Benchmarks are tricky and depends a lot on workload, data type and hardware. So we've refrained from doing that.