AMD submits SPECpower_ssj for Intel
Friday Jan 09, 2009
George Ou wrote an article called "AMD submits suboptimal SPECpower benchmarks for Intel", pointing out that AMD submitted some results for Intel-based systems.
Of course, call foul on AMD for using eight 2GB dimms instead of four 4GB DIMMS. AMD did however use Low-Power FB-DIMMs on their Intel submission ("Memory Details: DDR2-667 CL5 LP FB-DIMM;"). But it begs the question why aren't Intel-based vendors using the same standard DIMMs as they use on other benchmarks that don't require power reporting?
OK let's dig deeper, maybe the AMD published Intel result do reveal something. Maybe memory configuration/size is critical to a servers watts? Let's look for some tidbits...
- Why are all SPECpower_ssj on 8GB to 16GB, while other benchmarks using the same CPU family are on 32GB to 64GB?
- Why did the reporters miss that Intel-based systems use a modified (hacked?) BIOS for SPECpower_ssj? Vendors using different BIOS than what is default for customers???
- SPEC data does show servers at low utilisation (anywhere under 50%) have much different power-performance that even 50% utilisation. There has got to be a story there?
- ... other "interesting" configuration choices for SPECpower_ssj
Sun publishes measured watts, as I pointed out in yesterday's posting "Sun's 2008 summary of measured watts & watt/performance": http://blogs.sun.com/bmseer/entry/sun_s_2008_summary_of
Disclosure Statement
SPEC and the SPECpower benchmark name are registered trademarks of the Standard Performance Evaluation Corporation.
|
|
|
|
