Blender benchmarks, new feature needed ?

The results in the thread ‘2-61-Cycles-render-benchmark’ are not easily exploitable, I think it’s time to have a feature in Blender that can help benchmarking.

Description :
Create a simple feature in blender that can produce a « log » file that store datas of the benchmark .
The log file can be send (to blender.org?) .

The data is then aggregated and the results can be seen through a sheet.

Users :

  • main developpers of Blender. It can help to have feedback of new Blender revisions across many different configs,
  • Blender users, to improve the benchmark results and to see if their config is OK or how to improve it.

Inputs :

  • hardware config : CPU(s), main RAM, GPU(s)
  • software : OS revision used, Blender revision,
  • blend file used (with revision). The blender developpers should select what scene(s) can be interesting to benchmark.
  • all the render, sampling, light passes and performance options used.

Treatment :

  • store time datas in memory when the user render a scene,
  • produce a XML file and on screen.

Output :

  • a XML file containing all the datas, with a checksum. It will contain all input datas and differents time of treatment: total time, time for synchronizing objects, building BVH, …
  • the results in a screen in Blender, with an option to send it to the net.

Not a bad idea. I suspect Blenderheads would desire clear opt-in dialogs. I’d make a Blender Development Fund donation to that potential feature.

i sense endi’s response won’t be too long now…

If I am not mistaken, it was already here and now its gone again.

He said, “yes we need more features” or something like that.

Wouldn’t be too hard (for someone else) to knock that out as a python addon, might require some rna magic to get all the wanted stats though.

Uh, yeah, he always says something like that which is what makes him so annoying.

Perhaps the easiest thing would be a script that runs through the Blender regression suite with a timer.