RTX 5070 Ti vs RTX 4080 SUPER 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?
Both cards have 16 GB of VRAM, so the same models fit on both.
The RTX 5070 Ti has the higher memory bandwidth (896 GB/s vs 736 GB/s), so it should generate faster when a model fits on both.
RTX 5070 Ti · Specifications
- VRAM
- 16 GB GDDR7
- Bandwidth
- 896 GB/s
- Board power
- 300 W
- Architecture
- Blackwell
- Largest model fully in VRAM
- gpt-oss-20b
Official specData updated 38 days agoSource: NVIDIA specifications
RTX 4080 SUPER · Specifications
- VRAM
- 16 GB GDDR6X
- Bandwidth
- 736 GB/s
- Board power
- 320 W
- Architecture
- Ada Lovelace
- Largest model fully in VRAM
- gpt-oss-20b
Official specData updated 38 days agoSource: NVIDIA specifications
| Model | RTX 5070 Ti · 16 GB | RTX 4080 SUPER · 16 GB |
|---|---|---|
| Qwen3 32B | CPU/GPU offload3–5.5 tok/s | CPU/GPU offload3–5.5 tok/s |
| gpt-oss-20b | Excellent163–199 tok/s | Excellent160–196 tok/s |
| Qwen3 30B-A3B | CPU/GPU offload28–56 tok/s | CPU/GPU offload27–54 tok/s |
| Llama 3.1 8B Instruct | Excellent102–147 tok/s | Excellent87–123 tok/s |
| Qwen3 14B | Excellent61–85 tok/s | Excellent52–71 tok/s |
| Qwen3 8B | Excellent97–141 tok/s | Excellent83–118 tok/s |
| Llama 3.3 70B Instruct | Not recommended0.7–1.3 tok/s | Not recommended0.7–1.3 tok/s |
| Gemma 3 27B IT | CPU/GPU offload6–12 tok/s | CPU/GPU offload6–11 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