Hi James, The big question i would like to ask intel architects is whether the Xeon Phi architecture has a future, so what comes AFTER this Xeon Phi?
The next chip that's at least 2x faster than Xeon Phi, is it ever going to be there? if so, at what price? From what i understand it has cache coherency - otherwise it could not run x86 codes at a slow speed. The gpgpu hardware doesn't have cache coherency. This is why we have so many cores in such a short period of time at the gpu's. From answers from engineers i understand the reason why most normal cpu's do not have more cores is because of the cache coherency. More cores are a cache coherency nightmare. Cache coherency is very costly to maintain in a cpu. So the question i want to ask is whether Xeon Phi scales for future generations of releases of it. Are they going to modify the AVX2 code to AVX3, so vectors from 1024 bits in the future, in order to get some extra performance? I assume more cores is going to be near impossible to keep coherent. 60 is already a lot. How is intel going to win factors in the future at this Xeon Phi design? Are they going to drop the cache coherency simply? Resulting of course in a chip that no longer can run x86 codes at all - which seems to be the major salespoint right now. From my viewpoint seen the next version of the Xeon Phi, let's call it the 2 Tflop chip, X years from now, it's going to require modifying all codes again isn't it? They have to drop something from the chip. Either the 512 bits vectors and move those up to 1024 bits, which is the usual solution intel goes for, or drop the cache coherency and move from 60 to 120 cores (working cores), at which point your scripts no longer will execute as the x86 compatibility is gone then. Can you ask this question to one of the Xeon Phi architects next time you encounter them? On Dec 14, 2012, at 3:55 PM, James Cuff wrote: > Hi team, > > I've put a few notes up on our prerelease Phi engineering sample on > the blog: > > http://blog.jcuff.net/search/label/PHI > > Might be interesting to a few folks on the list. Sorry I can't talk > numbers, it's not fair to talk benchmarks on unreleased product but > there are some fun things we discovered over the last week or so. > > Best, > > j. > > -- > dr. james cuff, director of research computing & chief technology > architect harvard university | faculty of arts and sciences | division > of science rm 210, thirty eight oxford street, cambridge. ma. 02138 > tel: +1 617 384 7647 | http://about.me/jcuff > _______________________________________________ > Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin > Computing > To change your subscription (digest mode or unsubscribe) visit > http://www.beowulf.org/mailman/listinfo/beowulf _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf