David Kanter (dkanter@realworldtech.com) on 1/3/11 wrote:
---------------------------
>So apparently the embargo for Sandy Bridge reviews has lifted, ahead of CES. While
>we are not going to join the cohort of folks who are out the door immediately, we shall eventually have a review.
>
>In the mean time, several of our friends and colleagues have put together excellent
>reviews. I'd like to point out two of my favorites, from Scott Wasson and Michael Schuette:
>
>http://techreport.com/articles.x/20188
>http://www.lostcircuits.com/mambo//index.php?option=com_content&task=view&id=98&Itemid=1
>
>The bottom line seems to be that all the microarchitectural innovations have yielded
>performance and efficiency results. It will be a real uphill climb for AMD...
>
>David

Yes, clock-for-clock Sandy Bridge seem faster than Clarksfield, but I'd attribute the difference to Clarksfield's off-die memory controller.

Comparison vs Gulftown is more apple-2-apple and here Sandy Bridge loses more than it wins. Of course, Gulftown has more cores and it's not 100% obvious in which benchmarks it matters and in which it doesn't.

Overall, the impression is that core-for-core and GHz-for-GHz Sandy Bridge is few percents faster than Gulftown for FP/SIMD. Of course, when AVX-enable applications come out it's advantage in FP/SIMD is going to grow up.
As for integer work, there is simply not enough data. Diep scores (+11%) are encouraging so Vincent is probably very happy with SB, but I personally would want to see more tests than just Diep.