Four RTX 3080s hit 69 tok/s on Qwen3.6-27B for about $2,000
starkruzr · reddit · 2026-07-23
A benchmarker tested four 20GB RTX 3080s on Vast AI for code generation with Qwen3.6-27B and found the setup surprisingly strong.
- Reported around 69 tokens/sec decode at near-max context (256K) with MTP on
- Prefill dropped to 893 at max context with prompt cache off
- The author says the cards can cost about $400 each, with the full machine roughly $2K all-in
- Conclusion: a cheap 4×3080 box can run dense code models fast, with little quantization and no obvious degradation
More from coding & agent
- Non-CS Beginners Are Shipping Code Through CI/CD Pipelines with AI Help — curious_vii · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26
- Hugging Face uses small open models to reverse engineer a game binary byte-for-byte — yb2698 · 2026-10-26
- ContractSim: first framework evaluating LLM negotiation over contingent natural-language contracts — xuanalogue · 2026-10-26
- Hugging Face workshop: enterprise RL, environment design and scaling agent training with open models — Hugging Face · 2026-10-26