RTX 5080 vs RTX 4090 for local LLMs
VRAM decides which models fit entirely on the GPU; bandwidth decides how fast they generate. Here is how the two cards compare.
Which one for local LLMs?
The RTX 4090 has more VRAM (24 GB vs 16 GB), so it runs larger models entirely on the GPU.
The RTX 4090 has the higher memory bandwidth (1,008 GB/s vs 960 GB/s), so it should generate faster when a model fits on both.
RTX 5080 · Specifications
- VRAM
- 16 GB GDDR7
- Bandwidth
- 960 GB/s
- Board power
- 360 W
- Architecture
- Blackwell
- Largest model fully in VRAM
- gpt-oss-20b
Official specData updated 38 days agoSource: NVIDIA specifications
RTX 4090 · Specifications
- VRAM
- 24 GB GDDR6X
- Bandwidth
- 1,008 GB/s
- Board power
- 450 W
- Architecture
- Ada Lovelace
- Largest model fully in VRAM
- Qwen3 32B
Official specData updated 38 days agoSource: NVIDIA specifications
| Model | RTX 5080 · 16 GB | RTX 4090 · 24 GB |
|---|---|---|
| Qwen3 32B | CPU/GPU offload3–5.5 tok/s | Compatible33–45 tok/s |
| gpt-oss-20b | Excellent176–215 tok/s | Excellent212–243 tok/s |
| Qwen3 30B-A3B | CPU/GPU offload28–57 tok/s | Excellent125–209 tok/s |
| Llama 3.1 8B Instruct | Excellent108–156 tok/s | Excellent112–163 tok/s |
| Qwen3 14B | Excellent65–91 tok/s | Excellent68–95 tok/s |
| Qwen3 8B | Excellent103–150 tok/s | Excellent107–157 tok/s |
| Llama 3.3 70B Instruct | Not recommended0.7–1.3 tok/s | Not recommended0.9–1.7 tok/s |
| Gemma 3 27B IT | CPU/GPU offload6–12 tok/s | Excellent39–54 tok/s |
For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.
EstimateData updated 38 days agoSource: Run LLM Locally engine