Is it possible to stream multiple WebRTC sessions using 4 GPUs on a single workstation? Future roadmap?

[minsuk.ko]

Hello NVIDIA Team,

I’m currently working on a project that involves streaming multiple Omniverse Kit applications via WebRTC, using a single workstation equipped with 4 NVIDIA GPUs (Ada 6000 x4). The goal is to allow up to four users to remotely access and interact with separate Kit instances, each mapped to a dedicated GPU and port.

Here’s what I’ve attempted so far:

• Launched multiple .kit apps with unique WebRTC port mappings (e.g., 8211, 8212, etc.).
• Verified that only one WebRTC session works at a time — when a second Kit app is launched, the previous one stops streaming.
• Tried port remapping and multi-container strategies, but still no success in simultaneous sessions.
• WebSocket-based streaming works more reliably, but does not provide the full interactive GUI experience we need.

My questions are:

1. Is it currently possible to stream multiple WebRTC sessions simultaneously from one workstation by assigning each Kit instance to a dedicated GPU?
2. If not possible now, is there any planned support or technical roadmap to enable this functionality in future versions of Omniverse or Isaac Sim?
3. What is the best recommended method at this time to stream multiple sessions interactively via browser (e.g., WebSocket, GDN, or other mechanisms)?
4. Are there any official guidelines or configurations (e.g., Docker, Kubernetes, network namespace isolation) that can help achieve this setup?

Thank you for your support — we’re excited to keep building on Omniverse and would really appreciate your input.

Best regards,

[Richard3D]

Yes this is currently possible, IF you use Linux and Containers and assign each container to each GPU, with different streaming ports. However, as I say on here a lot, it is WAY way easier to take those 4 gpus, split them up into 4 machines and each machine, with 1 A6000 just streams kit normally. No Linux or Containers needed. The machines can be very basic and cheap. It is the A6000 that is expensive. Drop those cards into a $800 machine and you are fine.

That way, each GPU, has its own dedicated CPU, Hard drive, memory and SSD to work with. They are not 'sharing" resources.