title: llama.cpp RX 7800 XT benchmarks
pretty_name: llama.cpp RX 7800 XT benchmarks
license: mit
tags:
- llama-cpp
- gguf
- benchmark
- rocm
- rx-7800-xt
- local-llm
llama.cpp RX 7800 XT benchmarks
Sanitized benchmark archive for local llama.cpp/GGUF experiments on an AMD Radeon RX 7800 XT workstation. This dataset is benchmark data only: model weights, local system/network diagnostics, environment files, scripts, caches, and oversized raw dumps are excluded.
Start here
summaries/run-index.csv- index of benchmark result directories.summaries/practical-audit-index.csv- compact index of practical-audit runs.metadata.json- machine-readable sanitation and upload manifest.raw-clean/- selected raw benchmark logs with local paths and token-looking strings redacted.manifests/- included/excluded file manifests for auditability.
Sanitization policy
Redacted: local paths, temporary paths, hostnames, MAC addresses, and token-looking strings.
Excluded: model weights/checkpoints, system/network diagnostics, run.env / .env, local runner scripts, binary assets, compressed archives, files larger than 25 MB, caches, virtualenvs, and git directories.
Caveats
Some benchmark runs are exploratory, partial, failed, or intentionally blocked by model-fit/runtime constraints. Treat per-run summaries as authoritative when present and raw logs as audit evidence. Results are specific to RX 7800 XT local testing and should not be generalized to other hardware or backends without rerunning.