New V-Ray GPU Rendering Architecture At GTC 2018

[flat]

Chaos Group To Debut ?Over 80% Faster? V-Ray GPU Rendering Architecture At GTC 2018
https://techgage.com/news/chaos-grou...e-at-gtc-2018/

No information on forums?

Is this new V-Ray GPU engine different from the Next Beta 2 V-Ray GPU engine?

[savage309]

We have been working on that for a while and we haven't announced anything here yet.

Yes, it is different than Beta 2. However, currently the new rendering architecture should produce exactly the same result as the one in Beta 2. The difference will be the faster rendering.
It will allow us to continue to add features in the V-Ray GPU and also gives us many optimisations opportunities with different approaches for the future. It is also designed for the future GPU architectures in mind.

Also, we have developed a way to speed up the rendering even more than that, if you happen to have some GPU memory left after the scene is being loaded. Keep in mind that effectively this will make it look like V-Ray GPU Next Beta 3 uses more memory than before on simple scenes. This is not the case though - if you happen load bigger scene, it will still again fit in the GPU memory and it will be faster.

It is quite a major change, so it will be great if you can test that with Beta 3 (which is coming very soon) and report if there are any issues.

Best,
Blago.

[flat]

Thank you for information.
I am looking forward to Next Beta 3.

[super gnu]

woo now that sounds quite exciting!

[super gnu]

coming back to this.. care to shed any more light on this new architecture? or do we have to wait for beta3?

for example, i note it mentions nvidia Volta.. is it something card-specific? gains also on AMD cards? cpu based CUDA? 80% speedup is incredible... on cpu based cuda it might overtake normal Vray?!

is that best-case scenario, or typical speedup? with that kinda boost ( plus more says Blago!!) the titan V actually starts sounding like good value :P the days of rendering animations overnight on a single machine seem upon us (although of course, scenes will just get more complex and rendertimes will stay constant.. if its not a law, it should be.. maybe "Robin's iron law of rendering" -no matter the cpu or rendering speedup, final frames will always end up taking the same time.. )

[savage309]

The new rendering architecture affects primary the GPUs (CUDA GPUs tested far more). There will be not much of a difference CPU based CUDA. It is about making the code better for the GPU architectures.
We have seen up to 2 times faster rendering on some scene, on average the speed up is less. The speedup with Volta is bigger than the speedup with Pascal. This comes as no surprise, as the architecture is made with the future GPUs in mind.

We haven't seen a case where it is slower. Few studios are testing it in production for two months now. Hopefully, because of their help, it will be quite stable in the initial release.

What is more important for us is that it will give possibilities to improve and extend the GPU kernels for a long time ahead.

Best,
Blago.

[super gnu]

my maths is too bad to be working in 3d. my initial reaction was blimey! "80%faster =20% of original rendertime = 5x faster" doh. obviously i need either a maths class and/or more sleep.

[oglu]

Whats new on the table at GTC 2019.?

https://www.awn.com/news/chaos-group...acing-gtc-2019

[oglu]

Video is now online.
https://on-demand.gputechconf.com/gt...video/_/S9821/