Parallel samples default settings is way too high Tweet

[J.C]

I don't know why Parallel samples default settings is set to highest setting – 32?

I just did quick comparison in my sample scene:

- parallel samples set 4 = render time 37s
- parallel samples set 8 = render time 34s
- parallel samples set 16 = render time 33s
- parallel samples set 32 = render time 33s

Memory usage was 450MB in lowest and 1800MB in highest setting.


In other scene with one 4,5 mln triangles object:

- parallel samples set 4 = render time 11 s, memory usage 900 MB
- parallel samples set 32 = render time 9 s, memory usage 2800 MB

[frankmci]

TOOHIGH.jpg

I tried to resist, but I just couldn't help myself.

Seriously, though, I think it defaults to 16 Parallel and 32 Tile Samples, right in the middle of the range.

[J.C]

TOOHIGH.jpg

I tried to resist, but I just couldn't help myself.

Seriously, though, I think it defaults to 16 Parallel and 32 Tile Samples, right in the middle of the range.

Still too high. 4-8 is optimal.

[Lewis]

TOOHIGH.jpg

I tried to resist, but I just couldn't help myself.

Seriously, though, I think it defaults to 16 Parallel and 32 Tile Samples, right in the middle of the range.

Still too high. 4-8 is optimal.

Optimal for what ? How can they know how much VRAM users have ?

I tested it on much heavier scene and speedup was very noticeable, your scene is too small/fast to render to see much of benefit but speed difference is pretty decent when you have larger scene.
Also you can set you default/save preset as you wish so no harm done . Mine is set even higher than default .

you can check my examples here:
viewtopic.php?f=23&t=70008

cheers

[frankmci]

TOOHIGH.jpg

I tried to resist, but I just couldn't help myself.

Seriously, though, I think it defaults to 16 Parallel and 32 Tile Samples, right in the middle of the range.

Still too high. 4-8 is optimal.

Any setting that leaves VRAM sitting unused is less than optimal. If you've got eight or ten gigs of VRAM gathering dust instead of contributing to the render, you are just wasting resources. Why not use it and, as in your own example data, get an extra 12% faster performance?

[J.C]

If your scene fits easily you can increase samples to speed up rendering without any issues.
If you are using Octane for the first time and creating an empty scene with 16 parallel samples takes almost 1GB RAM you would be confused and surprised. Having eg. 8GB RAM, leaves you with 5.5GB available for your scene which is not much as we know Octane is quite memory hungry. (1GB takes parallel samples, 1,5GB is reserved for Windows).
I my sample scene the time difference between 8 and 32 samples was 2 seconds difference in a scene with 8 mln polygons. (1,8 GB vs 3,2GB)