Some thoughts about Keyshot

Started by gabrielefx, April 14, 2017, 10:29:28 AM

Previous topic - Next topic

0 Members and 1 Guest are viewing this topic.

gabrielefx

Hi all,
occasionally I like to test the new Keyshot features. I'm not a Keyshot user, I download the demo version and test it.
I'm attracted by the several renders available on this site.
I like KS and how it works, very simple and intuitive, specially for product design rendering.
I got some workstations with 4 years old Xeons 2660 2,2GH exacore and 4 Titans Maxwell. I purchased these machines with 4 GTX680s.
I'm a Vray RT 3.5 user.
I want to say that gpus can be upgraded, I will upgrade my "old" Titans with the new ones Xp. I can't do the same with cpus.
Unscrewing 4 screws I replace my old gpus. I can double the render speed without touch nothing except 4 screws.
Testing a product design model with Vray RT I can say that after 4 minutes I can get a 2000x2000noise free render, with Keyshot I need more than 50 minutes with my "old" Xeon. (raytrace depth 8, indirect 8, 128 samples, caustics off, gi cache off, 1 hdr, 1 emitting surface)
It's necessary to update the Keyshot render kernel for gpus because to buy the most powerful Xeon processors I have to spend more than 7.000€ for 2 cpus. These Xeons will stay forever inside a new workstation and they will be not upgradable.
Luxion programmers could say: ok, we can't have many Keyshot RT features because the complex algorithms, for example caustics or GI photon mapping pre-computation. Chaos solved the issue using cpus+gpus.
Gpus are a lot faster and cheaper than cpus, are upgradable, this is the point.
With programming everything is possible. I think that a lot of users like me with a gpu based render farm could benefit of this fantastic software because the art of programming not resides inside the cpus or gpus. The fact is that Nvidia is selling tons of gpus and many programmers are investing in gpu based software research.

Best regards

DriesV

Thanks for the feedback.

I think the issue with GPU rendering is still flexibility and platform dependency.
A very significant portion of our users are running KeyShot solely on notebooks for the sake of mobility.
While it's true that you can fairly easily add multiple GPUs to a big tower PC, that machine form factor is simply not desired by many of our users.
GPU rendering doesn't allow for much mobility.

I am not saying some sort of GPU support in KeyShot will never happen, but the way e.g. V-Ray, Octane etc. have implemented it is not viable for our target audience.

Dries

mattjgerard

"It is not that we do not want to use GPUs. The algorithms that make KeyShot fast do not map well to GPUs. We have made a GPU implementation internally, but it was too limited and it only ran on certain types of graphics cards."

"......  think I may quote KS-guys on this ... "GPU rendering in KS .. Ain't never gonna happen"  ......"

This was from a while ago, but this seems to be the sentiment, and some of the reason that some of the other major renderers are sticking with CPU rendering. There seems to be a stop-loss line where the scene complextiy gets to high for GPU renderers and CPU renderers are more efficent at loading huge sets of polygons.

The other issue is that it appears that the base code of KS was written before GPU rendering was really a thing, and the code doesn't really translate from CPU rendering to GPU rendering without significant re-writing.