I think size ranges of `1:100` and `logspace(100, 10_000, N)` are good, where `N` is the number of sampled sizes. I can add benchmarks for Skylake-X, Haswell (probably not all the way up to 10_000, perhaps stopping at just 1_000), and Tiger Lake. Having example results would be nice for BLASBenchmarks, while Octavian should showcase its performance.