Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Fri, 26 Mar 2010 09:53:26 +0300
From: Solar Designer <>
Subject: Re: core by core speed

On Mon, Mar 22, 2010 at 07:16:56PM +0100, wrote:
>  new test :

Thank you.  This shows that there's no impact from the Idle setting.
That's great to know.

It also shows some slowdown per-process as you increase the number of
running processes, but such slowdown is quite normal for a Core i7 CPU,
which lowers the clock rate as more cores are brought to use and which
runs two threads per core when you run 8 processes.  I've compared your
numbers against those for my benchmarks (different ones) of a Core i7 920
CPU, and they're similar.  When going from 1 process to 4, I was getting
a 28% slowdown per-process.  When going to 8, I was getting a 33%
slowdown per-process (compared to having only 1 process running).  This
was without MPI involved.  In your case, you're getting a 43% slowdown
per-process when going to 8 processes, which is slightly worse than "my"
33% slowdown, yet I think it is normal, especially considering that in
your case there should have been frequent L3 cache and memory accesses.

Core i7's SMT (Hyperthreading) behaves surprisingly well indeed, making
this quad-core CPU actually comparable to 8-core systems in benchmarks.


Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.