{"slug": "club-3090", "name": "club-3090", "category": "ai", "category_name": "AI tools and apps", "description": "", "recommenders": 8, "commenters": 8, "warners": 0, "score": 7.45, "rising": -0.29, "first": "2026-05-07", "latest": "2026-10-02", "why": "https://github.com/noonghunna/club-3090 is the way to go.", "why_is_quote": true, "has_verdict": false, "sentiment": {"recommend": 9, "neutral": 0, "compare": 0, "warn": 0}, "spark": [0, 0, 0, 0, 0, 0, 1, 3, 3, 1, 0, 1], "listed": true, "loved": [], "caveat": "", "short_quote": "https://github.com/noonghunna/club-3090 is the way to go.", "kind": "", "icon": false, "domain": null, "github_repo": "github.com/noonghunna/club-3090", "official_url": "https://github.com/noonghunna/club-3090", "pricing": "", "aliases": ["club-3090"], "verdict": null, "praise": [{"label": "", "summary": "", "quotes": [{"id": 799026784, "item_id": 49939174, "url": "https://news.ycombinator.com/item?id=49939174", "by": "gchamonlive", "date": "2026-10-02", "version": null, "stance": "recommend", "quote": "https://github.com/noonghunna/club-3090 is the way to go.", "excerpt": "small native inference engine optimized first for DeepSeek V4 Flash (including the experimental vision model), DeepSeek V4.1 Flash (Metal, and text inference on CUDA), and additionally GLM 5.2 and 5.3, GLM 5.3 Flash and DeepSeek V4 PRO, and Qwen3.8 Flash Next (Metal and CUDA) This is local targeting high end consumer hardware like DGX Spark or AMD Ryzen AI Halo. For our mere mortals that were kids not long ago and ca", "context": "", "use_case": "local Qwen on x090 GPUs", "switched_from": "", "story": "From the creator of Redis; run LLM locally with ds4"}, {"id": 788797569, "item_id": 49299848, "url": "https://news.ycombinator.com/item?id=49299848", "by": "m_ke", "date": "2026-08-14", "version": null, "stance": "recommend", "quote": "using the config from https://github.com/noonghunna/club-3090", "excerpt": "With spec decode should easily get to >100tps on my dual 3090s qwen 3.5 27b was running at around 110tps using the config from https://github.com/noonghunna/club-3090 make that 200tps on a single 5090, 4x faster than opus https://x.com/radixark/status/2088285681131110446 devs about to get handed a two 5090 box each and told to max that out", "context": "", "use_case": "", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 783597264, "item_id": 48974829, "url": "https://news.ycombinator.com/item?id=48974829", "by": "xiconfjs", "date": "2026-07-19", "version": null, "stance": "recommend", "quote": "you should look into https://github.com/noonghunna/club-3090", "excerpt": "you should look into https://github.com/noonghunna/club-3090", "context": "", "use_case": "", "switched_from": "", "story": "Qwen3.8 is launching and going open-weight soon"}, {"id": 780489376, "item_id": 48780586, "url": "https://news.ycombinator.com/item?id=48780586", "by": "gchamonlive", "date": "2026-07-03", "version": null, "stance": "recommend", "quote": "Run https://github.com/noonghunna/club-3090 with beellama, fast inference at the cost of a reduced 102k context window", "excerpt": "There's a sub 2k tier with a single 3090 that's also serviceable. Run https://github.com/noonghunna/club-3090 with beellama, fast inference at the cost of a reduced 102k context window", "context": "", "use_case": "local LLM on single 3090", "switched_from": "", "story": "Jamesob's guide to running SOTA LLMs locally"}, {"id": 780221041, "item_id": 48763815, "url": "https://news.ycombinator.com/item?id=48763815", "by": "mswphd", "date": "2026-07-02", "version": null, "stance": "recommend", "quote": "I've heard that club 3090 makes that relatively straightforward", "excerpt": "dense models are (more) compute heavy, so are generally worse to run on mac. mac tends to be better for (larger) MoE models. 