Using Adaptive Sampling and Optix

[Miss B]

I understand Cath, and the RTX and newer GTX cards for desktop computers have more abilities than in a laptop, such as mine.

I'm just very happy with the difference I'm seeing from what I've seen before when rendering.

 

[Ken1171]

OptiX don't render nor it is an rendering engine, it allows Cycles engine to access Cuda cores faster on RTX cards for 60% and more speed up rendering time . You can still use Optix rendering with GTX cards . OptiX can't send all shader nodes to the cuda cores so there are some limitations.
You always render with CUDA cores no matter Optix or not .

Oh Ok, thanks for the info. I missed the train on OptiX because when it first came out, I didn't have an RTX card, so I didn't look into it. There were claims it was possible to render with OptiX without RTX, but when I tried, it took 2X longer emulated in software - totally not worth it. Don't know if that had improved later, but with a GTX 980 Ti, it was impractical.

Still on this subject, the upcoming Cycles X is a redesigned Cycles engine optimized and streamlined for performance. The benchmarks indicate an 8-10 times rendering speed improvement. The article claims they rewrote Cycles from scratch to take advantage of modern hardware, and get rid of the limitations imposed by the original design. Bondware has officially announced CyclesX will be added to Poser, which is rather exciting news.

 

[Miss B]

That it is Ken. As a long time Blender user, I've wanted it's Cycles goodies in Poser for a long time.

 

[eclark1894]

I haven't used the Adaptive Sampling feature of Poser yet. Could someone please explain it to me?

 

[MEC4D]

Oh Ok, thanks for the info. I missed the train on OptiX because when it first came out, I didn't have an RTX card, so I didn't look into it. There were claims it was possible to render with OptiX without RTX, but when I tried, it took 2X longer emulated in software - totally not worth it. Don't know if that had improved later, but with a GTX 980 Ti, it was impractical.

Still on this subject, the upcoming Cycles X is a redesigned Cycles engine optimized and streamlined for performance. The benchmarks indicate an 8-10 times rendering speed improvement. The article claims they rewrote Cycles from scratch to take advantage of modern hardware, and get rid of the limitations imposed by the original design. Bondware has officially announced CyclesX will be added to Poser, which is rather exciting news.

Yes very exciting news Ken.
The best GTX for Cycles is GTX 1080 TI , it render as fast as RTX 2060 Super and RTX 2070 , it is faster than RTX 2060 , of course the best is Titan RTX and RTX 2080 TI , they performed almost the same on GPU level and Optix with Cycles .
The worse 4 speed on GPU- GTX 970 GTX 980 and GTX 1060 and GTX 1650 , all other RTX performing almost the same with Optix and GPU, I mean same level with OptiX and Same level when using GPU ..

so 3 hours rendering with GTX970 will be 20 min with Titan RTX using OptiX , my system I am using render little faster than Titan RTX on GPU around 23 min

 

[Ken1171]

I currently have an RTX 3090, which was a giant step up from the GTX 980 Ti. Like light years in performance boost. I bought it only for rendering, and gosh, it delivers! I haven't compared render speeds, since that becomes pointless when this card outclasses everything else. People who have purchased it for gaming are just out of their minds, and have been definitely ripped off. This card is THE absolute waste of money UNLESS you use it for 3D rendering. Since I do that for living, it was worth the pretty penny to me. 20 minutes renders have now been cut down to 15 seconds.

Now, if switching from raw CUDA to OptiX accelerated CUDA can speed up things so much - especially when combined with the new adaptive sampling, I wonder how much faster Cycles X would be in comparison? Looking at the benchmarks, it outclasses Cycles by a margin of 8-10 times faster, but I don't know if that comparison included OptiX or not.

I have the impression that Cycles X is coming to eliminate the need for OptiX for things to render faster. I guesstimate that it already optimizes things on its own to render faster, especially when they claimed it was designed to eliminate the original Cycles limitations that bogged performance down.

