I want to run Quantum Espresso - the GPU version on the DevCloud. However, because the GPU version is written in Fortran-CUDA only PGI can compile the code - It might be a dumb question but - can intel's new GPU support PGI/CUDA? my guess is no... In addition, Quantum Espresso comes with an OpenMP version, and I think it might be modified easily to work with the directives to offload it - teams, distribute etc. - i'm wondering if there's someone working on it already
You might be lucky, but I would be surprised as this is a pretty specialized physics package and it is huge. I tried the makefile with IFORT oneAPI and it came up with circular references.
I suggest LINUX and NVIDA and PGA. Unless you want a lot of messing around, it is 5800 files and some weirdness in the file arrangements.
Intel Fortran does not have CUDA support. The idea is that the oneAPI API will support GPUs from multiple vendors, but that hasn't happened yet, as far as I know.
I was reading about a academic in South Australia, he wanted a supercomputer, but his application failed, so he went around and scavenged a great number of old computers, linked them together with some code and created a behemoth. In LINUX of course.
Money can buy a lot
A really intelligent person with limited money can sometimes do as well.
Finally - it will be simpler when Intel buys NVIDIA