Tweet
I’d like to request independent Manual bucket size settings for CPU and GPU when using CUDA.
We have a lot of render nodes with Fast CPUs, but legacy GPUs. The default giant bucket size for GPU (and also too large bucket size for CUDA CPUs) is just way too big. These machines end up waiting for the GPU blocks because the CPUs are actually faster than the GPUs. Sure, we can disable the GPUs on those machines, but that is a waste. What we really need is the ability to set independent block sizes for GPU and CPU. The user knows best here what block sizes are going to work for their setup.
Ideally there would be an environment variable you could set on each machine if you wanted. This would be a fixed block size for CUDA CPU, one for CUDA GPU, and one for VRay CPU (important these are independent). In a perfect world this could be a multiplier to take the block size in the scene file and multiply it by, say 0.25 for a particular machine. This way we could tune relative block sizes per render mode.
Users want control over more aspects of VRay, not fewer. Removing settings, or not adding settings like bucket size for the GPU, for the sake of “Decluttering the UI” is never a good thing.
Thanks.
We have a lot of render nodes with Fast CPUs, but legacy GPUs. The default giant bucket size for GPU (and also too large bucket size for CUDA CPUs) is just way too big. These machines end up waiting for the GPU blocks because the CPUs are actually faster than the GPUs. Sure, we can disable the GPUs on those machines, but that is a waste. What we really need is the ability to set independent block sizes for GPU and CPU. The user knows best here what block sizes are going to work for their setup.
Ideally there would be an environment variable you could set on each machine if you wanted. This would be a fixed block size for CUDA CPU, one for CUDA GPU, and one for VRay CPU (important these are independent). In a perfect world this could be a multiplier to take the block size in the scene file and multiply it by, say 0.25 for a particular machine. This way we could tune relative block sizes per render mode.
Users want control over more aspects of VRay, not fewer. Removing settings, or not adding settings like bucket size for the GPU, for the sake of “Decluttering the UI” is never a good thing.
Thanks.
Comment