Processor Upgrade

[DanSHP]

The bigger challenges with exterior scene are time to first pixel and stopping the render, if the scene is big enough. We have been working on this in H1 of this year, expect improvements in the upcoming release

Best,
Muhammed

I keep hearing this but it never seems to materialize.

Is there anything you can share? Maybe a video testing a large exterior scene, with tons of assets comparing current release and the future one? Will these improvements affect IPR for both CPU and GPU?

[Muhammed_Hamed]

There are currently 2 changes in the nightly builds,

• First change is speeding up loading of scenes with many instances, Forest Pack, Chaos Scatter..etc, this affects RTX mode. This change fixes stopping the renders as well, I have tested with glorybound scene, it works well in his scene. This change slightly reduces GPU memory usage as well
• Second change is speeding up loading of bitmaps on the first rendered frame(we already have caching between animation frames). This change affects all scenes, time to first pixel is improved by around 30 to 50%, and for multiple GPUs the speedup is bigger

This doesn't affect the standard V-Ray(CPU) renderers, it is more about the gap in time to first pixel between V-Ray and V-Ray GPU. V-Ray GPU is much closer to V-Ray now in time to first pixel

We have many scenes that we used for debugging and testing the fixes, including this one
Scene compilation for this scene was more than 3 minutes, and after these 2 changes, it is around 40 seconds.

Best,
Muhammed

[Joelaff]

Do you mean 3Ds Max viewport?
Could it be about GPU memory, the 4070 has 12 GB. Getting close to the memory ceiling will tank the performance
3Ds Max viewport used to not care about high-end GPUs, I wouldn't see a noticeble difference between say a 3090 and 3070 for example. Hence we used to recommend mid-range GPUs for viewport, specially that they would have a lot of GPU memory these days like the 4060ti

I'm interested to test again, I have a 4070 in my office.

Best,
Muhammed

Interesting. Could be. Never did extremely thorough testing, just opened whatever the latest big scene was, turned on fps, orbited around and noted that before changing cards. Then did the same with the new card. The difference was always pretty decent, often doubling fps between x090 generations.

[Joelaff]

This should be better now with the new Out Of Core mode for textures, it reduces GPU memory usage drastically. All your bitmaps will be stored in system memory, with minimal impact on performance. In your case you have a motherboard and processor that uses PCIe Gen 4, V-Ray GPU will use this bandwidth to share data between the GPU and system memory

The bigger challenges with exterior scene are time to first pixel and stopping the render, if the scene is big enough. We have been working on this in H1 of this year, expect improvements in the upcoming release

Best,
Muhammed

Sounds like we should give GPU another shot. My biggest issue is that things revert so differently you can’t easily switch between them. This is the number one reason we rarely even try GPU. Everything would have to be re tweaked, which can be in an issue in a world where clients A/B each new render with the previous. Also, with high end threadrippers the difference seems minimal. Eager to try it with this 7995wx.

[Muhammed_Hamed]

You will need to start the project with V-Ray GPU, as you know V-Ray and V-Ray GPU are different renderers, the results will not match one to one although recently the results should be close enough. We have been trying to avoid objects rendering black for example or the big differences in bump mapping. It should be better in the current and upcoming versions.
Whenever we release a major new version, it is usually a good opportunity to evaluate V-Ray GPU

Best,
Muhammed

[Muhammed_Hamed]

Interesting. Could be. Never did extremely thorough testing, just opened whatever the latest big scene was, turned on fps, orbited around and noted that before changing cards. Then did the same with the new card. The difference was always pretty decent, often doubling fps between x090 generations.

I will have time to test this tomorrow Monday. It is been a while since I tested that myself
I will post the results here

Best,
Muhammed

[Joelaff]

You will need to start the project with V-Ray GPU, as you know V-Ray and V-Ray GPU are different renderers, the results will not match one to one although recently the results should be close enough. We have been trying to avoid objects rendering black for example or the big differences in bump mapping. It should be better in the current and upcoming versions.
Whenever we release a major new version, it is usually a good opportunity to evaluate V-Ray GPU

Looking forward to it. It is the picking the render engine up front that can be difficult. We are fully invested in CPU rendering anyway (farm is designed for CPU). More experimentation is definitely in order.

[Muhammed_Hamed]

I have been testing viewport performance for an hour between 3 different machines, here are some conclusions

• 3080, 3090 and 4090 have identical performance for variety of scenes, using the same CPU
• GPU usage is different between the GPUs, 4090 has the lowest GPU usage when panning or orbiting the viewport. Usually below 20% usage
• 3080 has around 50-60 % usage, which clearly says that the bottleneck is somewhere else(spoilers, CPU is the bottleneck )
• swapping machines to see how the CPU affects performance, the 7950X has nearly twice as much FPS as the Thredripper 3990X. Everything works faster in Max not just the viewport.
• I tested in C4D and Maya, the results are similar

So as long you don't run out of VRAM on your GPU, mid-range cards are more than enough for viewport performance. Say a 3080 or even better a 16-GB 4060 ti are ideal for 3Ds Max

Edit, I put a 3080, 3090 and 4090 in one machine. And the other machines had multiple 3090s

Best,
Muhammed

[Joelaff]

Very interesting. So a big laggy scene that orbits in perspective at less than 20fps does so at about the same speed between a 3090 and a 4090?

What sort of performance benefit do you get in GPU or OpenCL, etc. from moving a 4090 to be processing only (not driving a monitor)? Is it just that you save some VRAM, or is it actually faster?

[Muhammed_Hamed]

Very interesting. So a big laggy scene that orbits in perspective at less than 20fps does so at about the same speed between a 3090 and a 4090?

Yes, same for a 3080. They all have the same performance
I tested with a variety of scenes, from glorybound. If you have a scene you want me to test with let me know

On the other hand, moving the scene to a machine with a better CPU you get 2x the FPS and everything feels better/faster

What sort of performance benefit do you get in GPU or OpenCL, etc. from moving a 4090 to be processing only (not driving a monitor)? Is it just that you save some VRAM, or is it actually faster?

For GPU rendering or Vantage, you get both better performance and more GPU memory usage available..
For viewport performance, you get nothing. It is best to use the GPU connected to monitors, for viewport performance. That is the default behavior, unless you specify otherwise in Nvidia control panel

Best,
Muhammed