Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]


Groups > comp.soft-sys.math.mathematica > #13896 > unrolled thread

Re: OpenCL yes, Cuda no

Started byBill Rowe <readnews@sbcglobal.net>
First post2013-03-19 05:51 +0000
Last post2013-03-31 07:10 +0000
Articles 3 — 2 participants

Back to article view | Back to comp.soft-sys.math.mathematica


Contents

  Re: OpenCL yes, Cuda no Bill Rowe <readnews@sbcglobal.net> - 2013-03-19 05:51 +0000
    Re: OpenCL yes, Cuda no debguy <johnandsara2@cox.net> - 2013-03-31 07:11 +0000
    Re: OpenCL yes, Cuda no debguy <johnandsara2@cox.net> - 2013-03-31 07:10 +0000

#13896 — Re: OpenCL yes, Cuda no

FromBill Rowe <readnews@sbcglobal.net>
Date2013-03-19 05:51 +0000
SubjectRe: OpenCL yes, Cuda no
Message-ID<ki8uca$o3p$1@smc.vnet.net>
On 3/18/13 at 5:36 AM, janos@lobb.com (J=C3=A1nos L=C3=B6bb) wrote:

>This is a Centos 6.3 machine with Mathematica 9.0 Home Edition.  The
>Cuda 5.0 toolkit with the 304.54 driver was installed.  When I first
>issued Needs["CUDALink`"], it downloaded lots of stuffs from some
>Wolfram Server, although the cud a 5.0 driver was already installed.
>It is true that not in the expected location, but the needed soft
>links are all set at the right places.  For example all samples from
>cuda5 are compiled and those wo need just OpenCL 1.0 or 1.1 are
>executing at the Linux prompt.  So I am wondering why CUDAQ[] is
>unable to tell a True.

When I encountered a similar problem, I used
CUDAResourcesUninstall to clear things out then did
CUDAResourcesInstall to set it back up. This cleared the problem
for me. YMMV


[toc] | [next] | [standalone]


#13976

Fromdebguy <johnandsara2@cox.net>
Date2013-03-31 07:11 +0000
Message-ID<kj8ni4$1rh$1@smc.vnet.net>
In reply to#13896
First tell us if / test this.  Are both OpenCL and Cuda known to work
on the same box with any other application (other than what nvidia
provided, which might show a false positive).  There are many issues
people have when getting 3D working on linux machines and Mathematica
channel isn't a place to ask linux X 3D questions.

... and it could still be a linux library issue.  that's a big
possiblility.  when linux hacks the way (ie, the way a driver loads)
then Mathematica can't talk to a driver, it might crash a forked
process's stack, and in the front end look like "returned false".

With little info there's hardly a way to say more yet.

by the way just because nvidia says "~ that enables dramatic increases
in computing performance by harnessing the power of the ~" doesn't
mean it's true :)  They want you to do CUDA not OpenGL, GL, which is
another story, you have no choice anyhow :)

have you tried raytracing?  very slow but splendid output.

[toc] | [prev] | [next] | [standalone]


#13985

Fromdebguy <johnandsara2@cox.net>
Date2013-03-31 07:10 +0000
Message-ID<kj8nhf$1r2$1@smc.vnet.net>
In reply to#13896
It's important to say that OpenGL worked fine, OpenCL is a spank off.
It has everything to do with "forced upgrade" and making proprietary
what was free.

It's fair warning to tell you: Don't Blame Mathematica or Ubantu (and
Linux has 0 to do with it).  The OpenCL people know they are making
things difficult and mean to.

So don't get tangled up depending on it all working fairly for you.
All they want is buy more now it's broken buy more now it's broken.
Do it if you must :)

[toc] | [prev] | [standalone]


Back to top | Article view | comp.soft-sys.math.mathematica


csiph-web