NVIDIA · 2020-09
Best local LLMs for the RTX 3090
24 GB VRAM · 936 GB/s · ~23 GB usable. The value 24 GB card on the used market — still the entry to 32B-class models.
Check RTX 3090 price →Try it with your context & use case
Preset to the RTX 3090. Change the context length or filter by use case.
23 models run well and 1 run tight on RTX 3090 (any context — just fitting the weights) — 23 GB usable.
Sort:
| Model | Size | Fit | Best quant | Needs | Memory | Speed |
|---|---|---|---|---|---|---|
| Qwen3 32B | 32.8B | Runs well | Q5_K_M | 23 GB | ~29 tok/s | |
| Qwen2.5 32B | 32.8B | Runs well | Q5_K_M | 23 GB | ~29 tok/s | |
| Qwen2.5-Coder 32B | 32.8B | Runs well | Q5_K_M | 23 GB | ~29 tok/s | |
| DeepSeek-R1 Distill Qwen 32B | 32.8B | Runs well | Q5_K_M | 23 GB | ~29 tok/s | |
| Qwen3 30B-A3B (MoE)MoE | 30.5B | Runs well | Q5_K_M | 22 GB | ~187 tok/s | |
| Gemma 3 27B | 27.4B | Runs well | Q6_K | 23 GB | ~30 tok/s | |
| Gemma 2 27B | 27.2B | Runs well | Q6_K | 22 GB | ~30 tok/s | |
| Mistral Small 3 24B | 23.6B | Runs well | Q6_K | 19 GB | ~35 tok/s | |
| Qwen3 14B | 14.8B | Runs well | Q8_0 | 16 GB | ~43 tok/s | |
| Qwen2.5 14B | 14.8B | Runs well | Q8_0 | 16 GB | ~43 tok/s | |
| DeepSeek-R1 Distill Qwen 14B | 14.8B | Runs well | Q8_0 | 16 GB | ~43 tok/s | |
| Phi-4 14B | 14.7B | Runs well | Q8_0 | 16 GB | ~43 tok/s | |
| Mistral Nemo 12B | 12.2B | Runs well | Q8_0 | 13 GB | ~52 tok/s | |
| Gemma 3 12B | 12.2B | Runs well | Q8_0 | 13 GB | ~52 tok/s | |
| Gemma 2 9B | 9.2B | Runs well | FP16 | 19 GB | ~36 tok/s | |
| Qwen3 8B | 8.2B | Runs well | FP16 | 17 GB | ~41 tok/s | |
| Llama 3.1 8B | 8.0B | Runs well | FP16 | 16 GB | ~42 tok/s | |
| Qwen2.5-Coder 7B | 7.6B | Runs well | FP16 | 15 GB | ~44 tok/s | |
| Mistral 7B v0.3 | 7.3B | Runs well | FP16 | 15 GB | ~46 tok/s | |
| Gemma 3 4B | 4.3B | Runs well | FP16 | 9.1 GB | ~78 tok/s | |
| Qwen3 4B | 4B | Runs well | FP16 | 8.5 GB | ~84 tok/s | |
| Llama 3.2 3B | 3.2B | Runs well | FP16 | 7.0 GB | ~105 tok/s | |
| Llama 3.2 1B | 1.2B | Runs well | FP16 | 3.2 GB | ~272 tok/s | |
| Mixtral 8x7B (MoE)MoE | 46.7B | Runs (tight) | Q3_K_M | 23 GB | ~69 tok/s | |
| DeepSeek-R1 671B-A37B (MoE)MoE | 671B | Won't fit | — | 273 GB | — | |
| Qwen3 235B-A22B (MoE)MoE | 235B | Won't fit | — | 96 GB | — | |
| Llama 4 Scout (MoE)MoE | 109B | Won't fit | — | 45 GB | — | |
| Qwen2.5 72B | 72.7B | Won't fit | — | 30 GB | — | |
| Llama 3.3 70B | 70.6B | Won't fit | — | 29 GB | — | |
| DeepSeek-R1 Distill Llama 70B | 70.6B | Won't fit | — | 29 GB | — |
