View Full Version : AMD Hammer Benchmarks
Doomtrooper
07-Jun-2002, 01:15
Running on Alpha Hardware and clocked @ 800 Mhz, No operating system enhancements as WinXP couldn't recognize the CPU.
http://www.tecchannel.de/hardware/937/0.html
Lookin Good :P
I remember the early benchmarks Firingsquad ran on the Athlon and it ran dismally in everything. AMD was mighty pee'd off and FS was proved wrong when the Athlon showed how it truly performed.
Now we are getting Clawhammer benches that seem to destroy the Athlon and P4 even though it is running at a mere 800MHz and I am worried that when the actual product ships expectations will be too high and many people will feel disappointed.
Anyways.. time flies.. before you know it, it's XMAS :P
These benchmarks are misleading to say the least.
First of all since the clock speeds are so low memory and CPU speeds are pretty darn close. Caching systems of the CPUs don't offer much advantage and the real performance isn't reflected. Once you scale clocks, you start seeing the caching coming to play and telling the real story. It's not suprising the Hammer is absolutely smashing everything thing, that's due to the low clock and integrated memory controller, meaning very low latency memory sub-systems. Once you raise the clock this advantage will begin to evaporate very quickly and we'll soon find that the Hammer will be more inline to what many have said. 20% better than the K7 at the same clock.
Doomtrooper
07-Jun-2002, 02:57
I don't see anything misleading about them, If you could provide a link that states anywhere that a K8 was only 20% faster than a K7 ??
Looking at the architecture, especially runnning a 64-bit app I think 20% is VERY low and not realistic.
Based on current processors and the current rate of speed increases this is what was leaked from AMD and what performance they are expecting..thats more than 20%.
http://www.xbitlabs.com/news/images/2002-01/slide-1.gif
That's 20% faster in x86 code not x86-64.
You should be able to find that information out for yourself, just do a little surfing. I can't be bothered to fetch a quote for this instance.
Doomtrooper
07-Jun-2002, 03:05
I will call you on that 20% , revisit in 6 months :wink:
Deal.
BTW, about 3 months ago it was stated on Ace's hardware that AMD was getting a 15% boost after recompilation. That's 15% boost over Hammer's 32bit performance.
I don't know about these benchmarks....they seem a bit fishy to me.
Anyways...I thought that "leaked" SPEC score graph was thought to be fake a long time ago.
Ehmm 20% at the same clock would be a huge performance increase. Some people seem to have the mistaken impresion that the Hammer has not been designed to be able to be clocked higher than the K7 on the same process ... they might not succeed in that respect, but it has always been their aim. Higher IPC is just the icing on the cake.
Some people seem to have the mistaken impresion that the Hammer has not been designed to be able to be clocked higher than the K7 on the same process
I agree here, though they could have gone a little further, it seems pretty short.
... they might not succeed in that respect, but it has always been their aim. Higher IPC is just the icing on the cake.
I completely disagree here. IPC is a must for AMD. They know they can't win a clock race, because Intel can outstrip them in process technology. This may not have been the case with the PIII, but remember that core was in it's twilight years. Due to this, AMD needs the IPC advantage. Unfortunately, I don't think the K8 will clock that well and IPC is a problem with x86, since x86 code doesn't lead to much beyond 3 instructions per clock. I suppose this is why the K8 will be around for 2 years at which point the K9 will come into the picture, supposedly that has a longer life span.
It might have a short pipeline compared to the P4, but the impact of that is hard to judge. A couple of the difference between them can already be gained by not needing so much time to transport signals from one end of the die to the other (since its quite simply a lot smaller).
Performance is a must for AMD, which is the product of IPC and clock. They already have a high IPC, and as you say yourself there is only a limited amount of parallelism to work with ... I have little doubt that the hammer will break 2 GHz on .13u.
A couple of the difference between them can already be gained by not needing so much time to transport signals from one end of the die to the other (since its quite simply a lot smaller).
The Hammer is a 9 layer CPU while the P4 is 6 layer. The small die size is misleading. Further more, due to the heavy use of vias, keeping the signals clean will be difficult. The P4 besides it's pipeline has had a great deal of attention paid towards it's signal nets, which are remarkably small.
The die size is a must of AMD, seeing as they have limited fab capacity. Heat extraction might be another issue, SOI will make it worse due to the insulating factor.
Small? Well no smaller than the chip of course. Unless you assume that AMD did not pay much attention to routing I dont see how having more metal layers says much one way or the other. More metal layers could also just be there to give them extra options to reduce latency where they make sense (to better isolate signals for instance) and having more options is a good thing(TM).
I doubt self-heating is much of an issue for them, shrug ... hard to say without inside info, dont they have those Isonics wafers to help them with that?
arjan de lumens
08-Jun-2002, 18:38
The Hammer is a 9 layer CPU while the P4 is 6 layer. The small die size is misleading. Further more, due to the heavy use of vias, keeping the signals clean will be difficult. The P4 besides it's pipeline has had a great deal of attention paid towards it's signal nets, which are remarkably small.
