Author: blass uri
Date: 06:06:04 09/07/98
Go up one level in this thread
On September 07, 1998 at 07:19:22, Jouni Uski wrote: >> >>Is there any proof that hash tables help Fritz5 more than other programs? >> > >Don't be a fool! You can test it in 30 minutes by changing hash size and >comparing in couple of positions. The answer is definitely YES: when increasing >hash by 3x you get easily 20 - 40 % speedup in Fritz, but hardly any improvement >at all with other programs (e.g. Rebel, Mchess, Genius...). > >Other thing is that Fritz5 runs also fine with limited hash as with 8 MB in >my 486 (when hash is full it doesn't stop searching as claimed). > >Jouni 1)I believe that 9 MB is significantly better than 3MB but the question is if 120MB is significantly better than 40MB 2)I gave fritz5 some positions it needed hours to solve and I found that sometimes it needed more time with more hash tables I do not know if it is a problem of windows95 or a problem of fritz5. 3)I found under windows95 that too big hash tables are counterproductive for Junior4.6 (Junior4.6 needs more time to go to the same depth with 100Mbytes hash tables than with 20Mbytes(it looks at less positions but needs more time)) I know that in this case it is a problem of windows95(I think something between 20Mbytes and 70 Mbytes may be optimal). I did not check it for Junior5. 4)I think that doing the nunn match with small and big hash tables between fritz5 and Junior5 is a good idea to see which program earns more from hash tables(better than checking test positions). I think that it is better idea to use windows98 because windows95 cause problems. Uri
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.