memory issues

[TolgaDuz]

trafficstars

Hi @WPZgithub ,

I always get an cuda out of error warning. The code doesnt work on my gpu or CPU. For the gpu I get the error message from issue #4 and for the cpu I receive follwing error message:

Investigating in some threads, I found that it is often due to the immense size of your network. And a probable solution for this, would be to downsample the batch size. But Your code isnt working with batches right?

What could be done here?

Best, T

[WPZgithub]

Hi @TdzBAS , Our impelmentation uses full-graph rather than mini-batch processing by design. Sorry that our approach doesn't take more consideration in reducing memory requirements. I've successfully tested it with 3,000 highly variable genes on a single NVIDIA A100 GPU (40GB memory). Could you please provide: 1) Your GPU specifications and 2) the size of your input data?

Best, WPZ

[Yuzhang0102]

Hi @WPZgithub , I met the same error, I used the 2080ti GPU (12GB memory), and my dataset size shows below:

I use all expressed genes, not the highly variable genes. Is it better to use highly variable genes? Can I run CEFCON on two GPU?