Of course, server GPUs remain kings of the hill on FP64 workloads, as they have extra FP64 units. I focus on FP32 performance because it is representative of ge

Of course, server GPUs remain kings of the hill on FP64 workloads, as they have extra FP64 units. I focus on FP32 performance because it is representative of general compute performance on all other workloads, such as FP32, FP16, FP8, or integer.