Performance speed up with CUDA possible?

[David]

Hi Pete,

I observed that QuPath sometimes does computations on the GPU. I believe to remember it is during the classification/training of random forests.

Does it increase the performance, if CUDA is installed?

Best

David

[Pete]

Hi David,

Well, I wasn't aware that it did... but the actual random forests implementation is from OpenCV, so conceivably there is some optimization in there that I didn't know about that uses the GPU.  But I didn't think so.

Therefore I don't have an answer to your question; I don't think CUDA would increase the performance, but if you can benchmark it and find out this is wrong then that would be interesting to know.  This situation may change for QuPath at some point, and I'd personally rather have a CUDA-compatible GPU than not have one - given the direction the field is going, and the kinds of process likely to be required in the future.

I'd be less surprised if the rendering of the image (and overlays) on screen already used a GPU, if available, to improve the repainting speed.  Potentially, this might be more noticeable whenever training a classifier interactively, because the screen needs to update more frequently to recolor the objects as their classifications change.

Regards,

Pete

[David]

Hi Pete,

thanks for your answer. I observe the task manager nearly always, to be shure that QuPath is ready with all computations, because I made the experience that it tends to not respond anymore, if I try to do something while it is still computing. Often, it is not obvious if QuPath finished everything, because the progress bar sometimes disappears.

So I regularly see the GPU is used. Most of the time it is around 17% of its capacity. In case something could be parallelized, more GPU might be used.

I personally never used CUDA. In case I install it and observe an effect, I will let you know.

Best

David