Open dregala3 opened 6 years ago
Specifically, I only need to test Faster RCNN on one large image (4000 x 4000). Is it possible to use multiple GPUs for testing on one large image? Or should I just upgrade my GPU to a bigger memory size? Training the model works fine but testing on this image with one GPU results in a CUDA out of memory failure.
I need to use multiple GPUs to test Faster RCNN on large images. Microsoft has a link to set up multiple GPUs for distributed learning and parallel training https://docs.microsoft.com/en-us/cognitive-toolkit/Multiple-GPUs-and-machines#5-data-parallel-training-with-1-bit-sgd
However, I don't need to execute parallel training. I just need to have a large GPU memory during testing. If I follow the instructions on Microsoft's page, will it set up multiple GPUs for me to use during testing too? If not, how can this be enabled?