InvestorsHub Logo

Golfbum

06/03/15 8:26 AM

#141102 RE: mas #141100

Not exactly an impartial test. No mention of how benches are run, what compilers and switches used.

gb

wbmw

06/03/15 10:14 AM

#141103 RE: mas #141100

I see the measurements compare socket vs. socket. Is that the right comparison point? Are Power8 sockets as easy to scale and cost the same as Xeon sockets?

mas

06/05/15 10:01 AM

#141123 RE: mas #141100

I believe Intel could substantially close these throughput deficits to Power8 if it chose to double the threads on Xeon to 4. Power8 has 8 threads to match its eight-issue core, virtually nothing is left on the table in terms of unused pipelines whereas I believe Intel is leaving 20-30% throughput performance behind by not having quad-threads to match its quad-issue Xeon. Yes Xeon is cheaper, cooler and has generally higher single-thread performance but still there is no reason to give your competition any unnecessary edge. The Broadcom Vulcan ARM server chip will have 4 threads to match its quad-issue and the MIPS chip on which it is cloned from, XLPII, can beat the equivalent network Xeon in some network throughput benchmarks. This issue/thread mismatch is an oversight that Intel really should get round to fixing one day.