V-Ray Benchmark 1.0 released

[vlado]

Can we ask that in an update, there's an option to test both scenes with CPU AND also GPU. That way we can compare our own systems with a scene, CPU VS GPU. Currently we compare our own hardware, CPU OR GPU, to other users, which great. But it would be awesome if we can compare our OWN hardware CPU VS GPU on the same scene

The reason we don't want to do this yet is because the performance of CPU vs GPU is very scene-dependent. The benchmark might render the scene on the GPU X times faster, but that might not be the case for all scenes and we didn't want to mislead users.

In a later update, can we get a real world scene please with a lot more stuff in it and more "happening". The ChaosGroup 3D team did an AMAZING image a while back with a cartoony village with characters all over the place, foliage etc. THAT scene would be great for an exterior.

That scene took gigabytes and gigabytes to render, and took a few hours with DR on the Google cloud (the resolution was huge though!). I'm not sure you want that as a benchmark

But other than that, we will think of more interesting scenes for future versions of the benchmark.

Best regards,
Vlado

[OPEN_RANGE]

Thanks for setting up and providing this excellent and most cool app! You continue to exceed expectations and your commitment to your product and customers is unsurpassed. You are the benchmark <g>.

After running the benchmark app and clicking submit, the web page opened with what I assumed is my result on the top line of the table. The hardware config looked correct. The row is highlighted in blue. My only confusion is with the user name. They show first and last character and the characters in between are asterisked out. On the top line, that I assumed is my result, the username first and last characters do not match my user name. There is also an identicle user name on a row below that has a different hardware config. Should username be disregarded?

Thank you.

[vlado]

For the moment, results are only recorded if they are better than the current best result for the same configuration. If the highlighted user name is different than yours, it means that someone with the exact same hardware configuration submitted a better result. It could be handled a bit better and we'll think about it.

Best regards,
Vlado

[OPEN_RANGE]

OK thanks for that info. I wonder why my same spec cpu renders 1:16 versus the 1:04 that is shown as the highlighted comparable result? Not asking you to diagnose my machine...just curious very recently kicked down for new workstation build. Maybe I need to tune up or maybe 0:12 difference is negligible.

Edit: Tried it again and got 1:14. Still wonder why 0:10 slower than same spec on others system. i was confused by the results showing using Vray 3.57.01, cause I'm on 3.4. then came to understand that the benchmark utility is using that newer 3.57.01 version.

[larex39]

Hi Vlado, i have a question. Is there a reason why opencl gpus are not supported? Does it means that Vray will be more Nvidia Cuda centric in the future than universal Opencl renderer? I am realy nor fan of nvidia so i want to be prepared for the future thanks a lot

[vlado]

We have a lot of problems with OpenCL - driver issues, slow speed etc. It's not properly supported to the extent that we need - neither on AMD, nor on NVidia cards. It never worked properly on Mac OS either.

Best regards,
Vlado

[larex39]

Hi Vlado, thanks for the answer. I still think that opencl is the future because we are free to choose rhe hardware and not to be forced to use Nvidia. I hope that in future the drivers will be more solid each year and also the new opencl implementations will help to solve the issues.

[vlado]

I still think that opencl is the future

Unfortunately it doesn't seem this way.

Best regards,
Vlado

[Nicinus]

The reason we don't want to do this yet is because the performance of CPU vs GPU is very scene-dependent. The benchmark might render the scene on the GPU X times faster, but that might not be the case for all scenes and we didn't want to mislead users.

I understand that, but besides from what hardware configuration to pursue the other important question is which strategy makes most sense going forward. I feel the general preference is swinging back towards CPU power and IPR, after RT and number of Cudas have been all the rage for a while. It is kind of difficult for me as a user to know what I should focus in the longer term going forward, and it must be tough for Chaos as well, so it would be of interest to see how these two different scenes compare to each other using CPU vs GPU, don't you agree? Even if some scenes fit better for one or the other, that in itself is interesting to learn more about.

It's interesting, but I've always wondered if the C render people were out of their mind to only focus on CPU, with all the new Nvidia cards. I've always gotten the impression that it is a matter of relatively short time before CPU can't compare anymore, but maybe Ryzen, etc. is changing the equation?

Long term it looks as if RT and Adv/IPR will eventually converge and be identical solutions from the user's perspective, but which workflow has the upper hand for the next 2-3 years?

[FSGFX]

... but which workflow has the upper hand for the next 2-3 years?

as long as there are no graphics cards with 64gb of ram or more we render on CPU...

[vlado]

I feel the general preference is swinging back towards CPU power and IPR, after RT and number of Cudas have been all the rage for a while.

Hm, I feel it's the exact opposite

It would be of interest to see how these two different scenes compare to each other using CPU vs GPU, don't you agree?

I agree, but we are not ready to have this discussion yet. Next week at GTC we will be talking about something that is related to this topic though.

Best regards,
Vlado

[Nicinus]

I don't know, with IPR and subsampling you get very fast feedback and everything developed over Vray's lifetime works. With RT it takes a while to load into memory and with larger scenes the speed seems to suffer in comparison. At the same time architectural scenes becomes larger and larger. Perhaps RT will be the choice for product and animation, while CPU stays the better alternative for larger complex projects. Would love to go to GTC, seems like an interesting event to get some vision from the main players.

[biochemical_animations]

Perhaps RT will be the choice for product and animation, while CPU stays the better alternative for larger complex projects.

Yep. Until recently, I was unable to use vray adv for animations due to render times and flickering etc, workflow that took too long with ir prepass steps, animation gizmos, etc...I couldn't use it. I had to keep selling scan line animations.

Now with gpu, I can render crazy refraction brute force stuff, flicker free, denoised animations, in about the same time scan line took to make.

I agree gpu for animation, for sure.

[biochemical_animations]

Unfortunately it doesn't seem this way.

Best regards,
Vlado

Second this. I use cuda, and yes, eventhough the coding and drivers are still maturing, I like relying on the experts to run the coding. There is enough IT for me to handle with building a multi-gpu superstation and multi-gpu render farm without extra problems, but right now, I can run Windows 7 with 7x 980 ti's in cuda rt mode on a single system, and nvidia, windows, and vray handle it just fine.

[Alex_M]

as long as there are no graphics cards with 64gb of ram or more we render on CPU...

Wow, you must be rendering some insane stuff! May I ask what scenes did you render that require so much RAM? I would imagine some stuff for movies or something? Even on large exterior scenes with full 3D vegetation I very very rarely see the RAM spiking more than 16-18 GB. I would make a guess that no more than 5% of people work on scenes that require 64 GB RAM to render?