MI100 will be able to run Open

[davide445]

Wanting to test some compute+graphics app on this one, but looking better I understand has no video output (no pbm will be in the cloud), but seems has also no certification to be used with any graphics API.
Meaning any UI will just get software render on the CPU?

[agd5f]

Wanting to test some compute+graphics app on this one, but looking better I understand has no video output (no pbm will be in the cloud), but seems has also no certification to be used with any graphics API.
Meaning any UI will just get software render on the CPU?

MI100 has no GFX pipeline and no display hardware. It's basically a dedicated compute card. I don't see any reason to try and run a UI on it.

[davide445]

MI100 has no GFX pipeline and no display hardware. It's basically a dedicated compute card. I don't see any reason to try and run a UI on it.

This get me to a couple of questions
- my idea was to evaluate the possibility to create an interactive application using ROCm. If no graphics capabilities are inside CDNA I suppose I will need to use a second RDNA gpu coupled with a CDNA to get out the realtime graphics I need. But in this way the only way to communicate btw the two will be using PCIe? I didn't see any IF on RDNA. That will be unfortunate since i.e. a 3090 has NVLink. There is any example of graphics powered solutions using CDNA as acelerator?
- just for curiosity, the fact didn't make sense mean also its impossible to run any graphics on CDNA? Just since also didn't make sense to use CPU for graphics applications, but in fact I'm using since months Maya on a CPU only VM.

[agd5f]

This get me to a couple of questions
- my idea was to evaluate the possibility to create an interactive application using ROCm. If no graphics capabilities are inside CDNA I suppose I will need to use a second RDNA gpu coupled with a CDNA to get out the realtime graphics I need. But in this way the only way to communicate btw the two will be using PCIe? I didn't see any IF on RDNA. That will be unfortunate since i.e. a 3090 has NVLink. There is any example of graphics powered solutions using CDNA as acelerator?

There is no graphics engine on MI-100. If you need graphics capabilities, you'll need a GPU with a graphics pipeline. Vega20/MI-50/MI-60 have a graphics pipeline.

- just for curiosity, the fact didn't make sense mean also its impossible to run any graphics on CDNA? Just since also didn't make sense to use CPU for graphics applications, but in fact I'm using since months Maya on a CPU only VM.

You could theoretically emulate a graphics pipeline using compute shaders, but nothing like that exists at the moment as far as I know.

[davide445]

There is no graphics engine on MI-100. If you need graphics capabilities, you'll need a GPU with a graphics pipeline. Vega20/MI-50/MI-60 have a graphics pipeline.

Vega it's an old architecture. If we want to use CDNA2 and provide rendering capabilities we will need either to implement rendering as compute shader (possible and we will have a test on this) or use two GPUs, one for rendering (i.e. RDNA2) and one for compute (CDNA2).
What I didn't get is how are suppose to communicate so far. Using ROCm or OpenMP trough PCIe I suppose.