Estimates, computed in your browser — VRAM, quantisation and speed vary with your runtime and settings. How we estimate →
Top picks for the RTX 3090
Every model on the RTX 3090
| Model | Size | Fit | Best quant | Needs | Speed |
|---|---|---|---|---|---|
| Qwen3 32B | 32.8B | Runs well | Q4 | 22 GB | ~34 tok/s |
| Qwen2.5 32B | 32.8B | Runs well | Q4 | 22 GB | ~34 tok/s |
| Qwen2.5-Coder 32B | 32.8B | Runs well | Q4 | 22 GB | ~34 tok/s |
| DeepSeek-R1 Distill Qwen 32B | 32.8B | Runs well | Q4 | 22 GB | ~34 tok/s |
| Qwen3 30B-A3B (MoE)MoE | 30.5B | Runs well | Q5 | 22 GB | ~187 tok/s |
| Gemma 3 27B | 27.4B | Runs well | Q4 | 21 GB | ~41 tok/s |
| Gemma 2 27B | 27.2B | Runs well | Q5 | 22 GB | ~35 tok/s |
| Mistral Small 3 24B | 23.6B | Runs well | Q6 | 21 GB | ~35 tok/s |
| Qwen3 14B | 14.8B | Runs well | Q8 | 17 GB | ~43 tok/s |
| Qwen2.5 14B | 14.8B | Runs well | Q8 | 17 GB | ~43 tok/s |
| DeepSeek-R1 Distill Qwen 14B | 14.8B | Runs well | Q8 | 17 GB | ~43 tok/s |
| Phi-4 14B | 14.7B | Runs well | Q8 | 17 GB | ~43 tok/s |
| Mistral Nemo 12B | 12.2B | Runs well | Q8 | 15 GB | ~52 tok/s |
| Gemma 3 12B | 12.2B | Runs well | Q8 | 16 GB | ~52 tok/s |
| Gemma 2 9B | 9.2B | Runs well | FP16 | 21 GB | ~36 tok/s |
| Qwen3 8B | 8.2B | Runs well | FP16 | 18 GB | ~41 tok/s |
| Llama 3.1 8B | 8.0B | Runs well | FP16 | 17 GB | ~42 tok/s |
| Qwen2.5-Coder 7B | 7.6B | Runs well | FP16 | 16 GB | ~44 tok/s |
| Mistral 7B v0.3 | 7.3B | Runs well | FP16 | 16 GB | ~46 tok/s |
| Gemma 3 4B | 4.3B | Runs well | FP16 | 10 GB | ~78 tok/s |
| Qwen3 4B | 4B | Runs well | FP16 | 9.6 GB | ~84 tok/s |
| Llama 3.2 3B | 3.2B | Runs well | FP16 | 7.8 GB | ~105 tok/s |
| Llama 3.2 1B | 1.2B | Runs well | FP16 | 3.4 GB | ~272 tok/s |
| Mixtral 8x7B (MoE)MoE | 46.7B | Runs (tight) | Q2 | 21 GB | ~81 tok/s |
| DeepSeek-R1 671B-A37B (MoE)MoE | 671B | Won't fit | — | 303 GB | — |
| Qwen3 235B-A22B (MoE)MoE | 235B | Won't fit | — | 98 GB | — |
| Llama 4 Scout (MoE)MoE | 109B | Won't fit | — | 46 GB | — |
| Qwen2.5 72B | 72.7B | Won't fit | — | 33 GB | — |
| Llama 3.3 70B | 70.6B | Won't fit | — | 32 GB | — |
| DeepSeek-R1 Distill Llama 70B | 70.6B | Won't fit | — | 32 GB | — |
Similar hardware
FAQ
What is the best LLM for the RTX 3090?
For general use, Qwen3 32B is the strongest model that runs well on the RTX 3090. See the picks-by-use-case below for coding, reasoning and more.
How much can the RTX 3090 run?
The RTX 3090 has 24 GB of VRAM, of which about 23 GB is usable for a model. That runs 23 of our tracked models well and 1 more at a tight quantisation.
Estimates at 8K context — see how we compute these.