LLM-response data recoverable per query via LeftoverLocals (CVE-2023-4969) from un-scrubbed GPU local memory
≈181 MBobserved
| Value kind | observed — Reported measurements, counts, and specifications keep the precision and scope stated by their source; an exact specification is not treated as a range. |
|---|---|
| Scope | Measured on an AMD Radeon RX 7900 XT, which leaked about 5.5 MB per GPU invocation. NVIDIA said its devices were not affected, and the researchers saw only zeros on tested NVIDIA GPUs. |
| As of | 2024 |
| Source | Trail of Bits · Introductory findings: about 5.5 MB per GPU invocation and about 181 MB for each llama.cpp 7B-model query |
| Review | checking…review by 2027-10-16 · standard cadence |
| Recorded changes | last 2026-06-29 |
| Claim id | llm-response-data-recoverable-per-query-via |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.