Models › SmolLM3-3B
SmolLM3-3B: VRAM needs and speed by GPU
At Q4_K_M, SmolLM3-3B needs about 2.8 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.
3.08B parameters · license: apache-2.0 · 65,536-token context · GGUF published Jul 8, 2025 · GGUF files on Hugging Face
Memory needed per quantization
| Quant | File size | Needs (estimate) | Smallest card it fits |
|---|---|---|---|
| Q4_K_M | 1.9 GB | 2.8 GB | 8 GB cards |
| Q8_0 | 3.3 GB | 4.2 GB | 8 GB cards |
| F16 | 6.2 GB | 7.0 GB | 8 GB cards |
Expected speed on popular hardware
| Hardware | Best quant that fits | Expected tok/s |
|---|---|---|
| NVIDIA RTX 3060 | Q8_0 | 94.1 estimate |
| NVIDIA RTX 4060 Ti | Q8_0 | 74.7 estimate |
| NVIDIA RTX 3090 | Q8_0 | 200 estimate |
| NVIDIA RTX 4090 | Q8_0 | 222 estimate |
| NVIDIA RTX 5090 | Q8_0 | 351 estimate |
| AMD Radeon RX 7900 XTX | Q8_0 | 223 estimate |
| Apple M4 Pro (20-core GPU) (64 GB) | Q8_0 | 70.7 estimate |
| Apple M3 Max (40-core GPU) (64 GB) | Q8_0 | 103 estimate |
| AMD Ryzen AI Max+ 395 (Radeon 8060S) | Q8_0 | 63.5 estimate |
Too big for your card?
Run SmolLM3-3B on a rented 24-80 GB GPU by the hour.
Rent a bigger GPU on RunPodAffiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.