[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20121008145228.GA7975@openwall.com>
Date: Mon, 8 Oct 2012 18:52:28 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: CUDA tweaking to your actual GPU
On Mon, Oct 08, 2012 at 10:18:43AM -0400, Rick Zero_Chaos Farina wrote:
> On 10/08/2012 03:58 AM, Lukas Odzioba wrote:
> > 2012/10/6 magnum <john.magnum@...hmail.com>:
> >> - The OpenCL support for nvidia automatically picks your actual architecture so will likely have similar goods and bads. It's probably possible to pick an arch per format, if we want to.
> >
> > Actually we can do the same by compiling N kernels (each for different
> > sm) and choosing "most appropriate" - predefined for pair <GPU
> > name/format> or GPU arch if we do not have such data.
>
> As a distro maintain I think this is a FANTASTIC idea. I would love to
> be able build kernels for all supported opencl levels, or one, or
> anything in between.
I think there's some confusion here. Even though magnum mentioned
OpenCL in the quote above, I think Lukas' comment applied to CUDA.
For OpenCL, we're currently tuning/compiling kernels at runtime.
Alexander
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.