Considering unimesh should speed things up in Poser (when it comes out), faster rendering with Cycles X would make Poser a formidable program to work with. Most likely, the early unimesh Poser versions will be buggy, and that is normal and expected, due to how deep the internal changes will be. Bondware has been pushing Poser to the right directions, which hasn't happened in a very long time when it was owned by other companies. It has been a year now, and I still believe Bondware was the best thing to happen to Poser. It has been so long since I was excited about Poser, and this comes at a good time, when Dawn 2 is about to be released. ^_____^

 

[Ken1171]

I haven't used the Adaptive Sampling feature of Poser yet. Could someone please explain it to me?

"Adaptive sampling" (AS) works like the name implies. After a number of render iterations, AS looks at the image, and measures how much noise is left, and where. It then concentrates the next rendering iterations on the areas that need more work, instead of spreading it across the entire image. So it is a more intelligent approach, targeting ray casting processing to areas that need it more, and hence speeding up the render by periodically inspecting the current image noise to know what to do next.

As a result, you will notice rendering progress works differently when AS is enabled. No matter how high the pixel samplers you enter, AS will stop the render when the noise is below your desired threshold. You determine that by the only parameter AS requires, where the default is zero. The higher the value, the more tolerant AS becomes to noise in the final render, and also the shorter the render times.

There is a limit to how much noise the human eye can perceive in renders. I can't recall the average number, but I think it was something around 11% (from the back of my head). The bottom line is that we cannot see noise below that threshold, so the AS parameter should be set according to that. But mind you, noise in your render comes from many different factors, like for example, how well lit the scene is, or how many materials have things that need raytracing, like SSS, transparency, shadows over transparency, caustics, translucency, and refractions.

However, please keep in mind that AS cannot do magic. Each scene will require more or less pixel samples to clear the noise. AS just makes it quicker to do that, assuming we gave it at least enough pixel samples to begin with. The render will stop depending on which condition was reached first - the pixel samples limit was reached, or the AS noise threshold was reached - which ever happens first. The worst that can happen is when the pixel samples limit was reached first, which might leave more noise in the image than we had wished for. So a general recommendation is to aim higher with pixel samples when using AS. It will automatically end the render when the noise threshold is reached, which might skip much of the excessive pixel samples we had requested.

Hope this helps.

 

[eclark1894]

"Adaptive sampling" (AS) works like the name implies. After a number of render iterations, AS looks at the image, and measures how much noise is left, and where. It then concentrates the next rendering iterations on the areas that need more work, instead of spreading it across the entire image. So it is a more intelligent approach, targeting ray casting processing to areas that need it more, and hence speeding up the render by periodically inspecting the current image noise to know what to do next.

As a result, you will notice rendering progress works differently when AS is enabled. No matter how high the pixel samplers you enter, AS will stop the render when the noise is below your desired threshold. You determine that by the only parameter AS requires, where the default is zero. The higher the value, the more tolerant AS becomes to noise in the final render, and also the shorter the render times.

There is a limit to how much noise the human eye can perceive in renders. I can't recall the average number, but I think it was something around 11% (from the back of my head). The bottom line is that we cannot see noise below that threshold, so the AS parameter should be set according to that. But mind you, noise in your render comes from many different factors, like for example, how well lit the scene is, or how many materials have things that need raytracing, like SSS, transparency, shadows over transparency, caustics, translucency, and refractions.

However, please keep in mind that AS cannot do magic. Each scene will require more or less pixel samples to clear the noise. AS just makes it quicker to do that, assuming we gave it at least enough pixel samples to begin with. The render will stop depending on which condition was reached first - the pixel samples limit was reached, or the AS noise threshold was reached - which ever happens first. The worst that can happen is when the pixel samples limit was reached first, which might leave more noise in the image than we had wished for. So a general recommendation is to aim higher with pixel samples when using AS. It will automatically end the render when the noise threshold is reached, which might skip much of the excessive pixel samples we had requested.

Hope this helps.

Helps a lot.