Author: Robert Hyatt
Date: 10:10:39 07/01/04
Go up one level in this thread
On June 30, 2004 at 13:11:42, Vincent Diepeveen wrote: >On June 30, 2004 at 05:13:31, Uri Blass wrote: > >>In the following table you can see that the itanium2 is better than the opteron >>inspite having less processors >>see http://www.top500.org/list/2004/06 > >top500.org is dependant upon 2 things > a) how many floating point operations a second each cpu can execute, > and doing it in vectors is fine for it > b) how good the interconnects & routing system is. streaming bandwidth is more >important there than random access latency. > >>Itanium2 1.5 is place 9 in the list when opteron2.2 is place 10 in the list >>inspite of the fact that opteron2.2 ghz has 2560 processors against 1936 >>proccesors of Itanium2. >> >>The person who I talked with him tell me that there is already Itanium2 1.6 and >>1.7 and he claims that the Itanium is new technology and the only reason to use >>the opteron and not the Itanium in the world championship is probably money. > >the itanium is very slow processor for 32 bits engines. > >for diep a 1.3Ghz itanium2 using a very good compiler (intel c++) using PGO gets >the same speed like a 2.0Ghz K7 gets for DIEP. > >So you see that opteron is a lot faster than that. For crafty, you can take an Itanium or an opteron, divide the clock speed by 1000, and get _very_ close to the NPS crafty will produce on that processor. A 1.6ghz itanium is no better or worse for me compared to a 1.6ghz opteron. Opterons are clocked a bit faster, so they are winning for the moment. But should intel produce a 2.4ghz Itanium2, it would (for my program) be just as fast as the 850 I am using in the WCCC (2.4ghz opteron). YMMV. Mine doesn't.. > >computerchess is integer math, not floating point math by the way. > >the itanium2 is only fast for floating point and even there opteron outcheaps it >as you can buy more opteron processors to do the job. You can also buy far more 8086's to do the same job, but the programming is not that easy... > >>He claims that for the same money I can get better things from AMD but the best >>that I can get if money is not a problem is from Intel. > >itanium2 has many issues. like some instructions do not exist on the chip. > >for example for random generator i use rotate a lot and this instruction is not >there. Result opteron is 3 times faster there. > >itanium2 has no division instruction so it cannot divide with just 1 instruction >(there is some binary algorithm using 1/x though to approach divisions) and >loses therefore bits in accuracy when compared to a hardware instruction for the >same price. Strange logic. Cray supercomputers never had a divide either. Caused no problems at all to use the reciprocal computation instead, in terms of accuracy... > >>I want to know if this is really the case? > >for chesssoftware the opteron is 2 times faster nearly than itanium2. Hardly for me. But then I use all 64 bits. Only opteron advantage at present for my code is that it clocks significantly faster than Itanium has so far. > >i hope you realize the price differences. > >quad opteron 2.4Ghz = 14000 euro >quad itanium2 1.5Ghz = 55000 euro (last time i checked) > >the quad opteron is effectively 60% faster for computerchess. > >>I can add that I will not get the opteron for the world championship. >>They told me that they cannot get the opteron148 or opteron248 and the options >>that they can get for me is opteron246(1.4ghz) or Opteron142(1.6ghz) > >You should order it in USA or somewhere in europe, just pay some import tax then >on the processors. > >Is a lot cheaper still than ordering it in israel. > >>I decided that these options are not acceptable and the AMD3000 (2.0 ghz) is >>probably better. >>Uri > >for you AMD 2.2Ghz is even faster than that.
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.