Network Rendering with RTX and non RTX-Cards

[nils_gollub]

Hello,

At the moment I have a setup, where one machine is equiped with an A6000 and the rest with Quadro P4000 (No RTX). Now when I send a render job via backburner and have the Renderengine set to RTX, the machines with the P4000 switch automatically to CUDA but refuses to use the CPU in addition. When I do it the other way around and I send the job with CUDA selected, the machines use the P4000 and CPU, but the A6000 is using CUDA aswell. Is there a way to fix this?

I already tried to set the environment variable accordingly and chose the "select Rendering devices"-Tool, but nothing changes...

Thank you and Best regards

Nils

[Muhammed_Hamed]

Hi nils_gollub and welcome to the forum!

but the A6000 is using CUDA aswell

This should be fine, we have improved Cuda performance for Ampere GPUs in the recent update. It should be close to that of RTX in most cases
You will need a recent V-Ray 5 build and Nvidia driver 496.13 for this

On another note, P4000 could still use RTX mode, and the behavior you described seems to be an issue, I will see if we can reproduce this

Best,
Muhammed

[nils_gollub]

Thank you for the fast response. You are right, its possible to use the P4000 with the RTX Engine, I didn't know this. For this reason the CPU was not joining the work when I send the job with RTX selected. I was sure the P4000 is not able to work with RTX, maybe because in the V-Ray benchmark it is not possible to select it, when running the RTX-Benchmark.
In my case the performance of the RTX Engine on the A6000 is much faster then CUDA, I think at least 25% maybe more. But I dont have the latest Version of Vray (5.1) and also not the latest nvidia driver. I will try to get the IT to make an update and report if something changed.

I think i make a couple of test renderings to find out how the performance difference between two P4000 with RTX is compared to two P4000 combined with the CPU on CUDA. If it's faster, I simply will send all jobs as RTX and all machines work optimal. If not I need to find a compromise...

Best regards

Nils

[Muhammed_Hamed]

maybe because in the V-Ray benchmark it is not possible to select it, when running the RTX-Benchmark

Good point, I added to my notes about the Benchmark tool

In my case the performance of the RTX Engine on the A6000 is much faster then CUDA, I think at least 25% maybe more

yeah, this is expected. and the difference should be smaller with the recent update, RTX still can be faster within 5-15% or so
And in some scene with direct lighting, the difference will be a few seconds

I think i make a couple of test renderings to find out how the performance difference between two P4000 with RTX is compared to two P4000 combined with the CPU on CUDA

P4000 on Cuda + CPU should be much faster, because RTX acceleration on Pascal cards is software only. I have tested this extensively at some point
There is no dedicated raytracing hardware like the case with A6000

I commented about this above to point out the behavior with Backburner is odd

Best,
Muhammed

[nils_gollub]

P4000 on Cuda + CPU should be much faster, because RTX acceleration on Pascal cards is software only. I have tested this extensively at some point
There is no dedicated raytracing hardware like the case with A6000

Yes, i get the same result. It would be good to be able to set an engine per machine. Now both engines are a compromise at one machine or the other. I hope I get the updates soon, so I can do further testing to reach the best performance.

I commented about this above to point out the behavior with Backburner is odd

Would the normal behavior be that it uses the devices selected on the machine rather then the selected Engine in the setup of the job?

Best regards
Nils