Take the 2-minute tour ×
Mathematica Stack Exchange is a question and answer site for users of Mathematica. It's 100% free, no registration required.

I have Mathematica 8.0.4.0 installed on my new MacBook Pro running OS X 10.8. I also have the latest CUDA drivers (and even the CUDA SDK) from nVIDIA. I've noticed that commands like CUDAQ[] take a VERY long time to return 'True'. The CUDA resources paclet is already installed. Here's some examples:

In[1]:= Needs["CUDALink`"]

During evaluation of In[1]:= General::cdir: Cannot set current directory to authserver. >>

In[2]:= AbsoluteTiming[CUDAQ[]]

Out[2]= {112.003663, True}

In[3]:= AbsoluteTiming[CUDAResourcesInformation[]]

Out[3]= {7.734287, {{"Name" -> "CUDAResources"...

Has anyone else seen CUDAQ[] take 112 seconds to return? It's almost like Mathematica thinks the paclet isn't installed or something. Subsequent runs of CUDAQ[] in the same session return quickly. If I kill the kernel and launch a new one, it's back to 100+ seconds again. Other CUDA commands also take a very long time to initially run and then run quickly if run again in the same session. Restarting the kernel gets you back to the excessive times again.

A related question: is there a log somewhere where I can see what the heck the kernel is doing when I type in that initial CUDAQ[]?

Thanks!

share|improve this question
    
I should add that OpenCLLink doesn't have these problems. It seems to be only CUDALink. Also, the CUDA SDK samples build and run just fine on my Mac with no delays. –  Sean Wall Aug 21 '13 at 20:36
2  
Please mention that this question has been posted at community.wolfram.com/groups/-/m/t/101464?p_p_auth=aZ4uSZGg with number of answers there already so that folks do not waste time duplicating effort. –  Nasser Aug 21 '13 at 20:41
1  
Yes, that's true, however the suggestions there didn't seem to help. I was hoping to get a fresh perspective. However, I'll keep that in mind for future posts. –  Sean Wall Aug 21 '13 at 20:55
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.