AMD Clawhammer Benchmarks

1) They are not legit "benchmarks". First silicon on Clawhammer is confirmed (as of AMD earnings teleconference two days ago) *not* to have come back yet. It probably hasn't even taped out yet. At best these numbers are estimates.

2) But they're not legit estimates. The 1350 SPECint number comes from an unofficial remark Fred Weber (big guy at AMD) made at MPF (big microprocessor conference) along the lines of "Hammer will be around twice as fast its competition from today at SPECint". As the highest SPECint score when he said that was around 675, people started throwing around Hammer performance of 1350 or 1400 as confirmed fact.

Well first of all, not even AMD knows what score Hammer will get on SPECint yet. There's no Hammer silicon to test on, and simulations are way too slow to run SPECint, plus they're not all that accurate. Plus x86-64 optimizing compilers weren't available when he said that. GCC just released its x86-64 port this week, but it probably still has some optimization left. Plus it's unknown whether the benefit of using 64 bits will outweight the performance loss of using GCC instead of the Intel compiler for SPEC. Basically, no one has any idea yet.

Second "today's state of the art" doesn't necessarily mean the very fastest SPECint score then available. Maybe he meant twice as fast as the best available AMD chips at the time, which would mean a score of maybe 1150, not 1350. Who knows? Maybe he felt that the competition for Hammer was the Intel Xeon and was comparing to the best released Xeon scores (which would be even lower at that time IIRC). Or maybe he just made it up because he knew it was vague enough to mean nothing.

3) The "twice as fast" remark which gave rise to these numbers was probably referring to SledgeHammer, the 4-way+ workstation/server varient, rather than ClawHammer, the 1-to-2-way desktop version.

4) On the other hand, recent AMD roadmaps *have* revealed that a "model 3400" ClawHammer is due to be released by the end of this year, so the labeling was at least right. However, it's important to note that model numbers don't necessarily mean anything, and that unless AMD changes the way they assign numbers, a "model x" AMD chip may not be faster than an x MHz P4 once the P4 gets 533MHz FSB this spring.

5) This chart was completely made up by these Czech people to show how the 1350 rumor fits in with projected SPECint numbers for other architectures. Personally, I think that SPECint of 1000 for a 3 GHz P4 is about right (this will be in Q3). EV7 will launch at 1.2 GHz, not 1, and SPECint will be in the 900s if not break 1000.

As for ClawHammer, it will probably launch early next year. Assuming they do decide to call it "model 3400" it will probably be almost as fast as a 3.4 GHz P4 (which will launch around the same time or slightly earlier) as normal PC benchmarks, but not quite at SPEC. I'd say 3.4 GHz P4 gets ~1150, and model 3400 ClawHammer gets ~1050. When SledgeHammer launches around a quarter later, it may break 1200 but 1350 is quite a long shot.

OK, I found the English Version.

http://www.xbitlabs.com/news/story.html?id=1011297004

There you go droogies!

whatever the case, AMD will remain top dawg on the price/performance ratio.

this looks to me like a "hypethetically, it's this fast" type chart

AMD SMASH!!!!!!!!

AMD 3400? they just announced the 2500.

i think this page is bogus.

Man that page is so slow... I clicked it like 5 min ago and it's still loading.
Can you just give us the gist?

wlker wlkerj w