NVIDIA H100 for private AI

Any news on this GPU accelerator?

Where (in the cloud?) can I test it? More specifically, I would like to test its TEE capabilities (i.e., private AI applications & data). Is there any SDK/documentation available for this GPU?

I expect you have already have looked at the overview page here : H100 Tensor Core GPU | NVIDIA , on this page you will find links to some of the detailed specifications and links to white papers which dive into some of the newer features of this architecture.

There is also an technical blog here : https://developer.nvidia.com/blog/nvidia-hopper-architecture-in-depth/ , but I see your post on the blog asking the same question about availability - so let me find out and will get back to you.

Consider, e.g., SGX from Intel. They have drivers, SDKs, examples on GitHub etc. Basically, everything that is needed to start developing secure apps. I want similar resources to be available for H100. And also to have H100 available in the cloud (AWS, Azure or GCP). Any news on that?

Thanks for your suggestions - its always our objective to provide as much examples and guides to make it easy to incorporate such technologies, and I will pass on your suggestions to the team.
Here are some more links to places you can find more details:
Confidential Computing | NVIDIA
I would also strongly recommend you sign up for GTC coming up Sept 20th, there is a load of content about Hopper
Conference Session Catalog | GTC 2022 | NVIDIA

Full details of the confidential computing have not been officially released and at this moment there is no public access to exercise these capabilities. Attending GTC would be a great way to learn more and also meet some of the hopper team.

I will also send you a direct message - I would love to learn a little more about you use case.
Thanks so much for your patience and I look forward to sharing more news about this once I have it.

Hi, is there any news on the availability of an SDK/tutorials for confidential computing? Our H100 just arrived and I would love to try the confidential computing features to test strategies for securing DL with it.