I realize that hardware questions are off-topic here, but I think this question about documentation and optimization is not really hardware dependent - it's a question of being able to reproduce consistent results with different versions and installations of Blender.
It took me a while searching the web and this site to realize why I was getting such poor results rendering the classroom benchmark scene on my new system that came with an NVIDIA GeForge GTX 1070. The default settings were set to render with the CPU, and when I tried to switch it to CUDA-based GPU rendering, things only got slower. Searching the web I found others had gotten it to render in about 7 minutes on the same hardware (GTX 1070). Finally I found the answer: increase the tile size. It seems like a lot of effort to figure out some basic configurations if you're not intimately familiar with how rendering works. And I still haven't matched the other benchmark. I'm at 8:38, and the other benchmark was at 7:47.
So my questions are:
- Is there some way to know to what I should attribute this variation? My benchmark was on Blender 2.78c and the original was on 2.78a. So it could be that or something else.
- Is there any documentation covering optimal settings for the benchmark scenes? I realize optimization is very dependent on hardware and scene, but especially for benchmark scenes on the most typical hardware, it would be useful to know how to reproduce consistent results so we can, for example, compare the performance of different versions of Blender.
Edit: The classroom scene I downloaded from https://www.blender.org/download/demo-files/, and I just realized that it's not technically labeled as a benchmark.