Issues with Missing Blueprint Files and Optimization for Deploying MetaHumans

Omniverse Core Platform
1

Hi everyone,

I’m currently working on deploying MetaHumans using NVIDIA ACE and Omniverse. While setting up the project, I encountered a couple of challenges that I’d like to get some advice on:

Missing Blueprint Files

The documentation mentions using Blueprints for scalability and managing user interactions, but I’ve noticed that some key Blueprint files seem to be missing from the GitHub repository. Without these files, it’s been difficult to properly integrate and test the interaction frameworks for the digital avatars.

  • Are there alternative locations where the complete Blueprint files can be downloaded?
  • Or is there an updated version of the repository that I might have overlooked?

Performance Optimization for Concurrent Users

From my initial testing, even with four GPUs, the system seems to handle only about three concurrent users effectively. For our use case, this is quite limiting, as we aim to support a larger number of simultaneous interactions.

  • Are there best practices or configurations to optimize performance and extend the system’s capacity for concurrent users?
  • Could you recommend specific NVIDIA SDKs, frameworks, or settings (e.g., for Omniverse or TensorRT) that might help reduce GPU resource consumption per user?

Any guidance or insights into resolving the Blueprint issue and scaling concurrent user interactions would be greatly appreciated as we plan the next steps for our project.

Thanks in advance for your support!

Best regards,
Martin Lofrano
Founder, Medicina Consulta https://medicinaconsulta.com.br

3

Hello, and thank you for your inquiry. I see that we have two parts to this question.

Let me start by saying on the first one, with the blueprints, I will have to do some research on that. We do not directly support, digital humans, or the ACE workflow on this forum. Developers can discuss ACE through our NVIDIA Developer Discord Server NVIDIA Developer

For the second question about the memory use and consumption, I have some questions. You mentioned that you were trying to use it with 3 uses and one machine with with 4 GPUS. Is this a central Omniverse Server and you have 3 users all remoting into this one machine? This is not the ideal way of doing things.

Ideally, each user is using their own powerful local desktop machine and GPU, and each one is using USD Composer. Then Nucleus connects the 3 users and 3 live nucleus sessions into one multi user session.

If you have multiple users all using one session of USD Composer somehow, all 3 logged into one remote machine, there will be very significant lag yes. It is nothing to do with having multiple GPUs. The bottleneck will just be 3 users, all trying to use one session of USD Composer at the same time.

Can you describe your hardware setup in more detail please.