VRay Next Update 1 _IPR Feedback

Hi everyone
let me start by saying I’m super happy with the latest update so thank you very much for all the team!
I’m mainly interested in IPR as that what will really affect my workflow. It is already great but here are my notes/requests /questions : ) )
1- I’ve noticed that GPU IPR to be a bit more fluid and we have settings like Undersampling / IPR rays per pixel to have more control. would it be possible to have those for CPU IPR too?

2-In GPU IPR and especially for interiors where the GI is dominating there seem to be black areas that clean gradually while in CPU IPR we don’t have those areas. is it because of light cache in CPU we don’t get that? (image attached rendered for the same time limit)

3-Region renders as I’ve mentioned on other threads still start from scratch on IPR CPU instead of continuing cleaning like the case of IPR GPU.

5-IPR in the viewport is great but I would love for future to have a bit more control like what we have in VFB. Seeing the Color correction, pick a focus point, render region are the main features for me. if it is possible to hide all the lights/helpers/shapes that would be nice to have cleaner viewport while having them visible in other viewports for editing. I know that’s not the purpose of viewport IPR.In other words, what I’m asking for is having what we had with active shade before, but with the new interactive viewport if possible.

6- There are 2 cases that IPR doesn’t update and these I use like 50 times every day =D :
A- Adding any object to exclude list for Override Mtl under global switches doesn’t update in IPR.
B-hiding Geometry using SHhift +G will hide them in both IPR and viewport but when I unhide them IPR doesn’t update.
For both cases I have to stop IPR and run it again as “Refresh” doesn’t help here.

7- A small annoying bug I found with IPR ..try to click " Show correction controls in VFB while it is in full screen and the menus on the right won’t show up till you minimize or move the VFB which will force IPR to refresh. the workaround for this is to disable Fit IPR resolution to VFB and set it to match the full-screen resolution, but again for the final render, I will need to change it

As you can see most of these points are really simple and IPR is already great now, but I thought I should share them with you.

Best regards,
Nawras

1- I’ve noticed that GPU IPR to be a bit more fluid and we have settings like Undersampling / IPR rays per pixel to have more control. would it be possible to have those for CPU IPR too?

We’ll consider it.

2-In GPU IPR and especially for interiors where the GI is dominating there seem to be black areas that clean gradually while in CPU IPR we don’t have those areas. is it because of light cache in CPU we don’t get that? (image attached rendered for the same time limit)

You can experiment with LC in V-Ray GPU IPR with the maxscript command - renderers.current.ipr_forceBruteForceGI=false . Also you can raise the IRP rays per pixel. We are looking to improving this internally.

3-Region renders as I’ve mentioned on other threads still start from scratch on IPR CPU instead of continuing cleaning like the case of IPR GPU.

We’ll see what can be done about this.

5-IPR in the viewport is great but I would love for future to have a bit more control like what we have in VFB. Seeing the Color correction, pick a focus point, render region are the main features for me. if it is possible to hide all the lights/helpers/shapes that would be nice to have cleaner viewport while having them visible in other viewports for editing. I know that’s not the purpose of viewport IPR.In other words, what I’m asking for is having what we had with active shade before, but with the new interactive viewport if possible.

Some things like the CC and render region might be implemented, but for the rest we are limited by 3ds max. Don’t forget that this is 3ds max Viewport and it’s a viewport, not an image buffer.

6- There are 2 cases that IPR doesn’t update and these I use like 50 times every day =D :
A- Adding any object to exclude list for Override Mtl under global switches doesn’t update in IPR.

Will bump the issue in the system.

B-hiding Geometry using SHhift +G will hide them in both IPR and viewport but when I unhide them IPR doesn’t update.
For both cases I have to stop IPR and run it again as “Refresh” doesn’t help here.

Will bump the issue in the system.

7- A small annoying bug I found with IPR ..try to click " Show correction controls in VFB while it is in full screen and the menus on the right won’t show up till you minimize or move the VFB which will force IPR to refresh. the workaround for this is to disable Fit IPR resolution to VFB and set it to match the full-screen resolution, but again for the final render, I will need to change it

Will add it in the system.

Thanks for the feedback !

Thanks for the detailed response !

Really cool to know that it is possible to use lc with GPU IPR
@savage309 Is this WIP for an upcoming release? or can I do this in Maya?

You can experiment with LC in V-Ray GPU IPR with the maxscript command - renderers.current.ipr_forceBruteForceGI=false

I don’t really know, the Maya guys will know better.
@yolov @Mihail.Djurev

Best,
Blago.

I don’t recommend playing with the light cache for IPR with V-Ray GPU just yet; it’s still WIP and there are unresolved issues yet - f.e. with the adaptive lights. The option is hidden for a reason :slight_smile:

Your observation about the black areas is correct though - the CPU uses the light cache which improves the GI sampling. The GPU IPR is brute force only and it takes more samples to get the same result. I’m hoping that at some point we can enable the light cache for GPU IPR by default.

