|
From: | mpetch |
Subject: | Re: [Bug-gnubg] Performance Gains from 20090915 to 20110525 Codebases |
Date: | Sat, 28 May 2011 16:43:48 -0500 |
User-agent: | Internet Messaging Program (IMP) H3 (4.1.4) |
Quoting Philippe Michel <address@hidden>:
When I wrote 30-40, I meant depending on the CPU. Your big Xeons are a favourable case. I think something like an old Pentium M would profit less and a machine without sse instructions would see a gain well below 30%.
I think non SSE machines are likely the exception to the norm now a days. And I agree that this is specific to my system - was why I made it clear the hardware I was using. I do have some tests going with an older Core 2 Duo (Yes with SSE). I am effectively suggesting that your 30-40% improvement is reproducible.
I could run the experiments without SSE but I'm not sure it would be a good indicator of most typical usage patterns.
I think it also depends on what type of things you are doing in GNUBG. A checquer play rollout of say 46656 trials. Other data I have (most doesn't have to do with the performance gains) shows that Number of eval threads and cache as well as how many plys an evaluation is done at affects performance greatly.
Off topic note: I turned off all the Swap and used a Nice value of -19.
[Prev in Thread] | Current Thread | [Next in Thread] |