I have several servers equipped with 8 H100 GPUs each, but the CPU in these servers are too old and do not support TDX. Additionally, I have a workstation with a CPU that supports TDX, but it does not have an H100 GPU, as the cost is quite high.
I would like to test the construction of a confidential computing environment, but I want to avoid long periods of downtime for the servers. Is there a method or approach that would allow me to leverage the TDX-enabled workstation and the H100 GPUs on the servers to communicate and test the construction of the confidential computing environment in a non-disruptive way?
Is there a method or approach that would allow me to leverage the TDX-enabled workstation and the H100 GPUs on the servers to communicate and test the construction of the confidential computing environment in a non-disruptive way?
Not that I can think of. If you could get a TDX CPU (assuming it’s cheaper than a H100 for your workstation) for your eight GPUs, then in theory you should be able to spin up a TDX VM and assign a single GPU to it for test and development purposes, leaving the other seven GPUs for production workloads. I’m not sure about the security implications of that, but it should be fine for test and development.