help-octave
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: CUDA acceleration in GNU Octave


From: David Grundberg
Subject: Re: CUDA acceleration in GNU Octave
Date: Sun, 22 Nov 2009 17:02:22 +0100
User-agent: Thunderbird 2.0.0.23 (X11/20090817)

Chengqi Chang skrev:
On Sun, 22 Nov 2009 21:39:44 +0800, John W. Eaton <address@hidden> wrote:

On 22-Nov-2009, Chengqi Chang wrote:

| I think it would be great if someone can provide a "bridge" -- like MPI
| toolbox(http://atc.ugr.es/javier-bin/mpitb) -- connecting Octave and CUDA
| computing libraries. Besides, in my opinion, it matters little whether
| this "bridge" is universal or tied to a single vendor (Nv/ATI). Both of
| the two choices have merits and drawbacks. If it is tied to a single
| vendor, it can be easier to take advantage of some special features of the | GPU and it also has a larger possibility to get donation from that vendor.

Are the libraries that take advantage of this single vendor
distributed under a license that is compatible with the GPL?

If so, then I think this would be acceptable, but not great as it is
still ties you to a single vendor.

If not, then you would not be able to distribute .oct files that
require these libraries, and such an extension would never become a
part of the core Octave itself.  The Octave project is about software
freedom, not about promoting links to proprietary software, or
encouraging users to lock themselves into single-vendor solutions.

jwe
GPL compatibility depends on how the libraries use CUDA -- if the libs only call them, GPL can be retained. Since there are limited numbers of popular vendors, it is possible to initialize different projects. Once you buy your graphic card, you are actually tied to a vendor. BTW, I like the freedom philosophy as well, while I think whether the freedom is pure or conditional on some exogenous constraints can be equally acceptable.


So freedom has a price, and it is parallel computing? Come on!

Besides, I think you are underestimating the license clash between Nvidia's CUDA license and the GNU GPL.

I think OpenCL will be a more vendor-agnostic way to go.

David



reply via email to

[Prev in Thread] Current Thread [Next in Thread]