbridge is indeed integrated into ClawHammer, I would bet it'll run pretty damn hot. But who the hell cares, AMD will have the proper cooling solutions by then, there's no way AMD would leave a potential heat beast like ClawHammer as "bare" as the current Athlons, not with an integrated NB and all.

 

[WilsonTung]

<< Why? You've got to put things in perspective. Hammer is nothing like the current Athlon we love so well. It's a radically different design. >>


From RealworldTech (see link at bottom of post)


<< Although the 64-bit instruction set extension makes for attention grabbing headlines in the technical trade press, the major performance enhancements in the Hammer series are much more prosaic from a processor architecture point of view >>


The basic 3 Integer and 3 x86 units are still there, although the Hammer whitepapers apparently show that AMD has optimized the chip to make much better use of them. There is nothing 'radically different' here.

Integrating a memory controler into the chip comes at a cost in space. Adding registers takes up space. Supporting SSE/SSE2 takes up space. These additional components generate heat - you can't get around that. SOI may alleviate that. If the L2 cache size goes up that adds space.

Everything has costs and tradeoffs. No chip (or any other product, for that matter) can escape this. Motorolla decided to increase the pipeline length of the G4, they lost IPC. AMD used 3 FPU pipelines in Athlon, thus it ran hotter than Pentium 3. Transmetta's code morphing simplfies the archeticture of the chip and makes it smaller and cooler, trading performance in the process.

The only 'perspective' I see here is hype that rivals that of the Merced. I said the same things (big, hot, expenisve) about 'Merced' years ago and nobody beleived me. Perhaps I'm wrong this time, but I am inclined to be suspicious until credible benchmarks from several sites are available.

http://www.realworldtech.com/page.cfm?section=columns&AID=RWT010202033918

From Paul DeMone:


<< , I expect the AMD device will easily beat the much slower clocked IA64 server chip in SPECint2K and many other integer benchmarks, as well as challenge much faster clocked Pentium 4 devices in both integer and floating point performance >>




<< . Keep in mind that no Hammer device has even taped out as of the end of 3Q01 let alone been fabricated, debugged, verified, and benchmarked at the target clock frequency. >>




<< As been shown time and time again, it is best not to count performance chickens too closely before the silicon eggs hatch. >>

 

[Degenerate]

<< There's no denying that. It will be, unquestionably, the fastest x86 core on the market. >>


Why? where is the proof? Why make assumptions when there are no facts? I can also say the successor after the Pentium 4 will be "the fastest.. core on the market". Why bother? you seem very informed on a CPU that is still far from reach for any of us. However, i am excited as you are about a new CPU.

 

[WilsonTung]

<< where is the proof? >>



There is no proof.

<sarcasm>And a 2 GHz PPC G5 is going to make AMD sorry. Altivec 2 will destroy the hammer and Steve Jobs will laugh at all the slow PeeCee owners<sarcasm>

 

[Degenerate]

Showing one becnhmark, is about as credible as Intel showing that the fastest CPU is P4 by showing Q3 and Mac comparing, clock for clock, Photoshop benchmarks. And "Vapour" it is. So until it condenses, everyone breath.

 

[Rand]

<< The benchmarks are believeable to me. AMD has never thrown out any bullsh*t statements it cant back up. >>



Never?
It wasnt that long ago that AMD was hyping up their new K6-2 300MHz that would outperform a 450MHz PII in most gaming applications. We all know how that turned out.



<< Hammer engineering samples have been out for months now. I'm told these latest figures are compiled from .13 micron cores with SOI, running at a clock of 2.0GHz. >>



Wherein did you hear this when AMD has been claiming for months the exact opposite?
Jerry Sanders, Fred Weber, Hector Ruiz, and Drew Prairie have repeatedly claimed no Hammer has even taped out as of the end of 2001 let alone been fabricated, debugged, verified, and benchmarked.

AMD has stated architectural software simulations to have put the ClawHammer at a release of 3400+ to attain a SpecINT 2000 score of 1300-1400... which is dead on what these supposed "benchmarks" put it at, as well as the same supposed PR rating.
If these benchmarks are from a true Hammer then AMD's software simulations are incredibly accurate for a new core, espeially one that will be the first X86 core in existence to integrate a memory controller on-die. Also AMD must be clearly ahead of schedule despite all the rumours of late that the ClawHammer may slip to Q1 2003.

The statement of 2GHz is viable as that correlates with Ruiz's year ago statements that release ClawHammer's would run at 2GHz 'true' MHz... though latest rumoprs do put this in doubt.



<< Why? You've got to put things in perspective. Hammer is nothing like the current Athlon we love so well. It's a radically different design. >>



Nothing like it?
How so? All of AMD's tech documents thus far have indicated the Hammer is extremely similar to the K7.
The entire back end of the processor is almost an exact duplicate of the classic K7 core.
The entire pipeline process is extremely similar to it, a few new names but the functional uses are identical. Besides the two extra stages devoted to decoding the extra complexity X86-64 adds the basic execution of the Hammer remains unchanged from the K7, and the entire back end is totally the same.
The only real differences are in the front end of the processor.
What makes the Hammer special is it's vastly improved BPU, the three integrated bi-directional HyperTransport links and a memory controller. It looks top be quite easily the most scaleable X86 core ever derived, and seems it could quite easily scale to 8processors in a system without any chipset level innovation... a glue-less 8 processor Hammer system is extremely viable. Not to mention is seems strikingly easy to implement the Hammer into a dual-core processor. The real improvements to the Hammer are quite outside of it's actual architectural makeup which remains very similar to today's K7 based Palomino's....memory hierarchy and multiprocessing architecture are the big improvements, and in those areas the Hammer is unlike anything ever implemented outside of the RISC 64bit realm.

