Hello - thank you for all of your answers so far.
How can I tell if the nvidia backend is being used or not? I ask because for my little test loop, the CPU still pegs at 100% and the performance is about the same as before I put in the RTX3070. I’m running under ubuntu.
I am just averaging two lists of int16’s, and 100 calls takes about 5 minutes.
Is there some debug I can enable, or a config I can print that would let me know?
Thank you in advance.
Hi @ronhume, sorry for the late reply!
I’m afraid I will need more context to answer this. In which library / tool are you trying to use CUDA acceleration? The compiler?
Thank you. I am using Concrete. Sometimes with DIrect Circuits and sometimes using compiled circuits. But my primary question was how to know if I am getting CUDA acceleration. I guess I’m most interested in seeing if acceleration is helping with execution time. Compile time is not really a concern for me - but execution of some circuits - even when using large tensors on a 64 core threadripper is quite slow and I was hopeing that CUDA acceleration would useful in that regard. I’m getting good CPU parallelism but any additional boost would be really helpful. I hope that makes my question clearer?
Hey @ronhume, GPU acceleration is not exposed from Python at the moment. It’ll be configuration options in the upcoming releases
Ok thank you so much.
You guys are doing a great job. Thank you for the work you do.