I have an interest in doing lattice Boltzmann fluid dynamics simulation on GPUs, and this is something I've noticed with the P100's as well. That and memory ba

I have an interest in doing lattice Boltzmann fluid dynamics simulation on GPUs, and this is something I've noticed with the P100's as well. That and memory bandwidth is super important for such calcs, along with total memory. Commercial codes I've seen aren't using any ML ops compute optimizations or tensor cores and anything like that, it's mostly just OpenCL based and thus the P100s throughput per dollar is pretty high.