Llama 3.1 8B Instruct on AMD Radeon RX 7900 XTX
Measured this month.
Measurement
- tok/s
- 86.4
- TTFT
- 105 ms
- VRAM used
- 5.6 GB
- RAM used
- 1.4 GB
- Power
- 285 W
- Quant
- Q4_K_M
- Context
- 8K
- Run date
- 2026-05-04
- Source
- community
AMD RX 7900 XTX 24GB on Ollama with ROCm 6.2 backend. Throughput trails RTX 4090 by ~17% on the same model + quant — a real but narrowing gap as ROCm matures. Power draw is comparable. The architectural caveat: ROCm kernel coverage is narrower than CUDA, so models with non-standard attention variants may regress further. For the headline LLaMA / Mistral / Qwen architectures, the gap is what these numbers show.
Why this confidence tier?
Confidence is rule-based. Every factor below contributed to the tier. We never expose a single numeric score; the tier label is auditable through this explanation alone.
- +Source: community submission
- Reproduce this benchmark →An independent reproduction with matching numbers lifts the tier and reduces single-source risk.
- Read the confidence methodology →Full editorial standards for tiering.
- Why we don't use percentages →Tier labels — auditable, no opaque score.
Cohort intelligence
How this measurement compares to the rest of the corpus. Only comparable rows (same model + hardware first, with relaxations labelled) are used. We never average across runtimes or quant formats unless explicitly told to.
Same model, different hardware
8 matching rowsWhat this model looks like on adjacent hardware. Drives the 'should I upgrade?' question.
- 78.5 tok/sapple-m4-maxMLX-4bitEditorial
- 78.5 tok/sapple-m4-maxMLX-4bitEditorial
- 105.0 tok/srtx-3090Q4_K_MEditorial
- 55.0 tok/sapple-m3-maxQ4_K_MEditorial
- 118.2 tok/srtx-5080Q4_K_MEditorial
- +3 more
Reproduce this benchmark
Got the same model + hardware combo? Run the same measurement and submit your numbers. We'll pre-fill model, hardware, quant, and context — you just add your tok/s, VRAM, runtime version. If your numbers match within ±15%, this benchmark gets a confidence lift and a reproduction badge.
Related
Drill into the entity pages for this measurement.
Cite or export
Reference this benchmark in your work. Multiple formats; CC-BY attribution required.
Cite this benchmark or paste it into a README. Copy-to-clipboard; license is CC-BY-4.0 (attribution to RunLocalAI required).
<a href="https://runlocalai.co/benchmarks/332" rel="noopener">RunLocalAI: Llama 3.1 8B Instruct on AMD Radeon RX 7900 XTX — 86.4 tok/s</a>
Next recommended step
Got the same model + hardware? Run it and submit your numbers — successful reproductions lift this benchmark's confidence tier.