GPUs · NVIDIA
Can the RTX 5080 run local LLMs?
With 16 GB of VRAM (about 14.4 GB usable) and 960 GB/s of memory bandwidth, the RTX 5080 runs 21 of 39 popular open models fully on the GPU at 8K context. The largest is gpt-oss 20B at Q4_K_M, at roughly 130 tokens per second.
Best picks for the RTX 5080
Every model on the RTX 5080
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 |
|---|---|---|---|---|
| gpt-oss 20B | Runs well | Q4_K_M | 14 GB | ~130 tok/s |
| Qwen3 14B | Runs well | Q5_K_M | 13 GB | ~59 tok/s |
| DeepSeek-R1 Distill 14B | Runs well | Q5_K_M | 13 GB | ~59 tok/s |
| Phi-4 14B | Runs well | Q5_K_M | 13 GB | ~60 tok/s |
| Gemma 3 12B | Runs well | Q6_K | 14 GB | ~62 tok/s |
| Mistral Nemo 12B | Runs well | Q6_K | 12 GB | ~62 tok/s |
| Gemma 4 12B | Runs well | Q6_K | 14 GB | ~63 tok/s |
| Qwen3.5 9B | Runs well | Q8_0 | 12 GB | ~63 tok/s |
| Qwen3 8B | Runs well | Q8_0 | 11 GB | ~72 tok/s |
| Llama 3.1 8B | Runs well | Q8_0 | 10 GB | ~74 tok/s |
| DeepSeek-R1 Distill 8B | Runs well | Q8_0 | 10 GB | ~74 tok/s |
| Qwen2.5-Coder 7B | Runs well | Q8_0 | 9.8 GB | ~77 tok/s |
| Mistral 7B v0.3 | Runs well | Q8_0 | 9.4 GB | ~82 tok/s |
| Qwen3.5 4B | Runs well | Q8_0 | 6.4 GB | ~130 tok/s |
| Gemma 3 4B | Runs well | Q8_0 | 6.9 GB | ~140 tok/s |
| Qwen3 4B | Runs well | Q8_0 | 5.8 GB | ~150 tok/s |
| Phi-4 Mini 3.8B | Runs well | Q8_0 | 5.2 GB | ~150 tok/s |
| Llama 3.2 3B | Runs well | Q8_0 | 4.6 GB | ~180 tok/s |
| Llama 3.2 1B | Runs well | Q8_0 | 1.6 GB | ~490 tok/s |
| Gemma 3 1B | Runs well | Q8_0 | 1.4 GB | ~590 tok/s |
| Qwen3 0.6B | Runs well | Q8_0 | 0.9 GB | ~980 tok/s |
| Codestral 22B | Offload | Q4_K_M | 16 GB | ~6.6 tok/s |
| Devstral Small 24B | Offload | Q4_K_M | 17 GB | ~5.9 tok/s |
| Mistral Small 3 24B | Offload | Q4_K_M | 18 GB | ~5.8 tok/s |
| Gemma 4 26B-A4B (MoE) | Offload | Q4_K_M | 19 GB | ~16 tok/s |
| Qwen3.6 27B | Offload | Q4_K_M | 19 GB | ~4.9 tok/s |
| Qwen3 30B-A3B (MoE) | Offload | Q4_K_M | 21 GB | ~17 tok/s |
| Qwen3-Coder 30B-A3B | Offload | Q4_K_M | 21 GB | ~17 tok/s |
| Gemma 3 27B | Offload | Q4_K_M | 22 GB | ~4.5 tok/s |
| Qwen3 32B | Offload | Q4_K_M | 23 GB | ~3.7 tok/s |
| DeepSeek-R1 Distill 32B | Offload | Q4_K_M | 23 GB | ~3.7 tok/s |
| Qwen2.5-Coder 32B | Offload | Q4_K_M | 23 GB | ~3.7 tok/s |
| Qwen3.6 35B-A3B (MoE) | Offload | Q4_K_M | 23 GB | ~17 tok/s |
| Gemma 4 31B | Offload | Q4_K_M | 25 GB | ~3.8 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 5080 is comfortable with models up to about 20B to 27B parameters. Gemma 4 26B-A4B (MoE) is the most useful step up beyond that: it needs about 19 GB, so part of it runs from system RAM, but as a mixture-of-experts model it stays reasonably quick at around 16 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 20 GB: RX 7900 XT. Other 16 GB cards: RTX 5070 Ti, RTX 5060 Ti, RTX 4080 / 4080 Super, RTX 4070 Ti Super, RTX 4060 Ti, RTX 4090 Laptop.
Compare the RTX 5080
Questions
What is the biggest model the RTX 5080 can run?
At 8K context the largest model that fits entirely in its 16 GB of VRAM is gpt-oss 20B at Q4_K_M, at roughly 130 tokens per second.
Can the RTX 5080 run 70B models like Llama 3.3 70B?
No. It needs about 48 GB at Q4, which is far beyond 16 GB even with system RAM offload.
What is the best model for coding on the RTX 5080?
gpt-oss 20B at Q4_K_M fits fully and runs at about 130 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.