With GPU rendering, what is a good, reasonable setting to use for Samples limit and noise? I know that it “all depends” on the scene, lighting, output desired, etc…
With V-Ray 5, a friend of mine told me to “set the Sample limit up to 50,000” and the Noise limit to .003. That actually worked pretty well, but with V-Ray 6, it seems like it’s taking forever…
Back in the CPU days, I used to use a Noise limit of .01 for almost everything… What setting have you found work well for interiors with a dome-light?
CPU or GPU makes no difference here.
To know what it is you are doing, you will need to add a sampleRate RE to your renders.
Optionally, a noiseLevel RE may provide for more info.
The sampleRate RE is the thing that will let you know if you are hitting the end of the sampling V-Ray is allowed to bring to bear (i.e. max AA/samples), thereby leaving areas below the noise threshold you set (and those ought to be visible/detectable in the noiselevel RE), or if the noise threhsold is being achieved without using up all the sampling allowed.
Within a given max samples limit, simply moving up and down noise threshold will globally speed up/clean up the rendering, unless ofc you’re asking V-Ray to achieve too low a noise level, with too few max samples.
In addition to what my colleagues suggested, we have changed the default sampling settings in V-Ray 6 GPU. The new defaults should work well with most scenes, it uses 2500 Samples Limit and .01 noise threshold
There is no need to set samples limit to 50,000 , if you want cleaner results first try using a lower noise threshold like 0.005 (and samples limit of 2500) then you can double samples limit to 5000 if you have noise in GI/Refractions/Reflections..etc
We recommend using Progressive sampler type for V-Ray GPU, it is better than Bucket mode in performance and memory usage. And remove any render elements that you don’t need, this will speed up your rendering as well
If it is still slow on your side, upload me a screenshot of your beauty element + Sample Rate. I can take a look at your scene as well, if possible
The answer is yes to both questions
For animation sequences you will be able to use a time limit per frame. This is common practice in rendering animations
On my 3090 I set it to like 1-2 minutes depending on the scene and resolution
I would assume that you must use Denoiser then for the animations? Have you noticed any flicker or inconsistencies if you apply a hard “stop” of say, 2 minutes to each frame?
Currently, I am trying a Time limit of 6.0 minutes, .01 noise and 2500 Samples limit, but the renders still take 7.5+/- minutes to complete. Not sure why the 6 min stop is being ignored.
In general, the sampler will stop trying to lower noise until the time limit is reached.
However, if it stopped suddenly, noise level across the frame wouldn’t be uniform, so V-Ray (CPU too.) will exceed the time limit to bring all the remaining pixels under the N.T. defined at the stoppage time.
Add a sampleRate RE and stare at it while rendering is going: you’ll get a clearer idea of what V-Ray is doing, and why the time limit isn’t respected exactly.
edit: notice that setting a time limit for an animation is dangerous, as each frame noise level will be different (denoising will also, as a consequence, produce subtly different results.).
How different is of course dependent on what is being rendered (f.e. all of a sudden a divergent task comes into play, like SSS, and your performance tanks compared to the previous frames.), but conceptually, that’s what is going to happen: don’t use it for finals, unless you are 100% sure you know what you’re doing, and are fine with the varying noise amount per frame.
I always thought that bucket was better for memory usage, since it divides the scene into buckets instead of having to throw samples at the scene at once.
What is the status for optimizing vray gpu with many instances/heavy forest pack scenes ? You once said it is in active development or being looked at by the vray gpu team
Progressive is faster than bucket mode in the current builds of V-Ray GPU (up to 2x faster), and better on memory usage. swapping to progressive will save you at least 3 GB of GPU memory, it is part of our planning now to polish bucket mode and make it faster.
About Forest Pack and V-Ray GPU, this affects time to first pixel. It is near the top of our current priorities on the GPU roadmap. I hope to share some news on this soon
It depends on the scene of course, is there a chance I can take a look at your setup ?
I use 3090s as well, so on high end GPUs targeting one or 2 minutes per frame has always been simple for what I did in print. And yeah, I use V-Ray’s denoiser
What resolution are you rendering at? 7 minutes per frame on 2x 3090s seems slow
ok: what I find confusing is why GPU Samples limit value now (since the default become 2500) means “max samples per pixel *4”. what’s with the *4? previously Samples limit of 5000 corresponded to 71 max subdivisions (samples are subdivs squared) but now 5000 gives 35 max subdivs (when changed to cpu engine). what am I missing here?
Samples Limit is Max Subdivs, default value now is 2500 which is 48 Max Subdivs
Max Samples per pixel, is rays per pixel in the performance tab. it translates to min subdivs internally which is default at 8, you will not need to change that (it is not linked to the standard CPU renderer)
The GPU UI does all the conversions between V-Ray and V-Ray GPU, I can give you the conversion formulas, I don’t think it is necessary
The goal is having one slider to control sampling/quality in V-Ray GPU which is Sample Limit, everything else stays at defaults
GPU Samples Limit 2500 seems to be 25 subdivisions when you change the engine to cpu: sqr(2500/4). if just the default value of max samples per pixel (described in subdivisions) changed it would be Samples Limit of 576 not 2500. The default value and what the number means changed at the same time - question is why?
Just updated to Vray 6, are these changes to the GPU render settings reflected in the docs? I can’t find them if so.
Bit confused about Samples Limit and Rays per Pixel, I’m assuming samples limit has just replaced ‘Max Samples’ in Vray 5 but the numbers have shifted. For GPU I used to set it to max samples of 100 and just control quality with the noise threshold. @Muhammed_Hamed I see you’ve mentioned 2500 should be around the same as 48 max samples? is this still high enough for most GPU renders as that used to get me noise with glass materials.
And would this ‘samples limit’ now be comparable to number of samples in Octane render? I was always a bit confused about the difference in terminology between Vray GPU and Octane
V-Ray GPU will use double the Max Subdivs of what the standard V-Ray uses, in our testing this produces very close noise results
So 24 Max Subdivs on CPU, translates to 48 Max Subdivs on GPU which is 2500 Samples Limit
Yes to both questions
2500 Samples Limit is 48 Max Subdivs in V-Ray 5 for Maya. It should work well for most cases
You can go to 5000 Samples which is around 70 Max Subdivs if you need cleaner results
The goal of this change in the GPU UI is ease of use, in most cases you will use noise threshold to control quality/noise and leave samples limit to 2500 or 5000
On another note, if you have a scene setup for the standard V-Ray (CPU) your sampling settings will get translated automatically to V-Ray GPU. This helps people testing GPU rendering on their existing scenes, the noise results will be close between V-Ray and V-Ray GPU
