A complex pmc situation

[radiance]

Caustics on glossy materials are actually very fast and good looking. But on refractive they are relatively slow and on pmc it needs >1000 samples until they are as bright as they should be. Also on pmc it looks like it darkens refractive materials a lot, most with a high exploration_strength or with a very low direct-light-importance.

Anyway...
Radiance said I should use the pathtracing/mlt instead of biDir/mlt in Lux to get a fair comparison. Now I did. Now also Lux has a hard time to find new paths trough the glass because no directlight can be traced, but if it finds new ones it really samples the ones found way better and refractive caustics appear faster. After 7 Hours, both noiselevels of Lux and Octane nearly where the same and in my opinion the result in terms of realism looked slightly better in Octane.

Ok but I have to say that I just have a gt250m GPU in relation to a Core I7 at 1,6Ghz, so my GPU is very low end.

Hey,

You're GPU is 9x - 10x slower than a GTX580

Radiance

[telemmaite]

Caustics on glossy materials are actually very fast and good looking. But on refractive they are relatively slow and on pmc it needs >1000 samples until they are as bright as they should be. Also on pmc it looks like it darkens refractive materials a lot, most with a high exploration_strength or with a very low direct-light-importance.

Anyway...
Radiance said I should use the pathtracing/mlt instead of biDir/mlt in Lux to get a fair comparison. Now I did. Now also Lux has a hard time to find new paths trough the glass because no directlight can be traced, but if it finds new ones it really samples the ones found way better and refractive caustics appear faster. After 7 Hours, both noiselevels of Lux and Octane nearly where the same and in my opinion the result in terms of realism looked slightly better in Octane.

Ok but I have to say that I just have a gt250m GPU in relation to a Core I7 at 1,6Ghz, so my GPU is very low end.

Hey,

You're GPU is 9x - 10x slower than a GTX580

Radiance

Radiance any update on the caustics. How do you plan to solve this problem?

[radiance]

Caustics on glossy materials are actually very fast and good looking. But on refractive they are relatively slow and on pmc it needs >1000 samples until they are as bright as they should be. Also on pmc it looks like it darkens refractive materials a lot, most with a high exploration_strength or with a very low direct-light-importance.

Anyway...
Radiance said I should use the pathtracing/mlt instead of biDir/mlt in Lux to get a fair comparison. Now I did. Now also Lux has a hard time to find new paths trough the glass because no directlight can be traced, but if it finds new ones it really samples the ones found way better and refractive caustics appear faster. After 7 Hours, both noiselevels of Lux and Octane nearly where the same and in my opinion the result in terms of realism looked slightly better in Octane.

Ok but I have to say that I just have a gt250m GPU in relation to a Core I7 at 1,6Ghz, so my GPU is very low end.

Hey,

You're GPU is 9x - 10x slower than a GTX580

Radiance

Radiance any update on the caustics. How do you plan to solve this problem?

We're probably going to add bidirectional pathtracing support in a few months,
or another solution. We're currently doing research to see what's the most feasible on a GPU.
This is scheduled for post v1.0 final though, we need to finish what we have now and release our first final version (instancing, SSS and plugins)
We can't keep octane in beta forever...

Radiance

[pixie]

To make it a fair test you should use the same alghorithms,
if you use bidir/MLT in luxrender you will get efficient rendering of caustics due to the lightpaths.
PMC/pathtracing (what octane does with 'PMC' enabled) should be compared with MLT/pathtracing in any other engine for a fair comparison on this particular type of scene,
which is about the worst you can get for a pathtracer (not for a bidirectional pathtracer)

Radiance

Then if you want to be fair then add the bidrectional to GPU if possible, otherwise people might not care much on the level of fairness, rather the time available to get the work done, and it may even be that a CPU using clever algorithms be faster then a GPU, against all odds...

[Daniel79]

Great steps have been made by you...
when i ask for the first time about MLT has been a long time (May 2010)..and today is not yet clear how it will end.
Octane Render is a fantastic program and you are great but MLT or similar is very very important (for me), and a solution should be found as soon as possible, especially after so long that talk about it and after so many predictions of release...

Sorry for my bad english, good work...

Great

[telemmaite]

Great steps have been made by you...
when i ask for the first time about MLT has been a long time (May 2010)..and today is not yet clear how it will end.
Octane Render is a fantastic program and you are great but MLT or similar is very very important (for me), and a solution should be found as soon as possible, especially after so long that talk about it and after so many predictions of release...

Sorry for my bad english, good work...

Great

I think you misunderstood something. MLT (Metropolis light transport) is a variant of the Monte Carlo method. Octane has already very similar method which is called PMC (Population Monte Carlo) All this methods has to do with estimating and converging the correct sample radiance. The problem is that Octane shoots the samples only from the camera which is called Path-tracing . Implementing bi-dir will mean that the samples will be shot from camera and lightsource so the caustics will be a lot better and correct. If you are more interested about the PMC algorithm and whats even better you might want to check this link : http://www.luxrender.net/forum/viewtopi ... =4116&f=13

[radiance]

Great steps have been made by you...
when i ask for the first time about MLT has been a long time (May 2010)..and today is not yet clear how it will end.
Octane Render is a fantastic program and you are great but MLT or similar is very very important (for me), and a solution should be found as soon as possible, especially after so long that talk about it and after so many predictions of release...

Sorry for my bad english, good work...

Great

Hi, PMC is MLT. It is a cousing of MLT, and is better imo.
We already support it for a while now, you can use it as a kernel in octane 2.4+
It gives superior results than MLT, and is more GPU friendly.
To match other unbiased engines (that have been in development for 3-4x longer), and do complex indirectly lit interiors as efficiently,
we will need bidirectional path tracing aswell. which we might add in a few months, but as said we are investigating alternatives.

A properly configured scene, with portals, decent ligt setups and our PMC kernel will beat everything else out there, provided you have a decent GPU,
such as one or more GTX580.
A lot of tests compare octane with other engines where an expensive 6 core xeon CPU is pitted against a cheap laptop-grade GPU, no wonder it's slow...
Ask our customers with 2 xeon CPUs and 2-3 GTX580 (eg, the same cost of GPUs vs CPUs in purchase price) the same question

Even if PMC/pathtracing is 3x slower than MLT/bidir for these types of scenes, having 2 GTX5XX gpus in your machine will nevertheless make octane win by sheer computing power alone.

Radiance