Cloud Gaming - NVIDIA GRID vGPU Support

Question

A customer wants to use G4dn.xlarge machines to provide a cloud gaming service. They're testing with a small set of users (around 10-20) with the possibility - if working and cost-wise interesting - to add way more users to the equation.

They are evaluating the G4dn.xlarge to be way more powerful than necessary for the kind of games they wish to provide in their portal. Therefore, they would like to use [NVIDIA vGPU Architecture][1] to share the GPU with 2 (or more) VMs on top of the same EC2 instance. Is this possible? Are there customers doing this today?

My understanding from some of the researches I've done online is that [it shouldn't be possible][2], however you guys might know some workaround or alternatives. AppStream / Workspaces is not currently on the table. Thanks!

[1]: https://www.nvidia.com/en-us/data-center/virtual-gpu-technology/
  [2]: https://docs.nvidia.com/grid/cloud-service-support.html#aws-ec2

Accepted Answer

It is possible to launch multiple CUDA contexts on a single GPU, however I would not recommend it as performance with each context can be impacted. if g4dn.xlarge is actually too "powerful" then what is needed for the customer. They can look at the g4dn.metal. It does provide the flexibility of your customer to install their own hypervisor and NVIDIA Virtual GPU Manager and will allow them the flexibility to slice the 8 T4s as they wish with different profiles (minimum in 1GB increments).

Cloud Gaming - NVIDIA GRID vGPU Support

Conteúdo relevante