27B dense can fit on a consumer graphics card. Even without getting into various \"intrusive\" ways to shrink the size of a model (e.g. REAP), something like a NVFP4 quant of Qwen3.6 27b https://huggingface.co/nvidia/Qwen3.6-27B-NVFP4 should fit within ~22GB of VRAM. So easily on a 5090. It would", "context": "", "use_case": "multi-3090 local setup", "switched_from": "", "story": "Kimi K2.7 Code is generally available in GitHub Copilot"}]}], "caveats": [], "who": [], "tags": [], "use_cases": ["local qwen on x090 gpus", "local llm on single 3090", "multi-3090 local setup", "optimized local model configs", "local llm server setup guides"], "compared": [{"slug": "qwen", "name": "Qwen", "co": 4, "switched_from": 0, "switched_to": 0}], "makers": [], "threads": [{"url": "https://news.ycombinator.com/item?id=49936575", "title": "From the creator of Redis; run LLM locally with ds4", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=49299684", "title": "Qwen3.8-27B", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48966120", "title": "Qwen3.8 is launching and going open-weight soon", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48775921", "title": "Jamesob's guide to running SOTA LLMs locally", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48756602", "title": "Kimi K2.7 Code is generally available in GitHub Copilot", "mentions": 1}, {"url": "https://news.ycombinator.com/item?id=48721903", "title": "Qwen 3.6 27B is the sweet spot for local development", "mentions": 1}], "sentiment_by_month": {"2025-11": {}, "2025-12": {}, "2026-01": {}, "2026-02": {}, "2026-03": {}, "2026-04": {}, "2026-05": {"recommend": 1}, "2026-06": {"recommend": 3}, "2026-07": {"recommend": 3}, "2026-08": {"recommend": 1}, "2026-09": {}, "2026-10": {"recommend": 1}}, "versions": [], "version_unnamed": 0, "family_ref": null, "runs_on": null, "lines": [], "quotes": [{"id": 799026784, "item_id": 49939174, "url": "https://news.ycombinator.com/item?id=49939174", "by": "gchamonlive", "date": "2026-10-02", "version": null, "stance": "recommend", "quote": "https://github.com/noonghunna/club-3090 is the way to go.", "excerpt": "small native inference engine optimized first for DeepSeek V4 Flash (including the experimental vision model), DeepSeek V4.1 Flash (Metal, and text inference on CUDA), and additionally GLM 5.2 and 5.3, GLM 5.3 Flash and DeepSeek V4 PRO, and Qwen3.8 Flash Next (Metal and CUDA) This is local targeting high end consumer hardware like DGX Spark or AMD Ryzen AI Halo. For our mere mortals that were kids not long ago and ca", "context": "", "use_case": "local Qwen on x090 GPUs", "switched_from": "", "story": "From the creator of Redis; run LLM locally with ds4"}, {"id": 788797569, "item_id": 49299848, "url": "https://news.ycombinator.com/item?id=49299848", "by": "m_ke", "date": "2026-08-14", "version": null, "stance": "recommend", "quote": "using the config from https://github.com/noonghunna/club-3090", "excerpt": "With spec decode should easily get to >100tps on my dual 3090s qwen 3.5 27b was running at around 110tps using the config from https://github.com/noonghunna/club-3090 make that 200tps on a single 5090, 4x faster than opus https://x.com/radixark/status/2088285681131110446 devs about to get handed a two 5090 box each and told to max that out", "context": "", "use_case": "", "switched_from": "", "story": "Qwen3.8-27B"}, {"id": 783597264, "item_id": 48974829, "url": "https://news.ycombinator.com/item?id=48974829", "by": "xiconfjs", "date": "2026-07-19", "version": null, "stance": "recommend", "quote": "you should look into https://github.com/noonghunna/club-3090", "excerpt": "you should look into https://github.com/noonghunna/club-3090", "context": "", "use_case": "", "switched_from": "", "story": "Qwen3.8 is launching and going open-weight soon"}, {"id": 780489376, "item_id": 48780586, "url": "https://news.ycombinator.com/item?id=48780586", "by": "gchamonlive", "date": "2026-07-03", "version": null, "stance": "recommend", "quote": "Run https://github.com/noonghunna/club-3090 with beellama, fast inference at the cost of a reduced 102k context window", "excerpt": "There's a sub 2k tier with a single 3090 