Author: Dieter Buerssner
Date: 01:55:06 10/25/04
Go up one level in this thread
On October 25, 2004 at 04:12:08, Ed Schröder wrote: >On October 24, 2004 at 16:51:06, Ed Schröder wrote: > >>On October 24, 2004 at 14:14:29, Jouni Uski wrote: >> >>>Maybe this have something common with You problem. I have noticed, that UCI >>>engines (specially Yace) sometimes run 25%(!) slower than normally under >>>Fritz7. I discussed this with Dieter B: but we never cleared the reason. >>> >>>Jouni >>> >>>PS. Another bug in Fritz GUI still there with latest Fritz8: UCI parameters are >>>not used always only occasionally (depends which engine is loaded first)... >> >> >>Jouni, I hope to reply soon on this, if possible tomorrow. Need to check some >>things first. >> >>Ed > > >I had some informative talks with the ChessBase crew. They could not explain the >phenomenon I addressed but they found the issue interesting enough to do some >research themselves. > >On the issue of Yace and its loss in speed let me quote the words of Mathias >Feist (with permission) as a whole: > > > Quite some time ago I already wrote an article about that, but I don't > remember where. I think it may have been some ChessBase Magazine. > > It's indeed a problem with processor time distribution. The foreground > process gets considerably more processor time than the background > processes. > > If you play two Fritz engines against each other, you won't have a > problem because both are running in the same process. If you are running > two UCI engines against each other, both have their own process and > therefor no problems occur either. You have to watch out if you have a > Fritz and an UCI/Winboard engine playing each other. > > Switching the processer time distribution to optimization for background > processes would work, but generally it's not a good idea. It distributes > processor time evenly and makes the computer hard to operate. > > The UCI engine will be running in the background. Now, if Fritz is in > the foreground, the Fritz engine will receive more processor time. All > you have to do is to push the GUI into the background after starting the > match. The easiest way to do this is to open the task manager and have > it in the foreground (=title bar blue). Nice thing is that you can > easily monitor processor time distribution this way, too, because the > UCI engine appears in the process list together with Fritz. > > Mit freundlichen Grüßen > Best regards > Mathias Feist > ChessBase GmbH > >========= > >If I have understand his words correctly the solution would be to put the match >directly after it has started on the taskbar. This adresses the issue of unequal time usage in ponder on matches (or parallel analysis of more than one engine) on one CPU. Having some other window in the foreground was already enough here, and Mathias' explanation seems correct. It does not adress the issue, Jouni has raised (IIRC). That was with ponder off/one engine analysis. BTW. I never saw the problem Jouni had reported myself. 25% sounds much. I however have seen 10% difference in speed, running without GUI under very similar condoitions. Actually the same run, and the same (search) routine called twice from two slightly different places in the code. It was totally reproducable. Only the called function needs CPU time, caller overhead is zero. So, perhaps some strange cache issues can happen. Regards, Dieter
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.