I think it would be nice to have improved benchmarking:
Real simulation - benchmarking shoudl work in the same environment as real processing - build temporary source on pre-configured location and save temporary output to location, configured in preferences. “Pure performance” is often far from reality without this condition as I/O subsystem mostly is out of measurement.
Make it “all at once” - benchmarking is very rare action to perform. Why not perform full measuring cycle through SD->HD->FHD->QHD testing at once? Result may be displayed as table for improved readability. This way duplication of common frases (like “fps” can be omitted). Also using of fixed-width values can help with readability (for example, fps can be expresed as “12.3”, “12.3” or “123” depend on values).
Results export - add possibility to export benchmark results as CSV-file. So it can be used for performance reporting, bug reporting and just for comparison between configuration or in case of harware cofiguration changed on the same station.
All these changes are mostly “cosmetik” and not require too much resources for realization
I asked about RAM frequency and they told me it’s surprisingly hard to get that information programmatically. I think they intend to include that, but it’s not top priority since it did not turn out to be an easy thing to implement.
I disagree about real world numbers. With the amount of output options available, the results would become pretty useless, pretty fast. I always output to PNG. I think most users output to ProRes or NVENC H265. Just using those three examples, I know there are speed differences in them on my machine. If I see a benchmark outputted to ProRes, It’s going to mean as much to me as the numbers we’re getting now, since I only care about PNG speeds. It would mean even less if it’s ran on an old HDD or a cheap new SDD that maxes out at ~20MBs simultaneous read and write speed. All in all, I think they thought through all the situations and implemented the best approach for everyone.
