More and more GPUs come with a lot of RAM, but the max allocation size tends to be quite limited.
The idea would be to implement a multi-buffer generation in the OpenCL kernel to use the full GPUs RAM capacity.
The hard part is syncing between multiple buffers with possibly different sizes.
Or would it be better to instanciate multiple kernels at the same time ?
More and more GPUs come with a lot of RAM, but the max allocation size tends to be quite limited. The idea would be to implement a multi-buffer generation in the OpenCL kernel to use the full GPUs RAM capacity.
The hard part is syncing between multiple buffers with possibly different sizes. Or would it be better to instanciate multiple kernels at the same time ?