I'd really suggest anyone that hasnt already to take a look at Fred Weber's Hammer Presentation at the 2001 Microprocessor Forum. Hans de Vries also had a very good qick summary of the architecture.

BTW... I think the claims of $1000 for the ClawHammer are extremely unlikely. Remember this IS the ClawHammer we are talking about. It is to replace the K7 as AMD's high performance desktop core.
It's the SledgeHammer that will make headway into the corporate markets and high level server operations--- that's where you will see the over $1000 pricing.

 

[WilsonTung]

<< Nothing like it? >>


That's what I said - people are posting all sorts of nonsense on this board. I'd bet that most of the people who root for one side or the other don't know jack about digital design or CPUs.

 

[Sid03]

<< I'd bet that most of the people who root for one side or the other don't know jack about digital design or CPUs. >>

i couldn't agree more! pm, sohcan, burntkooshie, wingnut pez... are the more knowledgable people about cpu's on this forum. and you'll never see those guys with the stupid zealotry like many people here.

is it just coincidence? i don't think so.

 

[Degenerate]

<< If the Northbridge is indeed integrated into ClawHammer, >>


BTW, what advantages are there to this approach? faster?

 

[Random Variable]

I would much rather see the ClawHammer's floating-point performance.

 

[MadRat]

<<<< If the Northbridge is indeed integrated into ClawHammer, >>

BTW, what advantages are there to this approach? faster?>>

It destroys the dependency on the FSB for external memory bandwidth to the core.

 

[dullard]

<< Who cares about 64-bit Windows? The Hammer is 100% X86 compatible (32-bit). Not only is it compatible, but it will likely be the fastest X86 based chip ever constructed. This should make things interesting towards the end of 2002. The Northwood will have a tough time competing against Hammer. The Itanium doesn't support 32-bit apps in hardware. AMD said in their conference call that Clawhammer (desktop chip) will be the first to ship with the Sledgehammer to follow. I wouldn't be surprised if Clawhammer weighed in around $350-$450 either. >>



From what I've read AMD claims that the Hammer will run 32-bit applications no worse than current processors. No where do they claim state that it will run any faster. Thus I'd expect that it will run at or near those benchmarks for 64-bit applications and half that speed for 32-bit applications (Note: the current Athlon is about exactly half the speed on that benchmark). If the hammer is several hundred dollars more expensive than the Athlons that run 32-bit applications at the same speed, then the Hammer won't be an initial success. Only when 64 bit Windows and applications are around will we see the true performance.

 

[dullard]

<< I would much rather see the ClawHammer's floating-point performance. >>



Me too, since the programs that greatly benefit from 64-bit processors are almost completely focused on floating point calculations. But we have to start somewhere with the benchmarks...

 

[AGodspeed]

<<

<< If the Northbridge is indeed integrated into ClawHammer, >>


BTW, what advantages are there to this approach? faster? >>



Among the most important purposes of integrating the NB into the CPU is the role the memory controller plays. By having an integrated memory controller (via the NB) much higher FSB frequencies are possible (Anand mentioned 800MHz FSB before).

 

[Rand]

Dullard, I believe you have a few things confused regarding 64bit code.



<<
From what I've read AMD claims that the Hammer will run 32-bit applications no worse than current processors. No where do they claim state that it will run any faster. >>



The Hammer whitepaper claims improved IPC over the K7 core, and it is to debute at a higher clockspeed as well. Fred Weber was quoted as stating the Hammer core could offer performance as much as twice that of current processors. AMD has stated many times that the ClawHammer and SledgeHammer both will feature improved performance over the K7 core.
The improved BPU alone should improve per clock performance considerably as that is one of the weakest aspects of the current Athlon core.



<< Thus I'd expect that it will run at or near those benchmarks for 64-bit applications and half that speed for 32-bit applications (Note: the current Athlon is about exactly half the speed on that benchmark). >>



Half the speed in 32bit?
Having a 64-bit CPU vs. a 32-bit CPU doesn't imply that the 64-bit CPU is twice as fast. SISD arithmetic (single instruction- single data) operates on two operands and yields one result, if you do 1 + 1, it doesn't matter if you express the numbers as 32-bits or 64, the instruction will still do only one add. Floating-point arithmetic does benefit from wider adders in terms of accuracy, but the X86 FPU already have a high degree of precision (x87 uses 80-bit internal precision). If anything it should perform faster in 32bit code, as it shall be less complex and bandwidth intensive. The advantages of 64bit code are more in the areas of flat addressing of more memory then speed. If all else is equal 64bit code should realistically be about 5% slower.
In this case X86-64 does have a slight advantage of straight X86-32 though in that X86-64 adds 8 new registers (XMM8-XMM15) only available in 64 bit mode.



<< If the hammer is several hundred dollars more expensive than the Athlons that run 32-bit applications at the same speed, then the Hammer won't be an initial success. Only when 64 bit Windows and applications are around will we see the true performance >>



As stated above 64bit code will in absolutely no way inherently imprve performance. Most modern desktop appliocations don't even need 32bit. 64bit code is still a long way from being needed in a home environment. We're talking high end multi 4+ way SMP servers that need 64bit code.
Also the ClawHammer is set to replace the Athlon as AMD's high end desktop processor to compete against the Pentium 4, so pricing should be similar. It's the later SledgeHammer that will be priced much higher and is going for the high end server marker- that's wehere the 64bit capability will make a difference, and that's where the high prices will come in.