IPR_ADV vs AS_RT_CUDA vs IPR_RT_CUDA_Hybrid

[biochemical_animations]

Just a quick note...I found Active_Shade_RT_CUDA to be the quickest for look dev for me, (lighting/material adjustments) for a simple scene setup (using 7x 980 ti's)
IPR_RT_CUDA_Hybrid (vray 3.6) and IPR_ADV were pretty fast too (using 40 cores dual xeon 2670 v0...not the fasted cpu's), but full gpu_only above was quickest.

Now, for heavier scenes I haven't tested yet, I'm expecting IPR_RT_CUDA_Hybrid to win out when I need all powerful rendering to help reduce a production render preview.
I'm also expecting IPR_ADV to win out for more accurate and most supported features.

These were tested using BF+BF. Just thought I'd post to see if others saw this same trend. I really like have the various options between the 3 above, I think they all play a vital role. Just a data point for the devs in case they were curious how the end users results came out, also perhaps for future planning on where to combine/advance code and resources.

[kosso_olli]

I am only using IPR_ADV these days, because I need to see exactly the same result as in the high res production render. The others are out immidiately for this reason.
The speed is okay for me. On most scenes I set the IPR to BF+BF, because the LC recalc is distracting me, even if it is very short. Pretty stable, too. Whenever I come across a crash, I send the scene over to the devs or create a dump file.

[Syclone1]

I am only using IPR_ADV these days, because I need to see exactly the same result as in the high res production render. The others are out immidiately for this reason.
The speed is okay for me. On most scenes I set the IPR to BF+BF, because the LC recalc is distracting me, even if it is very short. Pretty stable, too. Whenever I come across a crash, I send the scene over to the devs or create a dump file.

I also stay away from anything GPU related for rendering as well for those same reasons. when GPU can handle anything that CPU does and can match it visually then Ill start doing tests in RT, otherwise Im still happy with the performance currently of IPR...

[biochemical_animations]

Curious if your main usage is for stills or for animations?

[savage309]

Just to be clear - the goal for V-Ray GPU is not to match the Adv render (it will never do). It shares the same UI and mostly the same features with it (Adv of course has more, but there are features that the GPU has, and Adv doesn't). It is a different render engine.

Best,
Blago.

[thanulee]

I wish we could switch to gpu, ipr is pain for complex stuff unless a very powerful ws. Unfortunately, we have so many machines on cpu in our farm, but while that helps in final renders, in look dev, my life would be a lot easier on gpu

[savage309]

I wish we could switch to gpu, ipr is pain for complex stuff unless a very powerful ws. Unfortunately, we have so many machines on cpu in our farm, but while that helps in final renders, in look dev, my life would be a lot easier on gpu

Probably the V-Ray GPU Hybrid mode can make use of those CPU machines 3.6 will be released very soon.

Best,
Blago.

[thanulee]

How would this exactly work Blago? I mean, if i have a bad gpu, will i still be able to utilise cpu with cuda to its full potential? Or in that case is better to go cpu instead of hybrid? cheers

[savage309]

How would this exactly work Blago? I mean, if i have a bad gpu, will i still be able to utilise cpu with cuda to its full potential? Or in that case is better to go cpu instead of hybrid? cheers

We have make it so it is possible to run CUDA code natively on the CPU. Running CPU code on the GPU is very slow, but the opposite, running GPU code on the CPU is not.
The CPU will behave as a regular CUDA device and it will be available in the list of CUDA devices. Because it is in-house implementation, it doesn't need any 3rd party drivers (so you can the CUDA on CPU even on machines that do not have CUDA GPU at all, even on macOS).

Best,
Blago.

[thanulee]

Oh this is great, so with adding good gpus, u just add to the cpu farm and dont care about vram. The limitation though would be some specific features? Also, will ipr lose its meaning like this?

[savage309]

The CPU is a regular CUDA device now. Still, all CUDA devices that you have must be able to fit the scene in their memory. With 3.5 memory optimizations and on-demand textures, this should not be big of a problem.
IPR will stil be around, because of the feature set of V-Ray and V-Ray GPU are different.

Best,
Blago.

[biochemical_animations]

We have make it so it is possible to run CUDA code natively on the CPU. Running CPU code on the GPU is very slow, but the opposite, running GPU code on the CPU is not.
The CPU will behave as a regular CUDA device and it will be available in the list of CUDA devices. Because it is in-house implementation, it doesn't need any 3rd party drivers (so you can the CUDA on CPU even on machines that do not have CUDA GPU at all, even on macOS).

Best,
Blago.

Bravo! Blago.

[Crayox13]

So to clear this, as I initially misunderstood it. Hybrid is not for production rendering at all but to speed up RT?

[biochemical_animations]

So to clear this, as I initially misunderstood it. Hybrid is not for production rendering at all but to speed up RT?

Well, my post was mainly talking about the interactive rending done by ipr or active shade either with adv or rt (namely cuda). I think the hybrid can be use (i'm just about to test it) for production rt cuda rendering. For animations, I had to move to gpu rendering, thus rt cuda was my choice, and the added hybrid could be a ~25% speed increase bonus.

[Alex_M]

So to clear this, as I initially misunderstood it. Hybrid is not for production rendering at all but to speed up RT?

You absolutely can use the new hybrid rendering for production. I think there's quite a bit of confusion about all this hybrid stuff. It's really simple, actually. What hybrid rendering is is just RT CUDA with the option to add CPUs to the rendering, not just video cards. That's all. RT is a production renderer if you set it as such in the render settings (F10) so you can use hybrid RT rendering for production no problems the same way you could use RT for production rendering before Vray 3.6.