Can Kimi VL A3B Thinking run on the DGX Spark?
Yes. At 8K context the best fit is F16, using about 31.5 GB of the 109 GB usable on the DGX Spark. Estimated speed ceiling: ~46 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) | 6.6 GB Fits | 6.7 GB Fits | 7.4 GB Fits | 10.2 GB Fits | 243 |
| 3-bit (Q3_K_M) | 8.3 GB Fits | 8.4 GB Fits | 9.1 GB Fits | 12.0 GB Fits | 186 |
| 4-bit (Q4_K_M) | 10.2 GB Fits | 10.3 GB Fits | 11.0 GB Fits | 13.8 GB Fits | 149 |
| 5-bit (Q5_K_M) | 11.7 GB Fits | 11.8 GB Fits | 12.5 GB Fits | 15.4 GB Fits | 128 |
| 6-bit (Q6_K) | 13.4 GB Fits | 13.5 GB Fits | 14.2 GB Fits | 17.0 GB Fits | 111 |
| 8-bit (Q8_0) | 17.1 GB Fits | 17.2 GB Fits | 17.9 GB Fits | 20.7 GB Fits | 86 |
| 16-bit (F16/BF16) | 31.4 GB Fits | 31.5 GB Fits | 32.2 GB Fits | 35.1 GB Fits | 46 |
• real GGUF file size. Speed is a bandwidth ceiling (273 GB/s), not a benchmark. Methodology.
More
Kimi VL A3B Thinking 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 (moonshotai/Kimi-VL-A3B-Thinking), NVIDIA specs. Data as of 2026-11-17.