LuxCoreRender Forums

Martini wrote: ↑Tue Jun 02, 2020 9:56 am Edit 6: Decided to test it - switched from CUDA back to OpenCL and rendered a few frames. The difference is quite noticeable:

memory-cuda-vs-opencl.png

So, is it a CUDA-only problem ?

So, is it a CUDA-only problem ?

On my side i get this error solved by switching from Opencl to CUDA but with lastest unified build.

Sharlybg wrote: ↑Wed Jun 03, 2020 12:38 pm

So, is it a CUDA-only problem ?

On my side i get this error solved by switching from Opencl to CUDA but with lastest unified build.

So it is a OpenCL problem ?

So it is a OpenCL problem ?

i can probably say today that it is a opencl bug only now.

Sharlybg wrote: ↑Wed Jun 03, 2020 3:28 pm

So it is a OpenCL problem ?

i can probably say today that it is a opencl bug only now.

The problem only occurred for me with OpenCL in the unified build.

Ok, this restrict the list of places where to look for the problem.

Dade wrote: ↑Wed Jun 03, 2020 7:43 pm Ok, this restrict the list of places where to look for the problem.

Btw it is such a pleasure to see luxcore render 1000 frames one after one without complaining. I feel so much peace. Will sleep happier tjis night. I have already 1055/1111 frames done currently.
Thanks for this beauty

The problem is easy to replicate but it starts to smell like a NVIDIA driver bug. I have modified the OpenCL code so it should be pretty much impossible to have memory leak across different renderings but the problem is still there.

Is someone with an AMD GPU able to replicate the problem ?