GPU Computing

I just tried running the following code:

 N = 6;
M = magic(N);
G = gpuArray(M);
??? Error using ==> gpuArray at 28
An error occurred during: setting up the CUDA device. The CUDA error was: (Unknown code: 10100).

And I got the above error and am not sure why.

I'm running Mac OSX 10.6.6 on a Macbook Pro, Matlab 2010_b, an NVIDIA gpu ( Chipset Model: NVIDIA GeForce GT 330M Type: GPU Bus: PCIe PCIe Lane Width: x16 VRAM (Total): 512 MB )

and I just downloaded CUDA Driver Version: 3.2.17 and installed it.

Any suggestions?

Answer 1

[Edric Ellis]

Unfortunately, that GPU is compute capability 1.2, and Parallel Computing Toolbox requires GPUs with compute capability 1.3 and above. (I'm not sure quite why you're getting such a cryptic error message though). There's a post over on the newsgroup explaining why 1.3 is required:

Newsgroup posting about compute capability constraints

Answer 2

[Jeremy]

How do I find the compute power of my (and other) GPU(s)? I tried searching online to no avail and I wasn't sure where to look in my hardware specs.

[Jason Ross]

The Wikipedia articles for some of the various nVidia product lines has the compute capability listed, but it looks like it's not there for the 330M. The GeForce and Tesla lines have the information.

You can also look at Appendix A of the CUDA Developer's Guide, although all cards are not listed there, either.

http://developer.download.nvidia.com/compute/cuda/3_2/toolkit/docs/CUDA_C_Programming_Guide.pdf

Finally, if you compile the examples provided by nVidia, the "Device Query" example script will tell you the compute capability of your machine.

http://developer.download.nvidia.com/compute/cuda/sdk/website/samples.html#deviceQuery

[Walter Roberson]

The 330M is the GT216 compute engine, which is CUDA 1.2 . The 320M is the same.

Unfortunately this means that none of the graphics boards currently sold by Apple for any of their systems are able to be used for this purpose. Apple sells Radeon boards for their higher end systems, Nvidea GeForce 330M for their iMac and MacMini, and Nvidea GeForce 320M for their MacBook Air.

[Mark Shore]

The Quadro 4000 is available in a Mac configuration ($1200 through Apple) and is CUDA Compute 2.0 capable. As far as I know it's the only one.

[Walter Roberson]

Thanks, Mark. I see that the Quadro 4800 is also available in the Apple store. Neither are configurable in the configure-your-own system (at least not for Canada), so it appears you have to buy the ATI and then add the NVIDIA as a separate purchase.

The ads and press releases from other years show the Quadro 4500 and 5400 as previously having been available for the G5 and Mac Pro as build-to-order options.

[Mark Shore]

I just noticed Jeremy had a question there. You can find the compute capabilities of different GPUs here:

http://www.nvidia.com/object/cuda_gpus.html

Answer 3

[Denis Zhegalin]

Hello, I've got the same error.

??? Error using ==> feval An error occurred during: setting up the CUDA device. The CUDA error was: (Unknown code: 10100)

I have two teslas c2050 + quadro for OS purposes. What to test ?

One remark, I use remote desktop access.

[Petter]

You cannot use the GPU via remote desktop access.

Answer 4

[Najeeb Ahmad]

I have TESLA C1060 (as compute only non-graphic device) and getting same error in MATLAB. What should I do???

Note: Using TELSA 263.06 WHQL driver (has TCC capability but I didn't find nvidia-smi.exe to exclusively switch it into TCC mode).

[Jason Ross]

Try the 263 developer drivers from

http://developer.nvidia.com/object/cuda_3_2_downloads.html

This is what we downloaded and it's definitely there.

Answer 5

[Najeeb Ahmad]

Thanks for your reply!

I finally found that Tesla TCC driver doesn't work with Windows 7 32-bit. Now I have installed Windows 7 64-bit and Tesla C1060 TCC driver. It is working without a problem.