Can Kimi VL A3B Instruct run on the H100 PCIe?
Yes. At 8K context the best fit is F16, using about 31.5 GB of the 76.0 GB usable on the H100 PCIe. Estimated speed ceiling: ~333 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) • | 7.0 GB Fits | 7.1 GB Fits | 7.8 GB Fits | 10.6 GB Fits | 1662 |
| 3-bit (Q3_K_M) • | 8.6 GB Fits | 8.7 GB Fits | 9.4 GB Fits | 12.2 GB Fits | 1319 |
| 4-bit (Q4_K_M) • | 10.6 GB Fits | 10.8 GB Fits | 11.5 GB Fits | 14.3 GB Fits | 1038 |
| 5-bit (Q5_K_M) • | 12.0 GB Fits | 12.2 GB Fits | 12.9 GB Fits | 15.7 GB Fits | 908 |
| 6-bit (Q6_K) • | 14.1 GB Fits | 14.2 GB Fits | 15.0 GB Fits | 17.8 GB Fits | 766 |
| 8-bit (Q8_0) • | 16.6 GB Fits | 16.8 GB Fits | 17.5 GB Fits | 20.3 GB Fits | 644 |
| 16-bit (F16/BF16) | 31.4 GB Fits | 31.5 GB Fits | 32.2 GB Fits | 35.1 GB Fits | 333 |
• real GGUF file size from mradermacher/Kimi-VL-A3B-Instruct-GGUF. Speed is a bandwidth ceiling (2000 GB/s), not a benchmark. Methodology.
More
Kimi VL A3B Instruct on every GPU and Mac · Every model that runs on the H100 PCIe · Other models that fit 64 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (moonshotai/Kimi-VL-A3B-Instruct), NVIDIA specs. Data as of 2026-11-17.