Author: Vincent Diepeveen
Date: 09:24:41 02/16/04
Go up one level in this thread
On February 15, 2004 at 15:26:45, Slater Wold wrote: >On February 15, 2004 at 15:22:03, Joachim Rang wrote: > >>On February 15, 2004 at 15:14:16, Slater Wold wrote: >> >>>On February 15, 2004 at 15:09:57, Bob Durrett wrote: >>> >>>>On February 15, 2004 at 15:06:26, Slater Wold wrote: >>>> >>>>>On February 15, 2004 at 14:34:08, Bob Durrett wrote: >>>>> >>>>>> >>>>>>I envision a standard equipment rack with 32 or 64 Hydra cards, a power supply, >>>>>>possibly a conventional computer for orchestration, and fans. >>>>> >>>>>Do you think Hydra was running on a single computer, with a single card? >>>> >>>>Actually, that is what I thought. What was actually done? >>> >>>4 Dual Xeon 2.4Ghz PCs & 8 FPGA cards (2 per PC) >>> >>>The PCs are probably $5k a piece. The cards, about $500 a piece. >>> >>>$25k worth of hardware, just to draw Shredder 8 ($50 program) on a PC. >>> >>>Still think Hydra is the 'engine of the future'? :) >>> >> >>be fair here: Shredder was running on a Dual Xeon 3.06 GHz, probably 10k > >Ack! I'm not sure about $10k, but it's not an average PC, no. 5x SCSI raid inside machine. 40 kilo machine. >>and think about the potential for Hydra. Soon we have PCI-Express more bandwidth >>and the FPGAs ware running @ 33 MHz this is easy to tune in contrary to the >>CPU-Clocks. > >Talk to anyone who knows anything about FPGAs. They will never catch the PC, in >terms of price OR performance. In theory they are faster, but in practice they aren't because of the huge cost and development path needed to develop hashtables on such a card. >Sure, 8 FPGAs are fast. But you ever see 8 Opterons going? :) >(And 8 Opterons would be cheaper!!) And development of a program is way easier. I hope you realize that compiling hydra takes half a day up to a full day. It's special fpga compilers that are a bottleneck here! Change 1 byte and another day time needed to compile! So software development on a regular basis is always better. In theory you could however order a 4 processor opteron system in order to compile quicker. >>I don't thin, that Hydra will be a commercial success but I think we might see >>beautiful performance in the near future. > >Correct. Neat idea. Not going anywhere. All depends upon the Sheikh of the emirates. When he wants to sponsor a big match for Hydra then we will see it more and more. >>regards Joachim
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.