[mokus]

For traditional floating point workloads, the RTX 6000 will probably not be 3x faster. For workloads that can use the tensor ops (integer matrix multiply, basically), the RTX 6000 may be as much as 10-100x faster.