Why not?

Because it’s good on benchmarks but not on real usage?

But OP said they've never used it. How would they know?