P028 Taps on scale-up and electrical interconnects (NVLink)
Can traffic on high-speed electrical links between accelerators (such as NVLink/NVSwitch) and other non-optical interconnects be captured or witnessed at full line rate by a trustworthy add-on device?
Capturing or witnessing NVLink/NVSwitch, copper and other non-optical interconnect traffic at line rate.
Terms: NVLink / NVSwitch: Nvidia's high-speed electrical links and switch chips that connect GPUs within a server or rack.
Status
Analyzed Desk analysis of two designs for copper (NVLink-class) links, a modified switch ASIC that copies sampled traffic from its digital fabric and a bump-in-wire retimer exposing its equalized bitstream, with the author's expectations of about 60 ns added per hop, dependence on the switch or retimer vendor, and at least a year of lead time; nothing built or measured. Milton & Reynolds (Amodo, 2026) likewise describe an optical-electrical-optical retimer tap and an electrical interposer at the transceiver port as designs 'not commercially available' that need to be designed and tested. Full line rate and the trustworthiness of an active device: argued (active electronics add attack surface), not tested. Bidirectional circulator-optics TPU pod links: no work. Peigne-Lefebvre et al. (2026) state NVLink traffic has no physical witness.
Bump-in-wire retimer tap: Insert a signal retimer into the copper link and extract a digital copy from its internal equalized bitstream.
Posed by
- Baker, Kulp, Marks, Brundage & Heim (2025), Verifying International Agreements on AI: Six Layers of Verification for Rules on Large-Scale AI Development and Deployment (RAND WR-A4077-1):
However, introducing network taps here would be very difficult.
(App. A.3, p. 42, L2547)
Also raised in: Cankaya (2026), Research Note: The Fundamentals and Feasibility of Secure Network Taps for Verifying AI Datacenter Use; Canary Institute research; Peigné-Lefebvre, Nguyen & Wang (2026), Zero Knowledge Verification for Frontier AI Training Is Possible, arXiv 2606.05433; AI Security Forum, Requests for Discussion; Dean, Verification Plan (AI 2040); Milton & Reynolds (2026), Network Tapping for AI Verification: A Technical Assessment; Proofworks, Living Verification Documents.
Sub-questions
| ID | Sub-question | What the sources report |
|---|---|---|
| P028.1 | Can intra-rack NVLink traffic be captured inline / tapped at line rate by a retrofit device, at what cost, to support sub-batch logging? | open |
| P028.2 | Can NVIDIA Confidential Compute hash NVLink collective payloads at a stated throughput (~NVLink line rate), a witness for tensor-parallel traffic? | open |
| P028.3 | Can bump-in-wire retimer / tapswitch / OEO taps on copper or 106 GBaud electrical lanes be made trustworthy and cost-effective (added latency per hop; 1:2 fan-out retimers available)? | open |
| P028.4 | What is the annual link-failure rate with an OEO tap in the hot path versus without, on 800G production links? | open |
| P028.5 | What fraction of node-end (GPU/NIC) links in new frontier clusters will use co-packaged optics, removing electrical tap points, by a stated date? | open |
| P028.6 | Has a tap been demonstrated on a bidirectional circulator-optics TPU pod link within its link budget? | open |
Possibly relevant work found by automated search
Found by AI agents searching for this problem's sub-questions, mostly by keyword and citation in the OpenAlex scholarly database and on the web; a second AI pass reviewed each one, and we spot-checked some by hand. They seemed reasonably related, as far as one can tell from outside each paper's field. Some come from neighbouring fields, such as arms control and financial audit, and bear on the problem by analogy. Not reviewed against the status above; a paper listed here does not change that status. 928 of 1052 quotes checked word for word against the paper's text.
For P028.2: Can NVIDIA Confidential Compute hash NVLink collective payloads at a stated throughput (~NVLink line rate), a witness for tensor-parallel traffic?
For P028.4: What is the annual link-failure rate with an OEO tap in the hot path versus without, on 800G production links?
For P028.6: Has a tap been demonstrated on a bidirectional circulator-optics TPU pod link within its link budget?
Data
- problems.csv — one row per problem
- problem_status.csv — status, scope and evidence per problem
- problem_sources.csv — the posed-by quote and other documents per problem
- leaves.csv — all sub-questions, including those without a checkable form
- leaf_sources.csv — every source row behind each sub-question