GPUs and increasing RAM usage by codes

[Old posts from the commercial version of ArrayFire] Discussion of ArrayFire using CUDA or OpenCL.

Moderator: pavanky

GPUs and increasing RAM usage by codes

Postby melonakos » Thu Oct 13, 2011 6:36 pm

We recently received the following question via email:

Apart from making the codes execute faster, Can GPUs also support in increasing the RAM usage by the codes. For example, If I try to find eigen values of a matrix of size (5000*5000) in MATLAB on my laptop of RAM 4Gb, then I get out of memory issues in MATLAB. This is because the memory required to find eigen value of this size of a matrix is more than my laptop's RAM. So is it possible through the use of GPUs to solve problems of this size or even more like a matrix of size (1million *1 million) which cannot be solved by the CPU working alone on my laptop of RAM 4Gb.


While theoretically it would be possible to use the GPU to increase the effective size of RAM available above what is available on the CPU only, the hassle of worrying about things fitting properly is not worth the price of buying additional RAM. We recommend that your CPU RAM be greater than or equal to your GPU RAM, otherwise you can run into situations where you want to move data back to the host and cannot.

Jacket does do some things to make more effective use of GPU memory through smart copy techniques, but it would not enable you to launch a problem like you describe above.

Finally, if you laptop only has 4GB of RAM, it is likely that your GPU only has 1GB or less and that it wouldn't really boost your effective RAM size much.

-John
John Melonakos (john@arrayfire.com)
User avatar
melonakos
 
Posts: 503
Joined: Tue Jun 10, 2008 9:49 am

Return to [archive-commercial] Programming & Development with ArrayFire

cron