Investigate in the context of GPU friendly tree evaluation. In ACF processing we compute features in shaders and pull down the textures to the CPU w/ platform optimized transfers, then evaluate regression and classification trees on the CPU. It would be nice to investigate ways to do this end-to-end on the GPU (my sense is it will be hard to beat the CPU even w/ transfer overhead).
Investigate in the context of GPU friendly tree evaluation. In ACF processing we compute features in shaders and pull down the textures to the CPU w/ platform optimized transfers, then evaluate regression and classification trees on the CPU. It would be nice to investigate ways to do this end-to-end on the GPU (my sense is it will be hard to beat the CPU even w/ transfer overhead).