So right now as it. Nvidia dose not support multi gpu or sli with there latest gen of graphics cards with older ones. I have a rtx 2080 and a gtx 1050ti… The gtx 1050ti is now utterly useless, but I would like to get it going on something. I am wondering if I can program a layer or extesion or something to use this…
or do I sell it. and get rid of it…
Also in the future I would like to get a AMD gpu going and have it working in tandem with the nvidia gpu. Is that feasible?
It all depends on how you define “in tandem”.
You can always create Vulkan devices for any hardware that supports Vulkan itself. So at the very least, you can create devices, perform operations in one of them, read the results back into CPU memory, and write the data to device memory for the other device.
It won’t exactly be fast, but it will function.
Any higher-performing inter-device communication is ultimately going to be based on those devices being able to talk to each others’ objects. Vulkan supports the idea of that, but it only really allows it if the two devices and their implementations are the same. So your RTX and GTX GPUs cannot use inter-device communication. Nor could NVIDIA/AMD devices communicate.
Again, outside of a GPU-CPU-OtherGPU round-trip.
I would also like to have GPUs A and B able to, at least, efficiently transfer bits between them selves. I get that device groups are for matching GPUs. But mis-matched GPUs that can DMA point to point would be super useful for the Workstation DCC and multi-card multi-monitor situations I deal with.