- Posts: 13
- Joined: Wed Nov 11, 2015 10:56 am
Dear all,
We have been using Fusion to generate small, personalized videos since 2015. Until now we used 2 dedicated servers with Rendernode Version 7.7.1. This has been working great and even still does.
Now we must move to a different computer centre where we must use virtual machines. We are now moving to a Proliant DL380 Gen10 with VMware ESXi, 7.0.3 where we have 2 VM’s with 8 vCPU’s and 32 GB Memory. Because a of the “GPU with at least 2 GB of VRAM which supports Metal, OpenCL 1.2 or CUDA 10” Requirement we added a NVIDIA GRID T4-4Q to the server. We are running Windows Server 2019 Version 10.0.17763 Build 17763
Because Rendernode 7.7.1 is very old, I decided to upgrade to v17.4.5.
Everything did run smoothly but I noticed a rather big performance degradation. I decided to also install version 7 again and do a back-to-back comparison with a small composition. Here are the results:
Rendernode Version 17
Render completed successfully at Wed 8:48AM - Total Time: 0h 0m 18.49s, Average: 4.97 frames/second
Render completed successfully at Wed 8:49AM - Total Time: 0h 0m 18.12s, Average: 5.08 frames/second
Render completed successfully at Wed 8:50AM - Total Time: 0h 0m 20.25s, Average: 4.54 frames/second
Rendernode Version 7
Render completed successfully at Wed 8:55AM - Total Time: 0h 0m 10.35s, Average: 8.89 frames/second
Render completed successfully at Wed 8:56AM - Total Time: 0h 0m 10.41s, Average: 8.84 frames/second
Render completed successfully at Wed 8:57AM - Total Time: 0h 0m 10.47s, Average: 8.79 frames/second
As you can see version 17 is about 40% slower
This morning I tried with the latest version 18.1.1 with similar results
Render completed successfully at Wed 10:38AM - Total Time: 0h 0m 15.01s, Average: 6.13 frames/second
Render completed successfully at Wed 10:43AM - Total Time: 0h 0m 16.97s, Average: 5.42 frames/second
Render completed successfully at Wed 10:46AM - Total Time: 0h 0m 14.40s, Average: 6.39 frames/second
Although a little quicker not as fast as version 7.
As I tested version 7 I noticed in the performance monitor that the GPU was hardly used at all. So I decided to test version 18 with GPU Options set to CPU and I got almost the same results as with version 7
Render completed successfully at Wed 10:48AM - Total Time: 0h 0m 10.54s, Average: 8.73 frames/second
Render completed successfully at Wed 10:49AM - Total Time: 0h 0m 10.75s, Average: 8.56 frames/second
Render completed successfully at Wed 10:50AM - Total Time: 0h 0m 11.54s, Average: 7.97 frames/second
That irritates me, did we buy that expensive NVIDIA Grid Card for nothing?
I hope somebody can help me sort this out:
We have been using Fusion to generate small, personalized videos since 2015. Until now we used 2 dedicated servers with Rendernode Version 7.7.1. This has been working great and even still does.
Now we must move to a different computer centre where we must use virtual machines. We are now moving to a Proliant DL380 Gen10 with VMware ESXi, 7.0.3 where we have 2 VM’s with 8 vCPU’s and 32 GB Memory. Because a of the “GPU with at least 2 GB of VRAM which supports Metal, OpenCL 1.2 or CUDA 10” Requirement we added a NVIDIA GRID T4-4Q to the server. We are running Windows Server 2019 Version 10.0.17763 Build 17763
Because Rendernode 7.7.1 is very old, I decided to upgrade to v17.4.5.
Everything did run smoothly but I noticed a rather big performance degradation. I decided to also install version 7 again and do a back-to-back comparison with a small composition. Here are the results:
Rendernode Version 17
Render completed successfully at Wed 8:48AM - Total Time: 0h 0m 18.49s, Average: 4.97 frames/second
Render completed successfully at Wed 8:49AM - Total Time: 0h 0m 18.12s, Average: 5.08 frames/second
Render completed successfully at Wed 8:50AM - Total Time: 0h 0m 20.25s, Average: 4.54 frames/second
Rendernode Version 7
Render completed successfully at Wed 8:55AM - Total Time: 0h 0m 10.35s, Average: 8.89 frames/second
Render completed successfully at Wed 8:56AM - Total Time: 0h 0m 10.41s, Average: 8.84 frames/second
Render completed successfully at Wed 8:57AM - Total Time: 0h 0m 10.47s, Average: 8.79 frames/second
As you can see version 17 is about 40% slower
This morning I tried with the latest version 18.1.1 with similar results
Render completed successfully at Wed 10:38AM - Total Time: 0h 0m 15.01s, Average: 6.13 frames/second
Render completed successfully at Wed 10:43AM - Total Time: 0h 0m 16.97s, Average: 5.42 frames/second
Render completed successfully at Wed 10:46AM - Total Time: 0h 0m 14.40s, Average: 6.39 frames/second
Although a little quicker not as fast as version 7.
As I tested version 7 I noticed in the performance monitor that the GPU was hardly used at all. So I decided to test version 18 with GPU Options set to CPU and I got almost the same results as with version 7
Render completed successfully at Wed 10:48AM - Total Time: 0h 0m 10.54s, Average: 8.73 frames/second
Render completed successfully at Wed 10:49AM - Total Time: 0h 0m 10.75s, Average: 8.56 frames/second
Render completed successfully at Wed 10:50AM - Total Time: 0h 0m 11.54s, Average: 7.97 frames/second
That irritates me, did we buy that expensive NVIDIA Grid Card for nothing?
I hope somebody can help me sort this out:
- Has someone experience with a VM and the NVIDEA Grid card?
- Might Windows Server 2019 be the problem?
- Due to the production-chain we use, we use *.tga input and *,tga output. I also tried with mp4 for input and output but still the CPU Option is quicker as the Auto or GPU option.
- We mostly use only simple Text and tracking nodes in our composition. This might not be optimal for GPU rendering. But if I set GPU option to CPU I will not be able to use the GPU if we are going to use 3D rendering in the future. And we are wasting an expensive graphic card.
- Are there any other settings I can try to improve performance?