Best regards,
Vlado

One of the main reasons I’m still not doing GPU rendering. IPR previews with GPU always come out dark and don’t correctly represent the final look as IPR CPU does no matter how long I wait for the GPU to clear the noise. I’m always wondering how are people doing interiors with GPU. Do they constantly rerender when they do changes like we did before IPR was invented? I’m curious to know. I can never go back to not using IPR again.

GPU rendering is all about IPR interactivity/speed. Vray GPU does this really well and many people have been using it in production.
Vray’s GPU IPR uses BF+BF(which is the case with RS, Octane and others) but it still can use adaptive lights..and, having the right hardware, it is very fast and feedback is almost instant..The experience is quite different compared to using IPR on CPU
In the testing above, M.Max used a single GTX 970 (BF+BF) vs 2x 2696V4 (72 threads and BF+LC) Still he talked about the GPU IPR being more fluid.
I do all my rendering on GPU since 3.6 and many other people have been using it in production, quality has been really good so probably you are doing something wrong.
https://www.behance.net/gallery/6628…cans-triplanar

I’m interested to see what I’m doing wrong. With the GPU in my signature I’m getting extremely slow light evaluation compared to IPR CPU (6 minutes vs 25 seconds). This makes my GTX 1080ti 14.5x times slower than my 1950X in IPR GPU (at least in this scene). Please check the screenshot below and notice how much longer IPR GPU took to resolve the noise to CPU IPR levels. All GPU settings are at their defaults.

IPR aside, I’m getting slow GPU production rendering performance compared to my 1950X too. Compare the 2 screenshots below. Both were set to resolve to Noise threshold level of 0.015. 1950X took 25 sec. and 1080ti took 3 full minutes. This time difference makes the 1950X 7.2x times faster than my 1080ti. This doesn’t seem right. Again, I’ve not change ANY setting, everything is left at default except for noise threshold which I set the same for both.

For the IPR specifically, on the CPU we use the light cache while on the GPU we use brute force, which would explain the large difference in render times for IPR. At some point we will enable the light cache for GPU IPR as well.

The production rendering case is more interesting and I would like to take a look, if possible?

Best regards,
Vlado

No, unfortunately this cannot be done in Maya at this point.

Thanks Alex and Vlado!

I will wait until it is officially released. Im quite happy with the GPU IPR
And considering Octane 4 and RS are using BF+BF for their IPR, I’m looking forward to light cache being implemented Into Vray’s GPU IPR. It will be a massive speed boost.
Even now I prefer working with Vray’s IPR for things like debug shading and the VFB.. and speedwise, The IPR is as fast as competition..

In my testing, noise threshold doesn’t match between CPU and GPU
Usually I have to use .005 for CPU to match .015 on Vray GPU
I would recommend you benchmark your 1080Ti as well and have low GPU thread priority off when you are rendering with Vray GPU.
And for this testing you might want to test render something that takes 5 minutes or longer.. In your screenshots you can see the reflection on TV and lighting on left wall, not matching the CPU render, so you might want to make both scenes as close as possible. You should probably use Bucket mode for both engines as well.
I’m using 3x 1080Tis and an i9 7900X @4.8 GHz.. This setup has been way faster than rendering on CPU alone, and with Hybird I can use my CPU in rendering as well.

Okay thanks for the clarification.
I don’t mind using BF for IPR now ..actually in the other renderer that we use (Corona ) there is an option to “Force Path tracing” for IPR to avoid the cost of re-computation to the UHD cache which comes handy in some cases ,so having such option could be good idea too imo .

@M.Max Agreed!
Considering that a 1080Ti is 4 times faster than a 970, your render times are the expected difference between the 2 engines.
There are differences between the 2 engines, which is expected to happen, like how the lights are brighter And bump mapping is strong on roof material compared to the CPU render
Also side by side, the GPU image seems to have less noise. So, trying to match the same noise pattern you will probably need to use lower noise threshold on CPU
So yeah, Something is going on with @Alex_M scene.
Here is another test I did, 3200x3200 Res
Using substance textures and VRscans to make sure that the 2 renders are matching as close as possible
Vray CPU >> 35m 11s
Vray Cuda using only C++/CPU >> 40m 30s
Vray Cuda with 3 1080Tis and C++/CPU >> 8m 24s seconds
For this scene I used .02 noise threshold for the GPU render and it is really impressive how Vray GPU deals with noise. Check out the sample rate screenshot.
To match the same clean render I needed to use .01 noise threshold for Vray CPU and still in dark areas, the image rendered on GPU is cleaner! See below
So rendering on GPU has been almost 5 times faster than my OCed i9 7900X @4.8 GHz, I can add 2 more cards to my setup, but with CPUs you are mostly limited to 1 or 2
And for animation, I usually do one frame per GPU using this tool by Alex Kurcera
http://vrscenegui.de
It saves a lot of render time.