Questions about multi-HMD CloudXR

Current test machine (i7-5930K, 64G, 1080Ti x1, WiFI 5GHz, Oculus Quest2 x2)

I have three questions:

  1. I have successfully used oculus to connect to the CloudXR server, but it seems that it can only support a single client connection? (The second HMD is disconnected)

  2. Discuss on the Internet “In virtual reality, if you use RTX 6000, RTX 8000 or NVIDIA A40, you can support up to two simultaneous CloudXR sessions, depending on the application requirements.” Is there a list of graphics cards that support multiple players?

  3. We have ESC4000-E10 (supports 8 GPUs). Can it increase the number of players of CloudXR?

Thanks for your response

Answer 1: Correct, CloudXR supports one HMD with one SteamVR instance.
Answer 2: How many users can be supported by a single, virtualized graphics card is largely dependent on the VR application. We know that for typical encoding workloads a single card Turing or Ampere generation card can support a maximum of two users for HMDs with specs similar to those of the Quest 2.
Answer 3: The max. with this server is 2 users per GPU, but the actual will depend on the application and the server (CPU, memory, etc…) and, of course, the network. If for instance you had 16 users in a single area over a WiFi 5 network the users would likely have a poor experience due to the network. Thus, there are a few variables on scaling, but I can say the max. is two users per GPU - the actual will vary with all the variables.

1 Like

Thanks for your clear answer.

According to current understanding:

  1. To support dual HMD VR on a single graphics card, I must use Nvidia turing or NVIDIA Ampere series graphics cards? All models are shown in the following two URLs? (Does GEFORCE RTX 30XX released after GEFORCE RTX 20XX also support dual HMD VR?)
    -NVIDIA Turing: Reinventing Gaming Graphics
    -NVIDIA Ampere Architecture | NVIDIA

  2. Currently CloudXR only supports one HMD with one SteamVR instance, I would like to ask whether it is possible to run two steamVR instances in CloudXR to support dual HMD VR? (If yes, how to do it)

  3. Is it possible to use ESC4000-E10 (supporting 8 GPUs) or a higher specification host with WIFI 6 for multiple HMD connections? (According to the official data, the recommend required bandwidth of an HMD is 60Mbps, 16 users is 16*60=960Mbps, theoretically one WIFI 6 should be able to meet this requirement)
    a. I try to run CloudXR on a host with T4 x 4, but only one HMD VR can connect. Is this correct?
    b. A single host has 4 GPU cards to provide CloudXR services for four HMD VR. Is it necessary to use four Docker Containers or other?

Thanks for your response

Just to the multi-HMD questions, yes you are correct, only one HMD can connect to one OS instance. Steam itself is not multi-instance. You’d need multiple VMs, one for each Steam+CloudXR instance needed.

@GJones-NVIDIA-XR-Team is better suited to answer the questions about GPU capabilities.

@cannady.will can check my answer. We’ve not tried to run multiple users on a multiple GPU AWS instance. Since SteamVR apps are windows-based a simple container strategy most likely won’t work. So, I believe the way to look at it right now is: 1 AWS Instance = 1VM = 1 SteamVR/CloudXR Driver = 1 User, no matter how many GPUs are in the AWS instance

Correct – 1 AWS instance per CloudXR user as vGPU segmentation is not available on AWS. Multiple GPUs help with applications that are written to take advantage of NVLink, but does not influence available user sessions on AWS per server.