VRay Next GPU + Irradiance map?

[narmer_visual]

Hello there,

I can see on the V-Ray GPU Supported Features site that the engine does not support Irradiance Map . Are there any plans to implement it into VRay GPU?
We are planning to do some hardware upgrades, and I would like to see if we usually go with IM+LC solution for our renders, is it worth to upgrade our GPU-s or rather the CPU-s.

I can see that VRay GPU has a lot of features implemented in the new version, but still, the CPU rendering has more options to optimize render times. One of these options is Irradiance Map, which is a huge time saver. I ran some test renders on VRay Next CPU and GPU, and I found that the CPU version with IM+LC still produces faster noiseless renders, than the GPU solution.

So what do you suggest, if we prefer the IM+LC solution (due to faster results), should we buy more CPU's instead of going for VRay GPU? Or we should forget Irradiance map, because a fast GPU will produce as much quick noiseless results?

Also an other question, what do you suggest if we plan to make a bigger upgrade only on our Workstation's hardware, so the render nodes would not have really fast GPU-s, so we could only use them for CPU rendering.
What could be a better construction?
- a Workstation with a fast GPU (2x Geforce 1080 Ti) but without render nodes?
- or a Workstation with reasonably good CPU + 5 nodes with basic CPU (i5)?

Thanks for the help!

Regards,
Gabor

[savage309]

We still have a version of the irradiance map in the kernels, but we have hiden it from the UI and we don't think that it is a better idea to use BF+LC or BF+BF for V-Ray GPU. We don't really have plans to put it back for V-Ray GPU.
Results are more predictable and stable with the BF/LC. If you want to get rid of the noise faster, you can use the V-Ray denoiser.

Keep in mind that V-Ray and V-Ray GPU are two different render engines. They share mostly the same workflow and features, but they will not render 1:1 the same.
V-Ray GPU can "emulate" (with zero overhead) the CPU to behave as a regular CUDA device (we call this "V-Ray GPU Hybrid mode") and in this case the results match perfectly.

If you want to go with V-Ray GPU, investing in GPUs is usually a better idea.

Best,
Blago.

[narmer_visual]

Hello Blago,

Thanks for the feedback. I agree that results are more predictable with BF/LC, they always were! Irradiance Map always needed to be supervised and fine tuned, but if someone knew what to do with it, it always produced very fast results. I'll be a bit disappointed if it won't be a feature in GPU. VRay always was good in providing a lot of options for opitmizing render times. So it would be nice to have it, even if it takes some skills to use it. And I'm sure I'm not the only one with this opinion, if i read this discussion, from 3 years ago:

https://forums.chaosgroup.com/forum/...p-please/page4

But still, I surely can live with it, if you won't support it, but it's hard to decide between CPU and GPU.
Can you tell if one machine with fast GPU (without render nodes) + denoiser can produce as fast renders and as good image quality, as a CPU construction with 5 render nodes using IM+LC? Or it will be definitely slower generally.

[PIXELBOX_SRO]

IM for GPU would definitelly be a step back IMHO.
Its unpredictable, fiddly and i am just so happy i dont have to use it anymore

[narmer_visual]

Sounds good, as long as the render times can be maintained low. As I said, IM needs skills and a lot of experience to use it well, but that pays off. I keep my opinion that VRay's strength always was to offer these options to optimize the render as you want. If you know the options, that is a very powerful tool.

I'm happy to hear that you prefer the GPU method, but are your renders not slower though?

[narmer_visual]

My other question that at least Irradiance Map HAS options! Denoiser does not really have any! What about animations? Does flickering not appear in case of a denoised animation? Is that not unpredictable?

[savage309]

There is no flickering in denoised animations. The V-Ray denoiser is specifically designed like that. V-Ray GPU also is designed specifically to need as few render settings as possible (and I think it achieved that goal very good), while being as fast as possible.
Because performance of rendering approaches depends on the scene, I can't really say which one will be best for you. It is always recomended to test yourself if possible.

Best,
Blago.

[Micha]

Hi savage309

I'm still using an old version VfR vesrion based on Vray 3 and from time to time I try to upgrade to Vray latest release. At the moment I'm testing VfR 4 next and I run in the same problem like narmer_visual . One of my main areas of my business are train and air plane interiors, where I have a lot of glossy plastic surfaces (floor, walls, ceiling, seats). Also, often I need a lot of high res shots of a scene, so if the setup is well adjusted, only render times counts.

I did a lot tests to optimize the render speed and found, that IM+LC still provide the best combination of speed, quality and noise. My hope was Vray 4 could be my start to work with GPU power (more power per machine and easier to upgrade). So, please consider to bring the good old IM to the GPU too. It's a really strong tool for still images.

Also very important for me is the usage of the GI caches. Often my clients needs from the same scene different versions, where some textures and colors needs to changed only. Per LC/IM cache quick done.

Best -
Micha

[Micha]

Today I run a test - a large complex train interior - LC+IM (CPU) approx. 30min and LC+BF (CPU) approx. 1h30min.

[M.Max]

Sorry to bring this again, but it would be really great if we could use Irradiane map with GPU renderer even if it is calculated on CPU first .
I've moved to Bf+LC for stills since 3.6 and will never look back ,but for fly through animations, irradiance map can save huge rendering time (2x to 3x in my tests with cpu) with accepted results and the clients wouldn't tell the difference.
I remember there was an option to use pre-cached irradiance map , is that still accessible some how using maxscript?