Technical Inquiry Regarding Non-Standard Resolution, Dual GPU Utilization, and Frame Rate Issues in nDisplay 2110 Workflow

Community Showcase Film, TV & Animation
,
1

We are currently building a large-scale virtual production system with the following configuration:

  • Number of Render Servers: 26
  • Server Specifications:
    • CPU: Threadripper PRO 7955WX
    • GPU: NVIDIA RTX 6000 Ada × 2
    • Sync Card: NVIDIA Sync 2
    • DPU: BlueField-2
  • Roles:
    • The first 12 servers are used for outer frustum rendering. Each server’s GPU 0 outputs two 2560×3072 signals (DP to HDMI), connected to Brompton SX40 processors
    • 8 servers are dedicated to inner frustum rendering (dual camera views)
  • Total LED Screen Resolution: 30720 × 5120

We have encountered several issues and would appreciate your support and clarification:

1. Resolution Settings
Currently, each output is set to 2560×3072, which is a non-standard resolution. When using NVIDIA Sync 2 cards (firmware 2.02), we observed that new batches of Sync 2 cards exhibit instability at this resolution (signal flickering, the sync indicator alternates between green and orange, HDMI signal drops intermittently). Replacing these with earlier batch Sync 2 cards resolves the problem and restores signal stability.
Could you please advise:

  • Are there any known compatibility differences between Sync 2 batches (even with the same firmware) when operating at non-standard resolutions?
  • Is it recommended to switch to a standard resolution (such as 3840×2160) to improve compatibility and stability?
  • Are there any firmware or driver updates, or best practices, to resolve issues with new batch Sync 2 cards at non-standard resolutions?

2. Dual GPU Utilization
Each server is equipped with dual RTX 6000 Ada GPUs, but currently only GPU 0 is used for video output.

  • Is it necessary or beneficial to utilize both GPUs for output in our current outer and inner frustum rendering scenario?
  • If so, what is the recommended configuration to maximize dual GPU performance?
  • For inner frustum (multi-camera) rendering, is single GPU output or dual GPU collaboration recommended?

3. nDisplay Frame Rate Issue
When testing nDisplay in Unreal Engine 5.4 and 5.5, we observe only 21fps even on an empty level. With our current hardware, there should not be a performance bottleneck.

  • Could there be a software or hardware configuration issue causing this low frame rate?
  • Does UE 5.4/5.5 impose any default frame rate limits in nDisplay 2110 workflow?
  • Do you have any troubleshooting guidance or recommended configurations to achieve higher frame rates?

4. Additional Recommendations
For such a large-scale, multi-region output system, do you have any additional suggestions regarding stability or compatibility? Are there recommended driver versions, Sync 2 firmware versions, or best configuration practices?

1 Like
2

I unfortunately cannot really offer any advice, but we’re running 24 render nodes here with 2 GPUs as well. We run 1 Unreal instance per GPU, which outputs to a single screen. Therefore we can distribute the load quite nicely on all GPUs, but I don’t think this can be used for your scenario, as only one GPU is connected to output.

However, we have the same performance issues as well - nDisplay seems to not scale nicely with number of nodes, and additionally adds quite a bit of a general overhead. I was hoping this was a Linux/Vulkan issue, but your post makes me think it happens on WIndows as well…

Regarding the sync cards, this is a very interesting issue. We’r ein the process of renewing our cluster, but this means I’ll defenitely keep the old batch around in case we’re running into similar issues. It’s probably best to contact Nvidia here directly, I doubt anyone will answer here.