that's also serviceable. Run https://github.com/noonghunna/club-3090 with beellama, fast inference at the cost of a reduced 102k context window", "context": "", "use_case": "local LLM on single 3090", "switched_from": "", "story": "Jamesob's guide to running SOTA LLMs locally"}, {"id": 780221041, "item_id": 48763815, "url": "https://news.ycombinator.com/item?id=48763815", "by": "mswphd", "date": "2026-07-02", "version": null, "stance": "recommend", "quote": "I've heard that club 3090 makes that relatively straightforward", "excerpt": "dense models are (more) compute heavy, so are generally worse to run on mac. mac tends to be better for (larger) MoE models. 27B dense can fit on a consumer graphics card. Even without getting into various \"intrusive\" ways to shrink the size of a model (e.g. REAP), something like a NVFP4 quant of Qwen3.6 27b https://huggingface.co/nvidia/Qwen3.6-27B-NVFP4 should fit within ~22GB of VRAM. So easily on a 5090. It would", "context": "", "use_case": "multi-3090 local setup", "switched_from": "", "story": "Kimi K2.7 Code is generally available in GitHub Copilot"}, {"id": 779601504, "item_id": 48725094, "url": "https://news.ycombinator.com/item?id=48725094", "by": "nsbk", "date": "2026-06-29", "version": null, "stance": "recommend", "quote": "Have a look at this repo with single/double 3090 optimized configs", "excerpt": "I beg to differ. Have a look at this repo with single/double 3090 optimized configs for Qwen and Gema models: https://github.com/noonghunna/club-3090", "context": "", "use_case": "Optimized local model configs", "switched_from": "", "story": "Qwen 3.6 27B is the sweet spot for local development"}, {"id": 777339552, "item_id": 48583722, "url": "https://news.ycombinator.com/item?id=48583722", "by": "eurekin", "date": "2026-06-18", "version": null, "stance": "recommend", "quote": "I'd jump right into verified set-ups and writeups, like this one:", "excerpt": "If I started today, with building a server, I'd jump right into verified set-ups and writeups, like this one: https://github.com/noonghunna/club-3090 You can find info about running a patched version of vllm for 1x24gb, 2x and 4x. There's also quite a few \"blackwell\" subreddits, where people seem to share a lot of substantial information, if you're going the 6000 route.", "context": "", "use_case": "Local LLM server setup guides", "switched_from": "", "story": "Local Qwen isn't a worse Opus, it's a different tool"}, {"id": 776934818, "item_id": 48558426, "url": "https://news.ycombinator.com/item?id=48558426", "by": "dstryr", "date": "2026-06-16", "version": null, "stance": "recommend", "quote": "club 3090's single card setup", "excerpt": "This is not my experience at all. Even the Nous Research guys have stated that \"Qwen3.6-27B is the canonical local model to use Hermes Agent with\" [https://old.reddit.com/r/LocalLLaMA/comments/1sz2y76/ama_with_nous_research_ask_us_anything/oiynr8i/]. I am finding the same when used with Pi and OpenCode. Gemma will just stop mid-tool call. It's been slower and I've had to reduce context size to run it. Qwen3.6 27b has", "context": "", "use_case": "", "switched_from": "", "story": "Running local models is good now"}, {"id": 768769875, "item_id": 48048117, "url": "https://news.ycombinator.com/item?id=48048117", "by": "pferdone", "date": "2026-05-07", "version": null, "stance": "recommend", "quote": "may wanna give it a go", "excerpt": "I can see that and I don't know your setup, but there are people pushing >70t/s with MTP on a single 3090, with big contexts still >50t/s. 64k is not a lot for agentic coding, and IIRC 128k with turboquant and the likes should be possible for you. r/LocalLLM/ and r/LocalLLaMA/ are worth a visit IMO. EDIT: just found this recipe repo, may wanna give it a go: https://github.com/noonghunna/club-3090 EDIT-2: this can als", "context": "", "use_case": "", "switched_from": "", "story": "ZAYA1-8B: An 8B Moe Model with 760M Active Params Matching DeepSeek-R1 on Math"}]}