LuxCoreRender Forums

Dade wrote: ↑Fri Jan 18, 2019 4:09 pm You mean PATHCPU, right ? PATHOCL is unchanged for the moment.

PathCPU of course, sorry.

epilectrolytics wrote: ↑Fri Jan 18, 2019 4:18 pm

Dade wrote: ↑Fri Jan 18, 2019 4:09 pm You mean PATHCPU, right ? PATHOCL is unchanged for the moment.

PathCPU of course, sorry.

BTW, the slowdown is usually the effect of having a very large cache search radius and having many thousands of photon falling inside the search radius. Try to change the radius and you will see how the samples/sec number goes up and down.

I may have to switch from a BVH look up accelerator structure to a Kd-tree in order to solve this problem but it is something I will look into when things are a bit more settled down.

Haha, it works, this is a rendering with 15 samples/pixel of LuxCore Benchmark and cache:


And this without cache:


Notice the huge difference of noise in indirect lighting (like under the table, etc.). And with only about 6 seconds of pre-processing:


I have to work on the samples/sec performance but PhotonGI is going to reduce the rendering times by a lot.

GOING to be more than we expected .coding Sensei Mode ON

Dade wrote: ↑Fri Jan 18, 2019 4:51 pm I have to work on the samples/sec performance but PhotonGI is going to reduce the rendering times by a lot.

I think the latest estimate I did way back using corona was about 8x speedup using PT+UHDC vs PT+PT.
Funny thing is that although caching is mostly recommended for interiors it still helps a lot even with exterior renderings and basicaly everything that includes shadow areas...
So unless its a super simple scenario, everything should benefit from using caching, that's why it's on by default for vray and corona aswell.

Dade wrote: ↑Fri Jan 18, 2019 4:51 pm I have to work on the samples/sec performance but PhotonGI is going to reduce the rendering times by a lot.

I have moved some BSDF evaluation step from the rendering phase to the pre-processing phase and now the rendering is about 4-5 faster (!) at the cost of a slight increase in pre-processing time:


If you check the previous post, the samples/sec were 0.17M samples/sec and are now 0.68M samples/sec (so it is 4 times faster) at the cost of an increase of pre-processing time from 6secs to 20secs.

The rendering with cache enabled is also now faster than without cache (0.68M samples/sec Vs 0.50M samples/sec). It is faster in term of samples/sec and it has a lot less noise. The result with 150 samples per pixel is quite good in term of noise. This thing is gone to be quite fast when you throw the GPUs in the mix

Would it be possible to get a preview of the cache while it is building?
To spot stuff like the artifacts on the doorframe quickly.

B.Y.O.B. wrote: ↑Sat Jan 19, 2019 1:41 pm Would it be possible to get a preview of the cache while it is building?
To spot stuff like the artifacts on the doorframe quickly.

Yup, this: viewtopic.php?f=5&t=840&start=30#p8884

I wonder why that doorframe even has those artifacts when other detailed geometry seems to look clean.
Fantastic progress btw.

Dade wrote: ↑Fri Jan 18, 2019 4:51 pm Haha, it works, this is a rendering with 15 samples/pixel of LuxCore Benchmark and cache:

Amazing!

I thought there is a poor rate of photons registered from sunlight but fortunately it seems I was wrong
And yet still without metropolis photon gathering, right?

Would this work with an HDRI as well?

Great progress, keep it up!