Why would vias be so problematic? Especially given that they in modern dual-damascene copper processes are just vertical wires. Other than that, the additional metal layers may allow routing of more wires/buses per unit area and thus reduce chip area and wire lengths (this helps particularly much in cases of register files and multi-banked SRAMs), or they could be used for power/ground places, which would help signal integrity substantially.
The die size is a must of AMD, seeing as they have limited fab capacity. Heat extraction might be another issue, SOI will make it worse due to the insulating factor.
Older SOI wafers, being made of massive sapphire, would probably have very real thermal condiction issues with modern processor designs. Newer SOI wafers, OTOH, generally use a thin oxide layer on top of bulk silicon to provide the electrical insulation - this layer is typically no more than about 3 microns (http://www.affixtek.com/SOI%20specifications.htm) thick, which is not nearly enough to cause any problems with thermal conduction beyond what bulk silicon already has.
I'm of the understanding that vias introduce noise.
Doomtrooper
09-Jun-2002, 05:52
AMD has a big enough road to travel, if you dispute having a alternative to Intel for your PC Processor and not having a second mortgage to pay for it..I for one appreciate what AMD has done for the 'blue collar' guy and hope it continues :)..and will
AMD has a big enough road to travel, if you dispute having a alternative to Intel for your PC Processor and not having a second mortgage to pay for it..I for one appreciate what AMD has done for the 'blue collar' guy and hope it continues ..and will
Grow the hell up. We aren't talking about who suxorz and who roxorz. We're talking about issues relevent to Hammer. Those like me who have trouble believing in the Hammer hype are not ones who wish to get screwed over by high prices. If they deliver then great, because it'll do me good whether I but one or not.
The discussion right now is what they will deliver and NOT AMD's effect on the market place.
Doomtrooper
09-Jun-2002, 16:26
AMD has a big enough road to travel, if you dispute having a alternative to Intel for your PC Processor and not having a second mortgage to pay for it..I for one appreciate what AMD has done for the 'blue collar' guy and hope it continues ..and will
Grow the hell up. We aren't talking about who suxorz and who roxorz. We're talking about issues relevent to Hammer. Those like me who have trouble believing in the Hammer hype are not ones who wish to get screwed over by high prices. If they deliver then great, because it'll do me good whether I but one or not.
The discussion right now is what they will deliver and NOT AMD's effect on the market place.
I'm grown up just fine kid ..
1)These benchmarks are misleading to say the least.
2)I believe it was a fake.
That Spec sheet came from AMD and could be seen linked right to AMD at one time , the entire presentation was originally covered here..
http://www.zive.cz/H/PCtuning/AR.asp?ARI=103690&CHID=4&EXPS=&EXPA=
We aren't talking about who suxorz and who roxorz
DUH!!! I can't believe you messed it up, Saem. It's "r0x0rz" and "su><0rz"!.....and they let you post news. :wink:
Here is an interesting tidbit. According to X-bitlabs AMD will be ranking there Barton AXP 2800+, because of the added 256KB L2 cache, otherwise the processor will remain the same. All of a sudden the Hammer 3400+ seems like quite the fairy tale.
What's really sad is the fact that this started out as an open effort to be create a benchmark suite that would allow one to judge x86 MPU, which has then turned into a retarded 66MHz = 100 points formula and then now it seems it's worse than Cyrix.
As for whether the new rating is true, X-bit labs from my understanding has been very accurate in it's rumour/inside information department.
I think you're right on the performance ratings, Saem. While there should be a standardized benchmark suite, AMD doesn't seem to be sticking to it if they do list the Barton cores that high.
Personally, I stopped listening to Xbit when they published that 3dfx was coming out with VSA-101, and a when a few other predictions were wrong. They're like the Inquirer, but their English is worse than most of Mike Mageek's crew.
LoL. Well that was a long time ago. They seem to have improved. Oh well. Then again I don't want to proport one rumour mill over the other.
;)
Some speculations about hammer pipeline improvments: http://www.chip-architect.com/news/2002_06_24_Hammers_Two_Extra_PipelineStages.html
Here is an interesting tidbit. According to X-bitlabs AMD will be ranking there Barton AXP 2800+, because of the added 256KB L2 cache, otherwise the processor will remain the same. All of a sudden the Hammer 3400+ seems like quite the fairy tale.
What's really sad is the fact that this started out as an open effort to be create a benchmark suite that would allow one to judge x86 MPU, which has then turned into a retarded 66MHz = 100 points formula and then now it seems it's worse than Cyrix.
As for whether the new rating is true, X-bit labs from my understanding has been very accurate in it's rumour/inside information department.
Frankly, I never took the PR ratings as any sort of serious means of comparing performance or as an indication of benchmarking or whatnot. It seemed clear from the beginning that it was just AMD trying to confuse the average consumer who equates MHz with performance. Which makes sense to me (I don't have a problem with it) considering Intel's clock speed advantage.
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.