AWS instance to run NVIDIA Omniverse Audio2Face

Question

I want to try NVIDIA Omniverse Audio2Face. I don't have an RTX GPU, so I want to run it on the cloud.

The requirements, shown in a splash screen of Audio2Face, are:

Minimum:

CPU: Intel Core i7 (7th Generation); AMD Ryzen 5

CPU cores: 4

RAM: 16GB

Storage: 50GB SSD

VRAM: 8GB

GPU: Any RTX GPU

Recommended

CPU: Intel Core i7 (9th Generation); AMD Ryzen 7

CPU Cores: 8

RAM: 32GB

Storage: 500GB SSD

VRAM: 10GB

GPU: GeForce RTX 3070 NVIDIA RTX A4000

I can't seem to find instances that use those specific CPUs and GPUs. I tried on a g3.8xlarge, whose page mentions:

vCPUs based on custom 2.7 GHz Intel Xeon E5 2686 v4 processors ... NVIDIA Tesla M60 GPUs

I set up that machine, install Remote Desktop Protocol, install Omniverse and Audio2Face. I launch it, the button changes to "Launching" and after a few seconds goes back to "Launch." I look at the logs (located at Profile Settings), but I don't see any mention of "failure", "fail" or "error". A thread on the NVIDIA Developer Forum from 6 months ago asks about running it on the cloud, with no answers. So I'm not sure:

• if the instance I set doesn't meet the requirements and I should try a different one

• if it does meet the requirements but running Audio2Face on the cloud requires a dedicated host rather than a virtual machine

• if running Audio2Face on the cloud is just not possible.

How can I find an AWS instance type that meets the minimum or recommended requirements and run Omniverse Audio2Face on the cloud?

Answer 1

I'm not guaranteeing that this is the best way to do it, but it works. I'm running it right now (as in, staring at it on my other screen) on a Windows Server 2022 version running on a g4dn.2xlarge (requires 8vCPU G quota). It was very slow on the xlarge instance (4vCPU).

To make it work on Windows:

1. Install Nvidia drivers
2. Enable the Windows Audio service

Caution: I'm only getting 13 FPS, even though it is not maxing out the resources (41% CPU, 28% Memory, 7% GPU) - I'm wondering if RDP or some other factor is limiting it, or if the system isn't using the GPU properly. I was searching for an answer to that's when I found your question.

I plan to try Linux after this to see if it's better, but the client uses Windows.

Edit: The key is to get an instance backed with a T4. That's backed with an RTX capable graphics card. RTX (ray tracing) is the key for Audio2Face.