I am testing Houdini 19 XPU render with the bar scene, it took 7M 44S to render with XPU on an 8cores AMD and a GTX 1650 with 4GB of VRAM, some bits don't seem to work and I get an Optix error, but it's darn fast, CPU 1H 23M 40S, but much cleaner render, but still, XPU is very promising indeed!
SIgor420 Nice, thanks for sharing. I wonder what are the minimum specs for XPU and Karma. For Renderman I know min spec for your GPU is at least 12GB or VRAM.
SIgor420 Nice, thanks for sharing. I wonder what are the minimum specs for XPU and Karma. For Renderman I know min spec for your GPU is at least 12GB or VRAM.
There isn't really a set minimum VRAM, it depends on the size of the scene. More VRAM means being able to render more complex scenes with more, larger textures.
From the system requirements page: GPU acceleration currently does NOT make use of multiple OpenCL devices (i.e. two or more graphics cards) but this may change at a future date. An exception to this is Karma XPU, which is currently in Alpha.
Chris McSpurren Senior Quality Assurance Specialist SideFX
GCharb Since they only support Nvidia, I wonder if they'll ever support NVLINK?
XPU uses Optix, an Nvidia library. Perhaps Mark can speak to Nvlink support.
I don't think we've tried Nvlink, but we have tried multiple GPUs. In fact, one sharp eyed person noticed that in this Entagma video, it looks like there are 4 GPUs working with alongside the CPU to render the image: https://youtu.be/tI4zqTezniI?t=602 [youtu.be] (top right corner of the Karma render).
mark I don't think we've tried Nvlink, but we have tried multiple GPUs. In fact, one sharp eyed person noticed that in this Entagma video, it looks like there are 4 GPUs working with alongside the CPU to render the image: https://youtu.be/tI4zqTezniI?t=602 [youtu.be] (top right corner of the Karma render).
Sharp eyed indeed, this is very cool, now if GPU prices could come down, that would be perfect timing, thanks for the info!
A second test, this time I build a simple Solaris scene with MaterialX shaders, I reduced the scan model to 280k polygons, this time I get no OptiX error, still rendered on an AMD 8-core CPU and a GTX 1650 GPU with 4 GB of VRAM, I used no denoise, I would love to know how fast this renders under XPU compared to CPU only, but with a decent CPU/GPU combo!
I rendered to mplay at 1280 X 720 pixels XPU: 10:49 CPU: 55:07