Replies to post #47589 on Intel Corp (INTC)

[Elmer Phud]

mas

How about a 4-socket set of predictions?

[wbmw]

Re: Anybody else want to have a go at these predictions ?

Very well.

c) SPECint_rate_base2006 (2 socket) - 80.2

If this were peak, it would seem right. Figure that AMD's stated simulation score for the 2.6GHz part, I would think a scaled down 2.0GHz part should get at least this much. The Intel X5365 gets 116, outclassing this estimate by 45%. Even Intel's 1.86GHz Clovertown gets this much in SPECint_rate.

d) SPECfp_rate_base2006 (2 socket)- 78.8

This sounds too high, IMO. AMD recently gave an estimate in the high 60s for this benchmark, and chances are, they were using peak. If you consider recent compiler enhancements (which are being released all the time), then maybe this is credible. However, I would estimate ~68 based on existing evidence. Even this score would slightly edge out the best submitted Clovertown score, however.

e) TPC-C (2 socket)- 236K

I'll guess 220k. Why not. The newest Clovertown submission is at 251k, almost 15% ahead. This benchmark does not scale well with clock frequency, so it will take more than 2.3GHz for AMD to catch up here.

g) JBB2005 (2 socket) - 143K

Get this from here: #msg-20928323 ? :-)

I come up with the same. Unfortunately, the newest Clovertown submission is at 238k, which is 66% ahead. Is this what you consider a "home run", mas? :-) #msg-22196496

That's at least 2 major benchmarks where Intel may have an IPC advantage over Barcelona. I'm not really interested in the scalar spec scores, or TPC-H, but I'm sure they will be part of the regular benchmark comparisons. It will be interesting to see 4-socket comparisons, too. Personally, I think Intel will look good in everything except for SPECfp_rate.

Sorry, mas, but that's not enough for AMD to claim leadership. <G>