If you are all getting lower CPU utilization with slower CPUs, then why does the GPU client take 1 core for itself?