I need to learn how Jcuda deals with enabling CPU cores to simultaneously run more tasks on the GPU.
I have not yet used „Hyper-Q“ as such, dedicatedly. All I could do would be to try and port the https://github.com/zchee/cuda-sample/tree/master/6_Advanced/simpleHyperQ sample to JCuda right now. It does not look so complicated, but I’m not sure that I could allocate the time for that right now. If you try to port it and have any questions, feel free to post them here.
I need Hyper-Q technology in Jcuda example. Marco, you sent me the link cuda-sample/6_Advanced/simpleHyperQ at master · zchee/cuda-sample · GitHub before. Unfortunately, it is with Cuda.
As I mentioned (also via mail): I haven’t used HyperQ/V. You could try to port the example from CUDA Runtime API to CUDA Driver API (which then could be similar to what you’d do in JCuda). Yes, this might take only half an hour for me (iff everything works as expected), but I don’t see a reason to do this right now. If you want to try to use HyperQ, nothing is preventing you from doing this. That’s a (tiny) part of what you’d get your PhD title for.
As you know, I am a beginner in cuda. Really, I learn from you. When you have time, can you try it, please? this will be wonderful.
I’m also not a CUDA expert. And there’s not much to „learn“ in that regard from porting a sample. You can look at the sample, and when you see a cudaEventRecord you look it up and replace it with a cuEventRecord from the driver API. Then you see a cudaMemcpy, replace it with cuMemcpy.... There’s a cudaStreamCreate, well, let’s try cuStreamCreate with that one. You could do this. You could ask more specific questions when you’re stuck. You could create a pull request on GitHub with the new sample, and contribute to the open source world. I’ve done that for 20 years, and essentially wasted the so-called „best time of my life“ with that. Why don’t you do that? Maybe I’ll port that friggin sample. But I’m not really motivated to do that right now.
I will try Marco. Thanks a lot.