Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Engine testing: Memory Speed vs NPS at 400 & 32mb hash settings

Author: Gian-Carlo Pascutto

Date: 15:31:25 08/08/02

Go up one level in this thread


On August 08, 2002 at 18:26:18, Aaron Gordon wrote:

>On August 08, 2002 at 16:40:14, Gian-Carlo Pascutto wrote:
>>
>>The latency will be important, yes, because you will spend the time waiting for
>>the data to come from the main memory. But bandwidth? Each probe will be exactly
>> the size of one cacheline or a small multiple thereof.
>>
>>You're not moving huge continous chunks of RAM back and forth in hashing, but
>>randomly distributed tiny amounts of data. What will determine your speed is not
>>the bandwith but the latency.
>>
>>--
>>GCP
>
>Thats odd.. because when I tested with Cachemem the latencies between the two
>settings were basically identical.. yet I get this boost from a 57% memory
>increase. If you'd like you can do similar tests yourself and I can provide you
>with cachemem to test your latencies (and it does a mem bandwidth test).

The speed increase could be due to some other part of the program having
advantage to more memory bandwith. Pulling in a routine that was swapped
out of the cache for example.

If I look at your numbers, the speedup is pretty much irrelevant of the
hash size for most engines, so that would support the above.

--
GCP



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.