Piotr Rybicki

@paul2t, @asny Hi there, do you plan to make it into main? Winit 0.30 api looks much more convenient and I'd love to have it in three-d before writing a...

> [22:14:18][22478887 us][critical]: Unrecoverable error (code=500): cuda error: operation not supported (code=801) @ /home/nimda/RobotecGPULidar/test/../src/memory/MemoryOperations.hpp:85 This points to: > CHECK_CUDA(cudaMallocAsync(&ptr, bytes, stream->getHandle())); Which used so-called [stream-ordered memory allocation](https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__MEMORY__POOLS.html), which has been...

Yes, this sounds possible, altough we never tested. Here's how to do it: In [MemoryOperations.hpp](https://github.com/RobotecAI/RobotecGPULidar/blob/develop/src/memory/MemoryOperations.hpp#L79) there's a function that returns a set of methods that manage given kind of memory...

This CUDA setup seems really broken. Have you tried running CUDA samples? The error says that it was impossible to get the primary CUDA context (which is a GPU counterpart...

Also, sharing output from `nvidia-smi` is a good idea.

@duongtd23 > Searching around the error code 801, it seems that we need to enable the so-called "Unified Memory" for virtual GPU from the host machine. Could you link the...