ggraphitti's current random number generation system consists of a series of C function calls, with RandomNormGPU as the kernel that is repeatedly invoked after the initial setup. This approach generates only a small number of random numbers per call, resulting in low GPU occupancy and frequent kernel launches. My plan is to introduce a dedicated class that generates random numbers asynchronously and in larger batches, improving GPU utilization while also enhancing encapsulation and modularity.
ggraphitti's current random number generation system consists of a series of C function calls, with RandomNormGPU as the kernel that is repeatedly invoked after the initial setup. This approach generates only a small number of random numbers per call, resulting in low GPU occupancy and frequent kernel launches. My plan is to introduce a dedicated class that generates random numbers asynchronously and in larger batches, improving GPU utilization while also enhancing encapsulation and modularity.