My current understanding is that this code is only set up to run on NVIDIA gpu’s given it was written in CUDA. I have one of the last mac book pros that has a CUDA enabled gpu and have successfully run this software…but I recently upgraded to a nice new shiny one that does not have an NVIDIA gpu (intel is the current one).
I’m wondering about how to run this software which I have counted on for a large portion of my dissertation work on my new faster machine. What options are there available for porting this over to openCL? Has anyone else run up against this issue? What are possible solutions?
Thanks!
Hello @spacegirl1923 and thanks for your interest in GPUSPH.
We are strongly aware of the limitations that come from being restricted to CUDA-enabled GPUs, and in fact an OpenCL port of the program is a long-standing item in our internal wish-list. Sadly, we have never had the resources to dedicate to this undertaking, which is far from trivial due to the heavy reliance on C++ for the device code.
Recent developments in the HPC and GPGPU world (such as SYCL and Intel’s OneAPI) would make it easier to migrate from CUDA and support a wider variety of hardware, but GPUSPH doesn’t support them yet, so I’m afraid that a CUDA device is still required at the moment. Help with porting to a more general platform would be more than welcome though 
1 Like