Consumer · GDDR6
RTX 4060 Ti 16 GB: which LLMs can it run?
On its own, it loads up to gpt-oss-20b in Q4. With several cards, memory adds up and opens bigger models.
Specifications
Memory16 GB
Bandwidth288 GB/s
FP16 compute~88 TF
Power165 W
Indicative price€450
Cloud rental€0.15/h
Against cards at the same price
Models and expected speed
| Model | Q4 | Q8 | FP16 |
|---|---|---|---|
| Qwen3 4B | ✓ ~62 tok/s | ✓ ~36 tok/s | ✓ ~19 tok/s |
| Llama 3.1 8B | ✓ ~32 tok/s | ✓ ~18 tok/s | 2 × · ~15 tok/s |
| Qwen3 8B | ✓ ~31 tok/s | ✓ ~18 tok/s | 2 × · ~14 tok/s |
| Gemma 3 12B | ✓ ~22 tok/s | ✓ ~12 tok/s | 2 × · ~9.8 tok/s |
| Mistral Nemo 12B | ✓ ~21 tok/s | ✓ ~12 tok/s | 2 × · ~9.7 tok/s |
| Phi-4 14B | ✓ ~18 tok/s | 2 × · ~15 tok/s | 3 × · ~12 tok/s |
| Qwen3 14B | ✓ ~18 tok/s | 2 × · ~15 tok/s | 3 × · ~12 tok/s |
| gpt-oss-20b | ✓ ~69 tok/s | 2 × · ~60 tok/s | 3 × · ~48 tok/s |
| Mistral Small 24B | 2 × · ~16 tok/s | 2 × · ~9.3 tok/s | 4 × · ~9.9 tok/s |
| Gemma 3 27B | 2 × · ~15 tok/s | 3 × · ~12 tok/s | 4 × · ~8.8 tok/s |
| Qwen3.8 27B | 2 × · ~14 tok/s | 3 × · ~12 tok/s | 4 × · ~8.5 tok/s |
| Qwen3 30B-A3B | 2 × · ~112 tok/s | 3 × · ~98 tok/s | ✗ |
| Qwen3 32B | 2 × · ~12 tok/s | 3 × · ~10 tok/s | ✗ |
| Qwen2.5 Coder 32B | 2 × · ~12 tok/s | 3 × · ~10 tok/s | ✗ |
| DeepSeek R1 Distill 32B | 2 × · ~12 tok/s | 3 × · ~10 tok/s | ✗ |
| Mixtral 8x7B | 2 × · ~30 tok/s | 4 × · ~34 tok/s | ✗ |
| Llama 3.3 70B | 4 × · ~11 tok/s | ✗ | ✗ |
| GLM-4.5-Air 106B | ✗ | ✗ | ✗ |
| gpt-oss-120b | ✗ | ✗ | ✗ |
| Mistral Large 123B | ✗ | ✗ | ✗ |
| Qwen3 235B-A22B | ✗ | ✗ | ✗ |
| GLM-4.5 355B | ✗ | ✗ | ✗ |
| Llama 3.1 405B | ✗ | ✗ | ✗ |
| DeepSeek R1 671B | ✗ | ✗ | ✗ |
| DeepSeek V3.1 671B | ✗ | ✗ | ✗ |
| Kimi K2 1T | ✗ | ✗ | ✗ |
8k tokens, one user. “2 ×” = number of cards needed, speed with tensor parallelism (vLLM). ✗ = doesn’t fit, even with 4 cards.