Possible to use CUDA and OpenCL side-by-side?

[lepido]

Hello,

I was wondering wether it is possible to use CUDA and OpenCL side-by-side on two different GPUs. Let's say, I have an NVIDIA GPU and an AMD one in my system and I would like two utilize both of them at the same time. Would that be possible?

---
Background info:
Why not just use OpenCL on both: The system is a Jetson Xavier for which only CUDA support is available. The dev-kit, however, has a PCIe slot, which I plan to plug another GPU into. Problem is that NVIDA does not provide aarch64 drivers for their GTX cards (or any GPUs I guess) which leaves me with the only option of using a RADEON. RADEON, of course, has no CUDA support and that's where OpenCL comes into play. In order to use the internal Volta GPU as well as the external RADEON GPU I would need to run CUDA and OpenCL side-by-side.

[undeath]

I can't imagine why this wouldn't be possible. Why not try it?

[philsmd]

yeah, we actually had a similar request/question like this on IRC and the "only problem" is to get the different drivers correctly installed at the same time.
It's easier on some operating systems compared to others. for instance on windows I think it's quite easy... while on linux it can be a problem in some very specific situations... and some users even had to manually change the xorg.conf file etc to get it working.
but all of this is not really a hashcat-related problem. if both drivers (OpenCL runtimes) are contemporaneously installed and working, you could just test it with e.g. clinfo on linux (to rule out that it could be a hashcat issue... i.e. if even clinfo won't be able to see the different devices/platforms, it probably won't work with hashcat and we are sure it's not a hashcat problem).

[lepido]

I can't imagine why this wouldn't be possible. Why not try it?

Thanks undeath and philsmd. I will try it eventually. So far I haven't received the Jetson Xavier yet (it's in the mail). I bought it for some AI stuff I am doing. My thinking was, I can also use it for hashcat while it is not bothered with transfer learning. I do not own a dedicated GPU, though, and wanted to verify it works (theoretically) before getting one.

[lepido]

if both drivers (OpenCL runtimes) are contemporaneously installed and working, you could just test it with e.g. clinfo on linux

Just to avoid any misunderstandings. I am talking about running one GPU via the CUDA interface (NVIDIA Volt) and the other one (RADEON) via OpenCL at the same time. I am not looking for running two different GPU models/vendors both under OpenCL. Is that the same understanding you had and I should read your message that way?

[philsmd]

oh, I see. yeah, I meant any backend supported (OpenCL or CUDA with newest beta version), not just the "OpenCL runtime", but any runtime. That's cleary my fault in writing that incorrectly.