Where are the settings to switch denoiser to be GPU accelerated?

[GidPDX]

I was just watching a MAX video on using the denoiser, when I noticed they used GPU acceleration for denoiser.

Where can we turn this on in Vray for MODO? Is it in Vray for MODO? If not any ETA?

I love the denoiser, but it can be a bit slow on large resolution renders.

[vladimir.nedev]

The GPU acceleration is enabled by default.
There is a hidden channel that you can find in the Channels viewport only.



Greetings,
Vladimir Nedev

[GidPDX]

oooh...things hidden in the channel view!

Ok, thanks for pointing it out. I was just finding it slow- probably because my paltry video card and limited Video RAM? The results really are amazing though in most instances. I can't believe the horrible render settings you can throw at it with high resolution renders and what a good job it does.

I'll be sure to try and poke around the channels too for some of these features- thanks!

Would be good to expose this feature though since it's on be default. People could be rendering with the GPU using RT and standalone, or redshift/octane and not know why things are slowing down as they do test renders in Vray thinking they are only using CPU.

[vladimir.nedev]

Would be good to expose this feature though since it's on be default. People could be rendering with the GPU using RT and standalone, or redshift/octane and not know why things are slowing down as they do test renders in Vray thinking they are only using CPU.

I am not sure about this one. The GPU version is much faster than the CPU one, I think. So even if you are using the GPU for something else as well, it might still be faster.
Also, I don't want to put too much options there, it will be confusing to the user which ones are important.
Maybe in the future, when I add support for the proficiency level of the Properties viewport, all these hidden options can go in the Advanced proficiency level.

Greetings,
Vladimir Nedev

[GidPDX]

The problem for me is when the denoiser is running with GPU enabled, my computer is practically useless. I only have one video card installed, and my whole system freezes on and off for 3-15 second stretches. Maybe this is only a windows 10 issue or something?

Would having enough Vram to fit the image into it solve the issue? I only have a 2GB card at the moment- if I had a 1070, or 1080, besides being faster, would the extra Vram mean fewer of these stalls?

[neverko]

Unless I'm missing something the denoiser seems to be running on the CPU by default, even with the mentioned channel set to 'true' (default).

When the denoiser is running my CPU is under 100% load while my GPU is running at 0% load.

Edit: I tried setting the 'Use GPU acceleration' channel to 'false' in case it was reversed, but it still only uses the CPU for denoising.

So the denoiser doesn't seem to be running on the GPU at all according to the system loads I'm seeing while denoising.

[vladimir.nedev]

It switches to CPU, if GPU cannot be used for some reason.
Can you check the Event Log ?
Maybe there is information there, which can help in understanding why it couldn't use the GPU.

Greetings,
Vladimir Nedev

[neverko]

I've gone through all entries in the log and there doesn't seem to be any errors at all or anything pertaining to denoising other than two entries, one for starting the denoising process and one for the time it took to complete the denoising process.

I have a GTX 980 (368.81), Windows 10. Running Modo 10.1v2.

The denoiser is definitely not running on the GPU as I see 100% CPU load in task manager while the process is running and 0% GPU usage in GPU-Z.

[GidPDX]

I wonder if this was broken at some point as I could have sworn earlier builds were using the GPU- I'll check on my system too and make sure.

[vladimir.nedev]

How long does it take to complete the de-noising ?
Is there a difference between GPU on and off ?

Greetings,
Vladimir Nedev

[GidPDX]

For me with GPU enabled for denoiser set to strong preset, during the denoise pass:

49.6 seconds

CPU usage 7.5-15%
GPU usage ~50%


For me without GPU enabled for denoiser set to strong preset, during the denoise pass:

52.9 seconds

CPU usage 80-98%
GPU usage ~15% (normal system levels)

******************************

Seems to be working. I was just testing on a small render region on a not so noisy scene. I have seen substantial differences in speed between GPU and CPU though- with GPU being at least twice, if not several times faster than the CPU at denoising.

Still causes my system to pause in fits and jerks when using GPU denoising though. Why is that- low video ram?

Also, why doesn't it use more of the GPU? Maybe my card is too old and the functions needed are not as prevalent or something on the chip?

[neverko]

I still can't get it to use the GPU for denoising. It's always 100% CPU load and 0% GPU load.

Some OpenCL weirdness? My card works great for OctaneRender, V-Ray RT GPU and other GPU related tasks.

[vladimir.nedev]

What GPU do you have ?

I don't think it could be the amount of RAM, more likely your GPU is very under-powered, so the CPU side of the code can utilize it to 100%.
With a faster GPU it is harder to reach 100% load, because the CPU is stalling it, maybe. Somebody from the GPU team might be able to explain.

Strange that your GPU load shows only 50%, what you describe sounds likes it is used at 100% by the denoiser, so nothing is left for the OS window drawing.

On my machine, CPU - i7 2600, GPU - Quadro M5000, GPU denoising takes 3.5s, while CPU is 26.7s. So, more than 7 times faster.
GPU load is close to 100%, but the Quadro M5000 is not connected to a monitor.

Greetings,
Vladimir Nedev

[vladimir.nedev]

I still can't get it to use the GPU for denoising. It's always 100% CPU load and 0% GPU load.

Some OpenCL weirdness? My card works great for OctaneRender, V-Ray RT GPU and other GPU related tasks.

I think there is a warning/error in the Event Log if it fails to use OpenCL and needs to fall back to CPU, can you write the log to a file and attach it here ?

Greetings,
Vladimir Nedev

[neverko]

I'll get back with a log tomorrow. I'll also try to investigate the issue a little further.