[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Optimal Setting for Eval Threads
From: |
Philippe Michel |
Subject: |
Re: [Bug-gnubg] Optimal Setting for Eval Threads |
Date: |
Thu, 13 Sep 2012 22:26:49 +0200 (CEST) |
User-agent: |
Alpine 2.00 (BSF 1167 2008-08-23) |
On Wed, 12 Sep 2012, M. J. Mannon wrote:
All the references I can find suggest setting "Eval Threads" to the
number of *cores* on your system. Most of these sources, I fear, refer
to CPUs that do not have hyperthreading.
I have a 4-core CPU. With hyperthreading (each core runs 2 threads), it
has 8 threads.
So, is the recommended setting for me 4 or 8?
Hyperthreading helps, but less than a real core. On the other hand
multiple threads have a cost related to contention issues.
Here is what I get with a real quad-core :
threads normalized speed
1 1 * 100
2 2 * 91.6
3 3 * 89.3
4 4 * 87.1
and with a dual-core with HT :
1 1 * 100
2 2 * 91.3
3 3 * 69.5 or about 89 * ( 1 + 1 + 0.3 )
4 4 * 58.2 87 * ( 1 + 1 + 0.3 + 0.3 )
Your mileage may vary (this was a match analysis on an old Xeon processor,
rollouts on a Core i-something may give different ratios), but I would
guess in your case the 5th to 8th threads would give some (possibly modest
for the last ones) improvement.
For an overnight job, I would use 8 threads ; if I was doing something
else at the same time probably only 4.