Multi-Tenant Support in NVIDIA Confidential Computing (CC) on a Single GPU

Hi,

I have a question regarding whether NVIDIA Confidential Computing (CC) currently provides protection for multiple VMs in a multi-tenant environment on a single GPU.

In Figure 5 of the initial NVIDIA CC whitepaper, there was an explicit use case mentioned as “Confidential computing for multiple tenants per GPU.” However, in the more recent documentation (Secure AI with Blackwell and Hopper GPUs Whitepaper), this specific use case seems to be no longer present.

So, I am wondering if NVIDIA CC currently supports multi-tenant protection on a single physical GPU. If so, what are the architectural requirements (e.g., specific MIG or vGPU configurations) to enable this?

Thank you.

The proposed multi-tenant protection is MIG-based. While the official document doesn’t seem to include this case, the GPU hardware might have supported CC in MIG mode.

Thanks for the reply.

I understand the theoretical possibility, but I’m struggling with the actual technical reproduction. If you have successfully enabled CC and MIG simultaneously, could you provide the specific steps or a guide?

I’m sorry but I haven’t tried. I guess it’s similar to the single-gpu passthrough case except that the pass-throughed GPU is a MIG