Author: Vincent Diepeveen
Date: 03:09:47 06/28/98
Go up one level in this thread
On June 28, 1998 at 02:30:52, Tom Kerrigan wrote: >>And if the DB hardware is too big or hot to be inside an ordinary PC, it can be >>placed in an external box with an expansion card to interface with it. But >>would the PCI bus present a bandwidth bottleneck when compared with the >>processor and DB chip speed? Or would the DB hardware have its own processor >>which does the coordination between the DB chips. > >The new 3D chips coming out are probably at least as big and hot as a DB chip, >and cooling them isn't a huge problem. > >Anyway, even if the PCI bus doubled the amount of time the chip took to search a >position, it would still be a tremendous gain. Let's consider these random-ass >figures: > >32 bit PCI bus, 33 MHz transferring a position (say, 256 bytes) = 30 ns * >(256/4) = 1920 ns * 2 (miscellaneous overhead) = >3840 ns to get a chip started on a position > >time for a chip to search the position: >4 ply search in 100,000 nodes (no null move, etc.) >say, 50,000 evaluations (1.5 clock cycles) >another 5,000 move generations (4 clock cycles) >clock cycle = 10 ns (100 MHz) -> >950000 ns to search a position > >So, taking these phenominally conservative numbers, transmitting the position >still only takes 0.5 percent of the time it takes to search it... > >Cheers, >Tom So suppose IBM permits certain people to try to put cpu at card in PC, what is your conclusion about it, is it very easy possible considering communication speed?
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.