Can deepseek coder 6.7b instruct run on the RTX 3080 10GB?
Yes. At 8K context the best fit is Q4_K_M, using about 8.3 GB of the 9.5 GB usable on the RTX 3080 10GB. Estimated speed ceiling: ~186 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | Max tok/s (est., 8K) |
|---|---|---|---|
| 2-bit (Q2_K) • | 5.1 GB Fits | 7.1 GB Fits | 269 |
| 3-bit (Q3_K_M) • | 5.5 GB Fits | 7.5 GB Fits | 230 |
| 4-bit (Q4_K_M) • | 6.3 GB Fits | 8.3 GB Fits | 186 |
| 5-bit (Q5_K_M) • | 6.9 GB Fits | 8.9 GB Tight | 159 |
| 6-bit (Q6_K) • | 7.6 GB Fits | 9.6 GB No | — |
| 8-bit (Q8_0) • | 9.1 GB Tight | 11.1 GB No | — |
| 16-bit (F16/BF16) | 15.0 GB No | 17.0 GB No | — |
• real GGUF file size from TheBloke/deepseek-coder-6.7B-instruct-GGUF. Speed is a bandwidth ceiling (760 GB/s), not a benchmark. Methodology.
More
deepseek coder 6.7b instruct on every GPU and Mac · Every model that runs on the RTX 3080 10GB · Other models that fit 12 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (deepseek-ai/deepseek-coder-6.7b-instruct), NVIDIA specs. Data as of 2026-11-17.