Joe Chang (jchang6@Xyahoo.com) on 11/9/08 wrote:
---------------------------
>Inquirer points out the SPEC Power results posted by AMD for Shanghai and Penryn (Xeon L5420) both at 2.5GHz.
>AMD results for Shanghai is 731 versus 561 for Xeon.
>SPEC Power results posted by others for the L5420 were higher, the SuperMicro results
>being 990, but not as close a system match as the 2 AMD results (Windows instead of Linux, and BEA vs Oracle).

Xeon L5420 (2.5 GHz) - 561 ssj_ops/watt
http://www.spec.org/power_ssj2008/results/res2008q4/power_ssj2008-20081017-00089.html

Opteron 2380 (2.5 GHz) - 731 ssj_ops/watt
http://www.spec.org/power_ssj2008/results/res2008q4/power_ssj2008-20081017-00090.html

AMD's has tried to create an apples-vs-apples comparison here. Same OS (SLES10SP2), same JVM version. Very few configuration differences and they did set HW prefetching to off for the Intel machine this time.

Note that the 8x2 GB configuration is the optimal choice for AMD hardware since the 4 DIMMs-per-channel-config will enable bank interleave while allowing max speed on the HT bus.

The best L5420 score with JRockit (from NEC) is *much* higher: 1010 ssj_ops/watt
http://www.spec.org/power_ssj2008/results/res2008q3/power_ssj2008-20080714-00064.html

This is on Windows (will yield a slightly higher score), with different BIOS settings and with 2x4 GB DIMMs instead of 8x2.

Interesting enough, the NEC submission has a higher peak score (288502 vs 267804) despite the fact that it is uses a smaller Java heap and the slower (?) 4 GB DIMMs. This is counterintiutive to me and makes me suspect that the AMD Xeon submission is on a machine with a less optimized BIOS.

The best Shanghai score is with a faster chip and on Windows:

Opteron 2384 (2.7 GHz) - 860 ssj_ops/Watt
http://www.spec.org/power_ssj2008/results/res2008q4/power_ssj2008-20081017-00092.html

-- Henrik