Repository navigation
feat(nvidia): validate H100 GPU CC verification on real silicon - #54
Merged
Merged
Conversation
Rewrite wcm.nvidia around the real NVIDIA H100 attestation format, confirmed against a live Standard_NCC40ads_H100_v5 capture (cross-checked with NVIDIA's own local verifier) and committed as a fixture: - the report echoes the RAW 32-byte nonce at offset 4 (not sha256(nonce) like the CPU SEV-SNP/TDX path); - the signature is the last 96 bytes, ECDSA P-384 raw r||s over report[:-96], SHA-384, by the leaf key; - the device cert chain roots in the self-signed NVIDIA Device Identity CA. New NvidiaGpuVerifier / build_gpu_verifier verify the cert chain to the pinned NVIDIA root + the report signature + the raw-nonce binding; parse_gpu_report + verify_gpu_report_signature are the primitives. The KBS gpu_report_verifier takes this verifier; GPU evidence bundles the report plus its device cert chain in GpuReport.quote_b64 (no manifest-schema change). Replaces the 0.20.0 PROVISIONAL placeholder that reused the generic QuoteVerifier with CPU sha256-nonce/DER-signature conventions. WCM ships only the public NVIDIA device root. Completes the real-silicon matrix: SEV-SNP + TDX + H100 CC. SDK 0.22.0, spec v0.15. pytest 279 passed, mypy --strict clean, bandit clean, mkdocs --strict clean, no em dashes. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]>
8 of 17 tasks
imran-siddique
added a commit
that referenced
this pull request
Sep 14, 2026
…ix (#131) docs/roadmap.md still listed NVIDIA CC quote verification and Intel TDX validation under Next, pending hardware. ROADMAP.md, the canonical copy, already records SEV-SNP, TDX and H100 CC validated on real silicon, and H100 landed in #54 on 2026-07-28. The docs copy now matches it. Also: SPONSORS.md and docs/sponsors.md name OPAQUE as sponsor without the ownership disclaimer. Claude-Session: https://claude.ai/code/session_013aK3gVWzNdcM3hZ2o2awK2 Co-authored-by: Claude Opus 5 (1M context) <[email protected]>
1 task done
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What
Flips the NVIDIA H100 GPU-CC verifier from PROVISIONAL to validated, against a real confidential-compute attestation captured on live silicon (a
Standard_NCC40ads_H100_v5, cross-checked with NVIDIA's own local verifier). Completes the real-silicon matrix: SEV-SNP + TDX + H100 CC.The real format (confirmed empirically, encoded here)
sha256(nonce), which is the CPU SEV-SNP/TDX REPORT_DATA convention).r||soverreport[:-96], SHA-384, by the attestation leaf key.GH100 ... GSP FMC LF) roots in the self-signed NVIDIA Device Identity CA.Changes
wcm.nvidiarewritten:parse_gpu_report,verify_gpu_report_signature,NvidiaGpuVerifier(cert chain to pinned NVIDIA root + report signature + raw-nonce binding),build_gpu_verifier. Removed the 0.20.0 placeholder that reused the genericQuoteVerifier(its sha256-nonce/DER-signature assumptions do not fit NVIDIA).kbs.py:gpu_report_verifiernow typed toNvidiaGpuVerifier; GPU evidence bundles the report + device cert chain inGpuReport.quote_b64(no manifest-schema change).python/tests/fixtures/gpu_h100_attestation.json+ pinned rootnvidia_device_identity_ca.pem(root only, mirroring the AMD stance).Verification (the point)
test_real_h100_attestation_verifies_offlineverifies the real captured H100 report offline against the pinned NVIDIA Device Identity CA: verified True. Negatives fail as they must: wrong nonce, a tampered report byte (signature fails), and an untrusted root (chain fails). Synthetic NVIDIA-format tests drive the KBS integration with controllable nonces.pytest279 passed,mypy --strictclean (24 files),banditclean,mkdocs build --strictclean, no em dashes.🤖 Generated with Claude Code