Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 200 Million NPS in 97 not 1 Billion NPS wrong again (nt)

Author: Robert Hyatt

Date: 08:13:18 07/21/00

Go up one level in this thread


On July 21, 2000 at 00:10:41, Dann Corbit wrote:

>On July 20, 2000 at 22:26:26, Robert Hyatt wrote:
>
>>On July 20, 2000 at 22:08:29, Dann Corbit wrote:
>>
>>>On July 20, 2000 at 22:02:57, Chris Carson wrote:
>>>
>>>200 million was average throughput for all phases of the game.
>>>
>>>1 billion peak.
>>>
>>>"Raw" throughput was 14,400,000,000 nodes per second (480 chips times 30 million
>>>NPS/chip), but the coordination cost was obviously very high.
>>
>>
>>The 30M per chip is wrong.  earlier versions of the chip were 2.0M nodes
>>per second.  The last batch or two (same exact hardware but the manufacturer
>>did a better job making the chips) ran at 24mhz, which was 2.4M nodes per
>>second.  so the right number is 480 * 2.2M.  I say 1B for simplicity...
>
>Quite right.  I was thinking of the numbers for Hsu's shrunken chip.


That's another frightening thought.  :)




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.