How to setup GPU to compare to normal render

[Crayox13]

Is there an easy way (other than setting noise level) to setup GPU render to be about the same in quality as a normal render? I use 1/48, and noise 0.005 in production, so I would want to setup hybrid render to give me the same quality and see how fast it goes.
Thanks

[hermit.crab]

For a reliable accuracy in the comparison, you should compare the images by noise threshold. Meaning measuring the time it takes for the different engines to reach a certain noise threshold. Increase the sampling amount and add a VRaySampleRate pass for the image quality's feedback. Note that the CPU and GPU engines work differently and there may be slight differences.

[Crayox13]

Hi,
sorry for the delay. By increase sampling amount, do you mean to increase "samples limit" from 5000?

[hermit.crab]

Yes, and the max. subdivs in V-Ray CPU respectively.

[squintnic]

Is there an easy way (other than setting noise level) to setup GPU render to be about the same in quality as a normal render? I use 1/48, and noise 0.005 in production, so I would want to setup hybrid render to give me the same quality and see how fast it goes.
Thanks

my god lol how do you make any money with a noise threshold so low - that must be insanely slow

[Julio_Cayetano]

Hey, I was trying to make this exact comparison between vRay GPU CUDA and vRay CPU, to be able to see how beneficial would be for us to buy some RTXs and change our current workflow:

We are in fact working almost 100% with Corona Render this year in our studio, and I was thinking of the possibility of changing either to vRay CPU (no hardware change) or even to vRay GPU (buying new RTXs with 24gb vRAM). We have prepared a typical interior scene to render almost as identically as possible in those 3 engines, so we can compare the render times we´d have in each case. We are using a RTX 2080ti (from a friend) and our old Titan (1st gener.), and a Threadripper for the CPU.
For the first test, I tried fixing the noise threshold at 0.02 in both VRay CPU and GPU, and the results were really similar, and the speed increase using the CUDA hybrid mode with both Geforce cards and the Threadripper is significant, but not that impressive I must admit. I think it was 56min (only CPU) vs 24min (2 GPUs+CPU). If I use only the RTX 2080ti it takes 45min.Nevertheless, adding some cards, specially the new 30 serie, the speed increase may be enough to convice us to gradually abandom Corona and adopt the GPU way (with the lack of features and stability, I know).
The problem I´ve found so far is that, when rendering with Corona, I fix its noise threshold to 2% (I believe it means the same as 0.02 in VRay), and the result looks way cleaner than the ones done with VRay GPU and CPU. So I´m not sure how to properly compare the speed of both engines, when their noise threshold doesn´t seem to stop the render at the same amount of noise. Anyone could help me wiht this? IS there a way to render the images in both engines to the same quality or amount of noise?

Thanks a lot in advance.

Best
Julio

[Karol.Osinski]

Hey Julio, thanks a lot of sharing your tests results with us, I keep a close eye on cpu vs gpu performance as our studio might too consider going GPU within one or two years.

Myself doing some testing every so often and unfortunately the gpu speed gains in our heavy archviz projects projects aren't significant or are none...I know it heavily depends on the hardware so to clarify, most of my testings are Threadripper 3970x vs x 2 x RTX SUPER and 32core cpu pretty much always faster.

In order to see the real speed gain I would have to open an almost empty scene with a sofa for example and render it with GPU...

I think the hype for GPU was more understandable before new, stable Threadrippers went out and they tipped the scale towards strong and stable cpu rendering again.

Although, with huge performance gains with new nvidia ampere the future seems bright for the gpu engine, but if Chaos Group won't improve scene loading times for heavier scenes and stability (YES, crashes are still expected) then most people will give up on GPU for good.

p.s I am sorry I could not answer your Corona related question but I believe Corona Render and Corona Render fan group on Facebook has many pro's that use both engines.

[Julio_Cayetano]

Thanks for your detailed answer. I also asked in Corona forums but received no answer, which feels strange I think, since I usually receive many fast answers there. Maybe this isn´t an interesting topic for most people, that may be just happy with their adopted software solution. I should be too, because Corona is a great engine, but I believe it is in my nature to be curious about the tools and technology I use, So I like testing and comparing my current engine vs other engines like VRay or Fstorm. Maybe that´s why my little studio has already changed the main renderer 4 times in 7 years... I think I´m just a bit crazy
A good friend that uses both VRay and Corona in their big Studio, always says that VRay is more stable and a bit faster. We are ususally having crashes almost everyday with MAX 2018 and Corona, and the render times are not that consistent, that´s why I wanted to give VRay 5 a try. But my preliminary results show that VRay is slower, so I´m pretty sure I´m doing things wrong.