Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 7 Oct 2012 23:28:14 +0400
From: Solar Designer <>
Subject: Re: CUDA tweaking to your actual GPU

magnum -

Thank you for running these tests!

On Sat, Oct 06, 2012 at 04:15:23PM +0200, magnum wrote:
> - This test should also include tweaking BLOCKS and THREADS but everything was left at default. Some auto-homing like we do in OpenCL should be fairly high priority for CUDA. The outcome *may* be totally different with optimal values.

Yes, or at least we should determine and document optimal settings for
some common GPUs.  I think that GTX 570/580/590 are most important now
(of NVIDIA's), so maybe the defaults should be tuned for these, if we
have compile-time defaults at all.  Maybe we should do that even if we
keep the default arch type at sm_10 for compatibility with much older
GPUs, or maybe we should make the default arch type sm_20 (and accept
that owners of much older GPUs will absolutely have to change that).


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.