AMD's Opteron marketing has been very good, mostly because they've taken the "high road" emphasizing performance, not just trying to compete on price. It didn't hurt that AMD held both integer and floating point performance crowns, and at the lowest power, for two years until Woodcrest arrived.
Most people don't realize that Opteron is STILL faster per-core than anything based on Woodcrest in floating point at even dual-socket, 4-way level. Take a look at SPECfp_rate2000 4-CPU results. 2 DC Opterons beat 2 DC Xeons by 43% (121 to 84.7). A pair of Opteron dual-cores beats the quad core Xeon by 84%!
Now get this: a pair of dual core Opterons (4-CPU SPEC_fp_rate) outperforms a pair of quad core Xeons in 8-way SPEC_fp_rate by 121 to 104, or 16%. Any scientist who buys Xeons is an imbecile. I would not be surprised if a SINGLE Barcelona outperformed a pair of quad Xeons in a few months.
OK, AMD is behind 33% in 2-socket SPEC_int_rate. They are getting a (4-way) score of 123 using DC Xeons and an 8-way score of 200 using QC Xeons. AMD can only get 90.4 out of a pair of DC Opterons.
But I just noticed something. SPEC_CPU2006 is replacing SPEC_CPU_2000, and it favors Intel a lot less in SPECint, probably because working set sizes have been expanded beyond Intel's cache capacity. But SPECfp2006 is a lot closer, but AMD still has an advantage, which will no doubt greatly increase with Barcelona.
In SPEC_int_2006_rate (4-way)
Woodcrest 2-socket dual core gets 55.2 (3.0 GHz)
Opteron 2-socket dual core gets 51.7 (2.8 GHz)
A pair of QC Clovertown Xeons (8-way) gets only 67.5 -- QC is getting hardly any benefit from the extra CPU'S!
I think its a cinch that a pair of Barcelonae will beat that score easily.
But there is something funny going on in that result --
spec.org
According to IBM, they are only using 2 GHz E5335's in that review, not the 2.66 GHz E5355's. What's up with that? Do the 5355's self destruct when running SPEC_int_rate_2006? Kind of strange because there's a gazillion SPEC2000 results using the 2.66 GHz E5355, but no SPEC_CPU2006 results using it.
Getting back to the original point of my post, I agree that desktop and notebook marketing has been poor.
Petz |
