Our catalog lists 6 GPUs with 8 GB VRAM (RTX 3070 8GB, RTX 5060 8GB, RTX 5060 Ti 8GB, RTX 4060 Ti 8GB, RX 7600 8GB and 1 more). The ranking uses the RTX 5060 Ti 8GB — the highest-bandwidth 8 GB card at 448 GB/s — and shows the 17 of 61 catalog models that fully fit it at Q4 / 8K context (weights + KV cache + 1 GB overhead within 95% of VRAM), fastest first. Methodology.
| # | Model | Params (B) | Q4 tok/s | VRAM needed (GB) | Fit check |
|---|---|---|---|---|---|
| 1 | Qwen3-0.6B | 0.8 | 356.4 est. | 1.6 | fit check |
| 2 | gemma-3-1b-it | 1.0 | 285.1 est. | 1.75 | fit check |
| 3 | Llama-3.2-1B-Instruct | 1.2 | 237.6 est. | 1.9 | fit check |
| 4 | Qwen3-1.7B | 2.0 | 142.5 est. | 2.5 | fit check |
| 5 | Llama-3.2-3B-Instruct | 3.2 | 89.1 est. | 3.4 | fit check |
| 6 | Phi-4-mini-instruct | 3.8 | 75.0 est. | 3.85 | fit check |
| 7 | Phi-4-mini-reasoning | 3.8 | 75.0 est. | 3.85 | fit check |
| 8 | gemma-3-4b-it | 4.3 | 73.1 est. | 4.15 | fit check |
| 9 | Qwen3-4B-Instruct-2507 | 4.0 | 71.3 est. | 4.0 | fit check |
| 10 | Qwen3-4B-Thinking-2507 | 4.0 | 71.3 est. | 4.0 | fit check |
| 11 | Mistral-7B-Instruct-v0.3 | 7.2 | 39.6 est. | 6.4 | fit check |
| 12 | Qwen2.5-7B-Instruct | 7.6 | 37.5 est. | 6.7 | fit check |
| 13 | Qwen2.5-Coder-7B-Instruct | 7.6 | 37.5 est. | 6.7 | fit check |
| 14 | DeepSeek-R1-Distill-Qwen-7B | 7.6 | 37.5 est. | 6.7 | fit check |
| 15 | Llama-3.1-8B-Instruct | 8.0 | 35.6 est. | 7.0 | fit check |
| 16 | Qwen3-8B | 8.2 | 34.8 est. | 7.15 | fit check |
| 17 | DeepSeek-R1-0528-Qwen3-8B | 8.2 | 34.8 est. | 7.15 | fit check |