NVIDIA GeForce RTX 5070 × Qwen3 8B
Can the RTX 5070 run Qwen3 8B?
Yes
Yes. The RTX 5070 can run Qwen3 8B at up to Q8_0 (needs ~11.0 GB of 11.4 GB usable), at an estimated ~98.1 tokens/sec at Q4_K_M.
Speed is estimated from memory bandwidth calibrated with a measured 127.54 tok/s Llama 2 7B Q4_0 result (source). No direct public benchmark of this exact pair is recorded yet.
By quantization (8K context)
| Quant | Memory needed | Fits? | Est. tok/s |
|---|---|---|---|
| Q3_K_M | 6.1 GB | ✅ | ~122 |
| Q4_K_M | 7.1 GB | ✅ | ~98.1 |
| Q5_K_M | 8.0 GB | ✅ | ~83.4 |
| Q6_K | 9.0 GB | ✅ | ~72.1 |
| Q8_0 | 11.0 GB | ✅ | ~56.0 |
| F16 | 18.9 GB | ❌ | — |
Fastest hardware for Qwen3 8B
- NVIDIA GeForce RTX 5090 — ~223 tok/s at Q4_K_M
- NVIDIA GeForce RTX 4090 — ~143 tok/s at Q4_K_M
- AMD Radeon RX 7900 XTX — ~142 tok/s at Q4_K_M
- NVIDIA GeForce RTX 5080 — ~140 tok/s at Q4_K_M
- NVIDIA GeForce RTX 5070 Ti — ~136 tok/s at Q4_K_M
No suitable GPU? Rent one by the hour on Vast.ai or RunPod (affiliate links).
Get new models & benchmarks in your inbox
No spam. Unsubscribe anytime by replying to any email.