Rohit and I briefly looked into this before, but there were issues then. It may just be that we need to uprev the Aquarium sources first. Then we need to ensure not too many benchmarks change with the new revision (especially power). And finally select more fishes! That said it probably mostly measures browser/driver CPU overhead. And 10k fishes may not be enough either to not max out at 60fps.

That said what would be slightly more interesting in these kinds of benchmarks is to measure other quantities and correlate them with the main measurement, like CPU utilization, GPU frequency, temperature, power draw. (Even if we run at 60fps flat we may actually use different GPU frequenceis to achieve that.) Maybe it would be sufficient to plot all these quantities in the same graph in the dashboard?

Thinking about this some more, if we get accurate enough RAPL power graphs, then pegging the workload on some artificially high maximum isn't going to be so interesting anymore. So lets defer this until we have powerized all benchmarks.

 Issue 489352  has been merged into this issue.