LuxCoreRender Forums

I was referring to the clamping value in general, having a clamping value per light group it's even better

B.Y.O.B. wrote: Tue Mar 31, 2020 2:21 pm In the hall bench scene, the VRAM usage is:
- without cache: 1029 MiB
- with cache: 1031 MiB

Are you sure ? I mean, if I can ignore the "additional memory problem" I can easily add the support for light groups (storing 8 x 3 floats values instead of 3 float, max. light group count x RGB instead of 1 x RGB).

The question is: the memory usage will jump from 2MB to 16MB (and no one care) or it will jump from 100MB to 800MB and people will start to load their guns

The question is: the memory usage will jump from 2MB to 16MB (and no one care) or it will jump from 100MB to 800MB and people will start to load their guns

We just need a warning Message on it. Being a special features it is acceptable to have such sacrifice to do. multiple render in one this is awesome.

Sharlybg wrote: Fri Apr 03, 2020 1:24 pm

The question is: the memory usage will jump from 2MB to 16MB (and no one care) or it will jump from 100MB to 800MB and people will start to load their guns

We just need a warning Message on it. Being a special features it is acceptable to have such sacrifice to do. multiple render in one this is awesome.

No, it is not optional, even if you use a single light group, it will use 16MB instead of 2MB (i.e. no problem) and 800MB instead of 100MB (i.e. big problem).

No, it is not optional, even if you use a single light group, it will use 16MB instead of 2MB (i.e. no problem) and 800MB instead of 100MB (i.e. big problem).

understood (This is Crazy )

So maybe boosted Path/Bidir will make the trick

Dade wrote: Fri Apr 03, 2020 1:13 pm

B.Y.O.B. wrote: Tue Mar 31, 2020 2:21 pm In the hall bench scene, the VRAM usage is:
- without cache: 1029 MiB
- with cache: 1031 MiB

Are you sure ? I mean, if I can ignore the "additional memory problem" I can easily add the support for light groups (storing 8 x 3 floats values instead of 3 float, max. light group count x RGB instead of 1 x RGB).

The question is: the memory usage will jump from 2MB to 16MB (and no one care) or it will jump from 100MB to 800MB and people will start to load their guns

Some tests, VRAM usage of PhotonGI indirect light cache (no caustics cache), as reported by LuxCore:

Danish Mood:
- without cache: 577 MiB
- with cache: 580 MiB
Diff 3 MiB

LuxCore 2.1 Benchmark:
- without cache: 479 MiB
- with cache: 486 MiB
Diff: 7 MiB

LuxBall scene from https://github.com/LuxCoreRender/LuxCor ... s/tag/v2.3:
- without cache: 269 MiB
- with cache: 296 MiB
Diff: 27 MiB

DLSC testscene:
- without cache: 374 MiB
- with cache: 376 MiB
Diff: 2 MiB

If these numbers look fishy to you, maybe LuxCore is not counting all VRAM usage of PhotonGI? However, the VRAM usage I see in the task manager looking mostly the same)

Dade wrote: Fri Apr 03, 2020 1:28 pm No, it is not optional, even if you use a single light group, it will use 16MB instead of 2MB (i.e. no problem) and 800MB instead of 100MB (i.e. big problem).

The memory can't be allocated only on demand?

B.Y.O.B. wrote: Fri Apr 03, 2020 2:40 pm If these numbers look fishy to you, maybe LuxCore is not counting all VRAM usage of PhotonGI?

No, if you are reading the numbers from the log, they are correct as far as I know.

Dade wrote: Fri Apr 03, 2020 1:28 pm The memory can't be allocated only on demand?

It is but is more complex (mostly in OpenCL).

I added the support for light groups to PhotonGI (both indirect and caustic cache)

Dade wrote: Sat Apr 04, 2020 11:41 am I added the support for light groups to PhotonGI (both indirect and caustic cache)

That was fast. We still need blendluxcore add to test i guess.

Sharlybg wrote: Sat Apr 04, 2020 12:34 pm

Dade wrote: Sat Apr 04, 2020 11:41 am I added the support for light groups to PhotonGI (both indirect and caustic cache)

That was fast. We still need blendluxcore add to test i guess.

Nope, it should work out of the box.