Author: Gerd Isenberg
Date: 02:44:38 10/15/03
Go up one level in this thread
On October 14, 2003 at 17:41:11, Vincent Diepeveen wrote: >On October 14, 2003 at 14:29:36, Gerd Isenberg wrote: > >>On October 14, 2003 at 14:15:33, Vincent Diepeveen wrote: >> >>>On October 14, 2003 at 14:13:08, Gerd Isenberg wrote: >>> >>>>On October 14, 2003 at 10:07:10, Ricardo Gibert wrote: >>>> >>>>> >>>>>http://www.wired.com/news/technology/0,1282,60791,00.html >>>>> >>>>>Can this be productively used in a chess program? >>>> >>>>I don't know, simular hardware ressources may be more productive for chess, if >>>>implemented as hyperthreading devices. I guess it's a kind of further >>>>development of SSE and AltiVec technology. With huge register files >>>>(N * 64 * 64|128|256-bit?) and probably SIMD-wise integer instructions >>>>(including popcount?) and fast memory interface, i can imagine that it is >>>>usefull for a lot of nice things, like some eval passes, e.g. a first square >>>>wise and a final scalar product pass. And fill-attack generation, e.g. square >>>>wise in all 16 directions with a specialiced dumb fill routine. >>>> >>>>Gerd >>> >>>this is just floating point arrays. >> >>Aha, well may be a matter of interpretation. >>I havn't seen any instruction set yet. >> >>On the other hand, if float and double arithmetic becomes as fast (or faster) as >>integer, why not use it for eval purposes? >>Gerd > >Where do you put the evaluation logics, those execution units are very small. > >Also you run in parallel at like 64 units or some insane big number and all you >can do is stream data to it. You cannot communicate with other processors very >well. > >It is an floating point array which can execute simple streaming programs. It's >not a supercomputer. > >See it as 64 tiny processors which can run tiny programs while streaming data to >and from it. > >So all you can do with it, with big effort, is calculate matrice with it. >That's all. > >Forget complex software unless you go program fulltime for a couple of years. >By then 25gflops isn't much anymore. > >1.3Ghz I2 is already like 5.2 gflop. > >I am not aware of the price of this programmable array processor, but i would be >amazed if for floating point operations it would be easier to use than simply >buying 10 dual itanium2s. You put in them 1.4ghz chips which cost around $1000 a >piece now or something. That's 5.6 Gflop a processor. Then for $10k you get a >fast router and you get a bunch of network cards for $10000. > >Price: $10000+$10000+10*$5000 or so = $70000 > >Hmm perhaps you can get quite something better than that for that price :) > >But well then you have 20 * 5.6 = 112 gflop. > >Factor 4 faster than this thing and you can only rewrite your program to work >over a network using streaming which works pretty well. In normal C code you can >write it in fact then. And it will run fast. > >For that array thing you need another 2 years fulltime work to get your complex >software to work. So if rewriting to a network is 1 year work and coding it for >that array thing is 2 years work, then trivially the network solution is faster. > >O yeah you also need to design a new mainboard to use that array thing and you >run the risk that your program will be too large to use at the array ;) > >In short faster is the software solution with GPRs *always*. > >If you really want a vector processor, then just buy some 500Mhz clocked vector >processors from Fujitsu. > Ok, i see, but the fact to do 64 parallel calculation "threads" is interesting. The challenge is Opteron/Athlon64, Itanium or G5, which seems really an interesting architecture, too. Considering the floating point performance of these processors, it may be an option to gain from these available ressources by using 32/64-bit float/double SIMD arithmetic for eval. Regards, Gerd >Best regards, >Vincent
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.