Author: Robert Hyatt
Date: 13:26:31 06/30/98
Go up one level in this thread
On June 30, 1998 at 14:19:37, Dezhi Zhao wrote: > >On June 30, 1998 at 13:24:45, Robert Hyatt wrote: > >>On June 30, 1998 at 10:47:26, Dezhi Zhao wrote: >> >>> >>>On June 29, 1998 at 18:57:34, David Fotland wrote: >>> >>>> >>>>Field programmable gate arrays range from 40,000 gates at $3 to over 500K >>>>gates with 270K bits of RAM, and can clock up to 80 Mhz. This is at least >>>>as dense and fast as the Deep Blue chips. See www.xilinx.com for example. >>>> >>>>It would be easy to put together a PCI card with one of these chips on it. >>>>Chess developers would then all have access to similar hardare >>>>technology as Deep Blue. If someone designed the basic hardware eval >>>>and search blocks, it would be easy for someone with a software background >>>>to modify the evaluation function and reprogram the FPGA. >>>> >>>>I did a short search and didn't find anyone already selling an FPGA PCI >>>>evaluation card, but there may be one. We may even be able to convince >>>>an FPGA vendor to design the PCI card for the publicity. If not I could >>>>do it, or anyone with a little bit of hardware background. >>>> >>>>Anyone interested in using or buying such a card? With a little volume >>>>the card would be inexpensive. Perhaps $25 since it just a small PC board, >>>>a cheap connector, and the FPGA chip (which would also be the PCI interface). >>>> >>>>David Fotland >>> >>>It seems to me a great idea. I just visited www.xilinx.com. Could you >>>specify a particular page in it for me to start with? If it could be around >>>$25, >>>your idea would certainly feasible. With the PCI card, commercial programmers >>>can get rid of dangles. >>>I had no FPGA experience. I got a question to hardware experts here: >>>How difficult to implement it? >>> >>>Best Regards >>> >>>Dezhi Zhao >> >> >>not hard. this is *exactly* how the original 160K nodes per second Belle >>was built... Note that it makes modification a process of throwing out the >>old and burning the new... > >Dr. Hyatt: >Oh, one has to buy a special burning device. How much speedup can >I expect from the card? A ballpark number is OK for me. My program >currently search around 100K nps. Can I reach 1M with the card? > >Thanks > >Dezhi Zhao hard to say because the "card" idea is bady flawed. A single chip won't be near enough. So it is doubtful that a single FPGA would work either. And before you get excited, check out the price of those 500K gate arrays. They are in 4 digits, *not* 2. So you start talking thousands of dollars to do something effective, and they are very much slower than the ASIC's in DB. maybe 100x slower.
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.