Can DeepSeek Coder V2 Lite Instruct run on the DGX Spark?
Yes. At 8K context the best fit is F16, using about 30.1 GB of the 109 GB usable on the DGX Spark. Estimated speed ceiling: ~51 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 6.2 GB Fits | 6.3 GB Fits | 7.0 GB Fits | 9.9 GB Fits | 274 |
| 3-bit (Q3_K_M) | 7.9 GB Fits | 8.0 GB Fits | 8.7 GB Fits | 11.5 GB Fits | 210 |
| 4-bit (Q4_K_M) • | 10.4 GB Fits | 10.5 GB Fits | 11.2 GB Fits | 14.0 GB Fits | 155 |
| 5-bit (Q5_K_M) • | 11.8 GB Fits | 11.9 GB Fits | 12.6 GB Fits | 15.4 GB Fits | 136 |
| 6-bit (Q6_K) • | 13.8 GB Fits | 13.9 GB Fits | 14.6 GB Fits | 17.5 GB Fits | 115 |
| 8-bit (Q8_0) • | 16.3 GB Fits | 16.4 GB Fits | 17.1 GB Fits | 19.9 GB Fits | 96 |
| 16-bit (F16/BF16) | 30.0 GB Fits | 30.1 GB Fits | 30.8 GB Fits | 33.6 GB Fits | 51 |
• real GGUF file size from lmstudio-community/DeepSeek-Coder-V2-Lite-Instruct-GGUF. Speed is a bandwidth ceiling (273 GB/s), not a benchmark. Methodology.
More
DeepSeek Coder V2 Lite Instruct on every GPU and Mac · Every model that runs on the DGX Spark · Other models that fit 64 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct), NVIDIA specs. Data as of 2026-11-17.