GPUs · NVIDIA
Can the RTX 3080 run local LLMs?
With 10 GB of VRAM (about 9.0 GB usable) and 760 GB/s of memory bandwidth, the RTX 3080 runs 14 of 39 popular open models fully on the GPU at 8K context. The largest is Qwen3.5 9B at Q5_K_M, at roughly 74 tokens per second.
Best picks for the RTX 3080
Every model on the RTX 3080
Results at 8K context, picking the highest quality quantization that fits. "Offload" means part of the model runs from system RAM (we assume 32 GB), which works but is much slower.
| Model | Status | Quant | Memory | Speed |
|---|---|---|---|---|
| Qwen3.5 9B | Runs well | Q5_K_M | 8.2 GB | ~74 tok/s |
| Qwen3 8B | Runs well | Q6_K | 8.5 GB | ~73 tok/s |
| Llama 3.1 8B | Runs well | Q6_K | 8.2 GB | ~75 tok/s |
| DeepSeek-R1 Distill 8B | Runs well | Q6_K | 8.2 GB | ~75 tok/s |
| Qwen2.5-Coder 7B | Runs well | Q6_K | 7.7 GB | ~79 tok/s |
| Mistral 7B v0.3 | Runs well | Q6_K | 7.5 GB | ~84 tok/s |
| Qwen3.5 4B | Runs well | Q8_0 | 6.4 GB | ~99 tok/s |
| Gemma 3 4B | Runs well | Q8_0 | 6.9 GB | ~110 tok/s |
| Qwen3 4B | Runs well | Q8_0 | 5.8 GB | ~120 tok/s |
| Phi-4 Mini 3.8B | Runs well | Q8_0 | 5.2 GB | ~120 tok/s |
| Llama 3.2 3B | Runs well | Q8_0 | 4.6 GB | ~150 tok/s |
| Llama 3.2 1B | Runs well | Q8_0 | 1.6 GB | ~390 tok/s |
| Gemma 3 1B | Runs well | Q8_0 | 1.4 GB | ~470 tok/s |
| Qwen3 0.6B | Runs well | Q8_0 | 0.9 GB | ~780 tok/s |
| Mistral Nemo 12B | Offload | Q4_K_M | 9.2 GB | ~10 tok/s |
| Gemma 4 12B | Offload | Q4_K_M | 11 GB | ~9.6 tok/s |
| Phi-4 14B | Offload | Q4_K_M | 11 GB | ~7.8 tok/s |
| Gemma 3 12B | Offload | Q4_K_M | 11 GB | ~9.4 tok/s |
| Qwen3 14B | Offload | Q4_K_M | 11 GB | ~7.7 tok/s |
| DeepSeek-R1 Distill 14B | Offload | Q4_K_M | 11 GB | ~7.7 tok/s |
| gpt-oss 20B | Offload | Q4_K_M | 14 GB | ~15 tok/s |
| Codestral 22B | Offload | Q4_K_M | 16 GB | ~4.3 tok/s |
| Devstral Small 24B | Offload | Q4_K_M | 17 GB | ~3.9 tok/s |
| Mistral Small 3 24B | Offload | Q4_K_M | 18 GB | ~3.9 tok/s |
| Gemma 4 26B-A4B (MoE) | Offload | Q4_K_M | 19 GB | ~13 tok/s |
| Qwen3.6 27B | Offload | Q4_K_M | 19 GB | ~3.3 tok/s |
| Qwen3 30B-A3B (MoE) | Offload | Q4_K_M | 21 GB | ~14 tok/s |
| Qwen3-Coder 30B-A3B | Offload | Q4_K_M | 21 GB | ~14 tok/s |
| Gemma 3 27B | Offload | Q4_K_M | 22 GB | ~3.1 tok/s |
| Qwen3 32B | Offload | Q4_K_M | 23 GB | ~2.6 tok/s |
| DeepSeek-R1 Distill 32B | Offload | Q4_K_M | 23 GB | ~2.6 tok/s |
| Qwen2.5-Coder 32B | Offload | Q4_K_M | 23 GB | ~2.6 tok/s |
| Qwen3.6 35B-A3B (MoE) | Offload | Q4_K_M | 23 GB | ~14 tok/s |
| Gemma 4 31B | Offload | Q4_K_M | 25 GB | ~2.7 tok/s |
| Llama 3.3 70B | Won’t fit | — | 48 GB | — |
| DeepSeek-R1 Distill 70B | Won’t fit | — | 48 GB | — |
| Llama 4 Scout 109B-A17B | Won’t fit | — | 73 GB | — |
| gpt-oss 120B | Won’t fit | — | 77 GB | — |
| Qwen3 235B-A22B (MoE) | Won’t fit | — | 155 GB | — |
What this means in practice
The RTX 3080 is comfortable with models up to about 7B to 9B parameters. gpt-oss 20B is the most useful step up beyond that: it needs about 14 GB, so part of it runs from system RAM, but as a mixture-of-experts model it stays reasonably quick at around 15 tokens per second. Models from Llama 3.3 70B upward need 48 GB or more and are out of reach. NVIDIA cards have the broadest support: Ollama, LM Studio and llama.cpp all use CUDA out of the box.
The next memory tier up is 11 GB: GTX 1080 Ti.
Compare the RTX 3080
Questions
What is the biggest model the RTX 3080 can run?
At 8K context the largest model that fits entirely in its 10 GB of VRAM is Qwen3.5 9B at Q5_K_M, at roughly 74 tokens per second.
Can the RTX 3080 run 70B models like Llama 3.3 70B?
No. It needs about 48 GB at Q4, which is far beyond 10 GB even with system RAM offload.
What is the best model for coding on the RTX 3080?
Qwen3.5 9B at Q5_K_M fits fully and runs at about 74 tokens per second. Pair it with an editor extension like Continue pointed at Ollama.
Estimates based on memory size and bandwidth; real speed varies with drivers, software and context length. See how we calculate VRAM.