Writing a Matlab program for a GPU card in a different machine (in the same local area network)?

1 view (last 30 days)
Can someone please help me write a Matlab program to a GPU that is plugged in a different machine in the same local area network?
I'd like to; - Configure CUDA drivers in the machine that runs MATLAB, to be able to connect with the other machine that has the GPU plugged - Writing a simple Matlab code, possibly a matrix computation - Accessing data to/from the GPU and the Matlab work space
Regards,

Answers (1)

Jason Ross
Jason Ross on 28 Aug 2012
Edited: Jason Ross on 28 Aug 2012
You will need to have MATLAB installed on the remote machine with the GPU to be able to use the two together. As far as I know, there's not a way to just use the GPU remotely without also running MATLAB there, too. The CUDA driver is essentially a display driver, so it's very much tied to the hardware and doesn't have any provision for working over a network. Also, the network bandwidth and latency are going to be very slow compared to the PCI bus on the local machine -- so even if doing GPU calculations over a network were possible, any performance advantage would likely be swallowed by the network cost.
If you have MATLAB installed on the other machine, it's fairly straightforward to use. Install the CUDA drivers for the platform, log in remotely, start MATLAB and start writing code. The only caveat here is that you may need to use the nvidia-smi utility to configure the remote GPU to work properly with Remote Desktop, but that's covered in the installation instructions from nVidia.
You can use the Parallel Computing Toolbox and the Job Manager (or another scheduler) to access a GPU in a worker machine, but that will still require install and setup of MATLAB in some fashion on the remote machine, as well as configuring the scheduler and so on, so this is likely not what you are looking for.
  2 Comments
Friedrich
Friedrich on 28 Aug 2012
Maybe a bit too much for this issue but that should work too:
Buy MATLAB Distributed Computing Server, start one worker on the machine with the GPU and connect to the cluster from the machine on which you are working. In that way you can run your MATLAB code on the Machine with the GPU.
This would be the most expensive way I can think of. Buying a Tesla would be cheaper :D
Jason Ross
Jason Ross on 28 Aug 2012
Yes, I added that as an afterthought in an edit but I'm betting it's not really what is being looked for here. It would most certainly solve the problem, though.

Sign in to comment.

Categories

Find more on Get Started with GPU Coder in Help Center and File Exchange

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!