How do I effectively manage and troubleshoot the use of H100 GPUs in Confidential Computing (CC) mode, especially in multi-GPU systems with NVLink, and ensure seamless integration with both Intel TDX and AMD SEV-SNP environments? I am encountering issues such as SPDM session failures, GPU pass-through problems, and attestation errors, particularly when running non-CC workloads or mixing CC-enabled and CC-disabled GPUs. Additionally, I need clarification on the compatibility of H100 GPUs with different guest OS configurations and ensuring secure communication across multiple GPUs in CC mode. Could someone provide insights or resources on optimizing GPU setup, firmware updates, and handling DMA traffic in these complex scenarios?
sbellock
2
The questions asked are too broad to be answerable. You’ll have better luck asking for help with specific issues.
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
How to run H100 GPU without CC Mode? | 5 | 452 | February 28, 2024 | |
Userguide to get started with H100 GPUs? | 6 | 1185 | January 23, 2024 | |
Can I test Confidential Computing with H100 and TDX on Different Machines | 2 | 132 | January 6, 2025 | |
TDX Confidential VM with non-CC GPU | 4 | 209 | October 24, 2024 | |
Is it possible for an H100 without CC mode to run in CVM, like TDX? | 1 | 90 | August 12, 2024 | |
Does AMD SEV or SEV-ES support H100 CC? | 8 | 129 | August 18, 2024 | |
Announcing Confidential Computing General Access on NVIDIA H100 Tensor Core GPUs | 2 | 269 | May 14, 2025 | |
If H100 use SPDM mutual authentication? | 5 | 823 | January 17, 2024 | |
Confidential Compute on GitHub? | 2 | 670 | July 26, 2023 | |
Use CC in multi-GPU system (with NvSwitch) | 6 | 1009 | October 28, 2023 |