Scale up the number of processors while not scaling up the problem size. It’s intended to test the latency (how fast the system can process the same problem).
Scale up the number of processors and scale up the problem size. It’s intended to test the throughput (how much more problems can the system process).