R2011a on Windows HPC Server 2008 R2 cluster can't see CUDA card
2 views (last 30 days)
I have computers with a GeForce GTX 470's running Windows HPC Server 2008 R2. I use 3 computer as client, Head Node and Compute Node. Compute Node computer only have GeForce GTX 470. But others have a ATI graphic card.
However, when I run below matlab code in client, it errors out.
matlabpool hpc % hpc is a scheduler of HPC Cluster
Error message is below.
??? Error using ==> spmd_feval at 8
Error detected on lab(s) 4
No device supporting CUDA was found.
currentDeviceIndex.m at 7
deviceProperties.m at 8
GPUDevice>GPUDevice.current at 34
gpuDevice.m at 18
Is there additional configuration I need to do to matlab to get it work with GPU?
Jason Ross on 4 Oct 2011
In the HPC Server setup, you need to indicate that the GPUs need to run with a desktop/console open.
This applies to cards that require a display driver to run CUDA and cannot be run in "service" mode.
More Answers (2)
Elwin Chan on 5 Oct 2011
The easiest way to set a job environment variable from MATLAB is to use the "JobDescriptionFile" property on the HPC Server scheduler object. You can set this using the Configurations Manager or directly on the scheduler object returned from findResource().
You need to use the HPC Cluster Manager or Job Manager UI that is supplied with HPC Server to create the job XML file. When you create a new job in the UI, there is an "Environment Variables" section where you can add in job environment variables. Once you have done this, you can click on the "Save Job XML File" button and then use that file as the "JobDescriptionFile" in the MATLAB configuration.