Author: Robert Hyatt
Date: 14:09:18 09/08/03
Go up one level in this thread
On September 08, 2003 at 15:06:31, Vincent Diepeveen wrote: >On September 08, 2003 at 14:53:57, Robert Hyatt wrote: > >>On September 08, 2003 at 14:30:41, Vincent Diepeveen wrote: >> >>>On September 07, 2003 at 14:37:17, Joost Buijs wrote: >>> >>>diep is getting tested on the 248 right now by Johan de Gelas. both amd64 and >>>quad opteron. >>> >>>www.aceshardware.com >>> >>>regrettably only they want to test 32 bits mode... ...but well diep is NUMA. I >>>have shipped a NUMA version of diep to them that can go up to 4 cpu's. So that's >>>interesting to see how it does. Johan is very good in testing. also he does not >>>cheat. >>> >>>some other sites that tested diep in the past cheated. when the production >>>version of those machines was released and i tested with same executable, it >>>appeared that intel had cheated by 5% in speed. >>> >>>Also the speedup from SMT they had with diep was higher than when some others >>>who had purchased dual Xeons, could not get to. >>> >>>www.aceshardware.com doesn't do that. Johan is a very objective testers. >>> >>>His main intention is to figure out how much the added branch prediction table >>>will do for DIEP. >>> >>>P3 has 512 entries BTB (branch target buffer) >>>K7 has 2048 entries BPT (branch prediction table) >>>opteron/amd64 has 16384. >>> >>>Intel runs behind for complex software. >> >>That is irrelevant (the size). The Intel BTB is _far_ better than >>that on any of the other processors you mentioned. >> >>I'll let you do the research to find out why the new Intel branch >>prediction is better. Hint: It detects patterns and predicts based >>on the pattern of past taken/not-taken cases. It's clever. >> >>If bigger were always better, 2M of direct-mapped cache would be better >>than 512K of 16-way set associative. It isn't. >> >>Quality still counts... > >You are wrong in one respect. You underestimate how advanced the AMD prediction >works. > >The K7 branch prediction of 2048 entries was using the alpha 21264 way of >predicting. > >That is *very* advanced. It is OK. It is not as good as the PIV. > >For complex software when both are very advanced, more is better of course. > >First of all branch prediction of opteron works better. Secondly it's like 16384 >entries. I forgot how big the P4's is. Probably also quite bigger than the poor >512 entries they have put in the P3. > >But still Opteron outguns everything at *all* fronts > - 64 bits, p4 and k7 are all 32 bits cpu's > - 16 GPR registers, only cpu's like itanium from $8000 a piece have more > - highest clocked RISC cpu > - bigger branch prediction table than any cpu i know > - BIGGER LEVEL 1 CACHE, even than itanium which has a poor 32KB. > - shorter stage than P4, so smaller branch mispredicton penalty > - capable of getting a high IPC and even at floating point it can > effectively retire 1 multiplication a clock (without cheating). > >And very important: you can buy it NOW. Can we turn this rambling recital into something about automobiles, airplanes, and wombats as well? The point was BTB. Not cache size. Not word size. Not clock speed. Etc. The PIV is better, period, in how it does BTB to recognize patterns that span several outcomes for the same branch. My point is that "bigger is _not_ always_ better." All you talk about is "bigger". I will have some opterons to play with soon. We'll see if "it is better on all fronts." > >> >>> >>>Best regards, >>>Vincent >>> >>> >>>>On September 07, 2003 at 11:43:26, Michael P. Nance Sr. wrote: >>>> >>>>>On September 07, 2003 at 11:31:30, Gian-Carlo Pascutto wrote: >>>>> >>>>>>On September 07, 2003 at 10:54:47, Michael P. Nance Sr. wrote: >>>>>> >>>>>>>If I may,how can I get that version of WS 2003 (64 bit) so I can run Deep Sjeng >>>>>>>on My dual (246) Operton also? I'm at a Crossroads now, on what OS to use with >>>>>>>My Operton. I want to be able to go 64 bit.I wouldn't be able to accomplish >that with Windows XP Pro. Please fill in the gaps for Me.>>>>Mike >>>>>> >>>>>>You cannot get it. According to other posters here, there's a beta, but >>>>>>I don't think it is available to members of the general public. >>>>>> >>>>>>The only way to run 64 bit programs on the Opteron right now is to >>>>>>use Linux. >>>>>> >>>>>>-- >>>>>>GCP >>>>> >>>>>When might a commerical version of WS 2003(64 bit) be available to be purchased >>>>>by Us Peons? I don't suppose that there is some kind of a restriction on using >>>>>Linus with My Opteron or is that a dead Horse also? If so,what would be the cost >>>>>compairson comparired to WS 2003,would It work?>>>>Mike >>>> >>>>It will take a few months before a 64 bit AMD version of windows will hit the >>>>streets. There seems to be a beta version of windows XP for AMD-64 to. >>>>GamePC already did a preview. >>>>http://www.gamepc.com/labs/view_content.asp?id=amd64xp&page=1 >>>>Probably it is worth while to wait somewhat longer, AMD will release the 2200 >>>>MHz. 248 Opteron very soon according to the inquirer. >>>>http://www.theinquirer.net/?article=11371 >>>>This will be a real beast!
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.