Intel Core i7-965: 50% more performance than QX9770

[JumpingJack]

Do you think it is likely that AMD cannot produce bug-free steppings of a die-shrink?

Well, I think AMD can produce a quality part that is more than fit for the market from a shrink or any other design rev for that matter. I have yet seen any CPU released that is bug-free, they all contain errata... typically, the errata are so minor they never even trigger, only occurring under highly specific circumstances. C2D has some > 150 errata listed last time I looked, some closed, some labeled no fix intended, etc. K10 has > 300 errata indexed, some (many) have been fixed are are not even documented in the last errata revision, others also have no fix intended, etc.

If AMD is similar in that regard to Intel we know how K10.5 performs +- a couple percent due to an early leak. Surprisingly the preview shows ~8% better performance, the same IPC gains that behemoth competitor Intel achieved with their die-shrink.
I know this is not a proof, but if AMD cannot produce early bug-free steppings of a die-shrink, why should they be able to incorporate some magic that improves performance by 15% or more?

Both companies, Intel or AMD, when they make 'spokes person' statements related to 'expected performance increase' cloud the statement with ambiguity. For example, the 20% number AMD touts I absolutely believe in this case -- for many reasons, but my question is 20% in what? Sever based apps, perhaps just the SpecFP_Rate -- if this is what they are referring to then this 20% number will not transfer over to DT most likely because a good portion of that 20% comes from BW improvements moving to HT3 in the intersocket communications.

However, if it s 20% on average for SpecFP, SpecINT, etc. (non BW apps), then it bodes well that 15-20% is expected for DT (as the non-rate Spec benches do not press BW to it's hilt).

Frankly, we just need to wait for benchs ... if you want my personal prediction -- Shanghai will see good improvements in Server, making it much more competitive to Harpertown over a wider range of benchmarks, and extending the SpecFP_rate lead. They will also get a good power improvement from the shrink. In desktop, ~ 8-10% clock for clock improvement over Agena on average with a select few apps getting as high as 15-18% -- gaming, around 5-8% if you just measure gaming code performance ... so very 'Penryn over Conroe' like relative increases but just a bit more in this case (Penryn was on average in the 3-7% range over Conroe IPC-wise).

Total guess on my part, so those numbers should not be construed as my trying to spout facts. ... they are totally my opinion.

Jack