Can Moonlight 16B A3B Instruct run on the H100 SXM?
Yes. At 8K context the best fit is F16, using about 30.7 GB of the 76.0 GB usable on the H100 SXM. Estimated speed ceiling: ~558 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | Max tok/s (est., 8K) |
|---|---|---|---|
| 2-bit (Q2_K) • | 7.0 GB Fits | 7.1 GB Fits | 2710 |
| 3-bit (Q3_K_M) • | 8.5 GB Fits | 8.7 GB Fits | 2151 |
| 4-bit (Q4_K_M) • | 10.6 GB Fits | 10.8 GB Fits | 1691 |
| 5-bit (Q5_K_M) • | 12.0 GB Fits | 12.2 GB Fits | 1480 |
| 6-bit (Q6_K) • | 14.1 GB Fits | 14.2 GB Fits | 1249 |
| 8-bit (Q8_0) • | 16.6 GB Fits | 16.8 GB Fits | 1050 |
| 16-bit (F16/BF16) | 30.6 GB Fits | 30.7 GB Fits | 558 |
• real GGUF file size from mmnga/Moonlight-16B-A3B-Instruct-gguf. Speed is a bandwidth ceiling (3350 GB/s), not a benchmark. Methodology.
More
Moonlight 16B A3B Instruct on every GPU and Mac · Every model that runs on the H100 SXM · Other models that fit 64 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (moonshotai/Moonlight-16B-A3B-Instruct), NVIDIA specs. Data as of 2026-11-17.