I am not aware of that right now. Internally we measure the performance by looking at the theoretical peak possible (like the scaling efficiency when you use distributed training). C2 has been doing pretty well; if you find performance degradations, send an issue to us and we are usually very good to figure out the perf.