Score breakdown
Popularity is tracked separately. Support, ads, sponsorships, and tips never affect these signals.
Why it matters
Useful for home-lab developers and small teams running one or two RTX 3090s (or 4090/5090) who want a measured, reproducible LLM-serving stack without picking vLLM vs llama.cpp vs ik_llama from first principles: club-3090 is the Apache-2.0 community recipes repo that ships validated Docker compose variants for each (vLLM for max throughput, llama.cpp for max context + cliff-immunity, ik_llama for
Who should use it
Who should skip it
Skip noonghunna/club-3090 unless the captured evidence suggests it solves a problem you are actively working on.
About this signal
noonghunna/club-3090 is tracked by RepoRadar as a rtx 3090/4090/5090 community llm in the Apache-2.0 community recipes for serving LLMs on section. It was first seen on 2026-06-25 and last updated on 2026-06-25. The current verdict is 'try now' with a Gold tier and moderate setup difficulty. The standout signals for noonghunna/club-3090 are workflow potential (9.6) and maturity (9.1), while setup ease (6.4) trails — that balance shapes where it fits best. This page summarizes the evidence RepoRadar has captured from captured source metadata. The score, tier, risk label, and verdict on this page are never influenced by sponsorship, ads, or tips — they reflect only the usefulness, popularity, novelty, momentum, maturity, and evidence signals described in the RepoRadar methodology.
How this item is evaluated
RepoRadar assigned noonghunna/club-3090 a composite score of 8.5 out of 10, placing it in the Gold tier. This score combines weighted sub-signals: usefulness (35%), novelty (18%), momentum (14%), maturity (10%), open-source/build quality (7%), evidence quality (6%), workflow potential (6%), and setup ease (4%). Popularity is tracked separately at 1468.0 and never affects the composite score or tier. The risk label of 'low' reflects inherent user-impacting hazards, not generic novelty. Items with no risk flag may still require normal code review before production use.
Risk explanation
**Linux + Docker-first stack; Windows requires WSL2.** The repo's primary deployment target is Linux + Docker compose; the README is explicit that native Windows runs only the upstream llama.cpp binary — none of the repo's tooling. Windows users must follow `docs/WSL_SETUP.md` start-to-finish before the Docker steps. macOS is documented but most measured TPS numbers come from Linux + Ampere/Ada/Blackwell NVIDIA hardware; Apple Silicon users should not expect cross-rig reproducibility with the published numbers; **Ampere 24GB prefill-cliff wall on single-card vLLM at >50K context.** The README is explicit that on 24GB single-card vLLM, prefill OOMs at >~50K single-prompt context because of the head_dim=512 FA wall (Genesis v7.72.2 PN59 was intended as the fix but does not engage on chunked-prefill). Workarounds are measured and documented: `vllm/dual` with TP=2 escapes it, or `llamacpp/default` (different engine, no cliff). Verify the team's context-length requirements against the published Cliff table (`docs/CLIFFS.md`) before adopting single-card vLLM for long-context workloads; **Community fork (`beellama.cpp`) is unofficial and unvalidated on sm_89/sm_120.** For Gemma 4 31B on Ampere 24GB single-card, the README ships the `beellama` config with the explicit warning that the `beellama.cpp` community fork builds with `FA_ALL_QUANTS=ON` and the multi-arch image is unvalidated on sm_89/120 — users adopting this path should pin the commit and re-run the published benchmarks on their own hardware, or prefer `ik-llama/iq4ks-mtp` if the single-card code-fast profile is not critical.