i just try to test kamra xpu and i'm notice that if i render the same scene (large scene) in linux and windows It behaves differently

in windows both cpu & gpu working
and in linux the gpu failed

i test it in a few scenes (also very large scene) and all of them are the same (in linux it failing and in windows it working)
(I guess it will be like that for everyone)

why in linux its failing and in windows not? its the same vram .
and if its the nvidia windows drivers why its happening?

( i have 4090+3090)

On Windows, Microsoft/NVidia have implemented a way for Optix GPU memory to "spill over" to main CPU memory.
https://nvidia.custhelp.com/app/answers/detail/a_id/5490/~/system-memory-fallback-for-stable-diffusion [nvidia.custhelp.com]

There is not a lot of documentation or info about this feature, but we have formally asked NVidia to give us more and they're working on it. One thing we know is that it doesn't work if there are two GPUs in the machine.

you can disable the feature by setting the "CUDA system fallback policy" to "Prefer No Sysmem fallback" in the NVidia control panel. Once you have done that then windows and linux should behave the same.

brians
On Windows, Microsoft/NVidia have implemented a way for Optix GPU memory to "spill over" to main CPU memory.
https://nvidia.custhelp.com/app/answers/detail/a_id/5490/~/system-memory-fallback-for-stable-diffusion [nvidia.custhelp.com]

There is not a lot of documentation or info about this feature, but we have formally asked NVidia to give us more and they're working on it. One thing we know is that it doesn't work if there are two GPUs in the machine.

you can disable the feature by setting the "CUDA system fallback policy" to "Prefer No Sysmem fallback" in the NVidia control panel. Once you have done that then windows and linux should behave the same.

well i just test a large scene with 2 gpu and karma xpu in windows working perfectly , both gpu doesn't failing
but in ubuntu gpu it failed easly again.

soo i'm guessing thats this feature work also with 2 gpu? if not then I have no idea why its happening (and i dont have problem with this situation because its not failing and the rendering is faster but i want to know out of curiosity why its happening).
by the way with this feature even if the driver send data to the CPU memory the rendering in xpu in windows its always faster then in linux when the gpu failing in linux.