RunsOnMy ModelsGPUs & MacsBy VRAMAPI pricesVRAM calculatorNew

Can LightOnOCR 3 4B run on the H100 SXM?

Yes. At 8K context the best fit is F16, using about 9.0 GB of the 76.0 GB usable on the H100 SXM. Estimated speed ceiling: ~398 tokens/sec.

Memory by quant and context

Quant4K ctx8K ctx32K ctx128K ctxMax tok/s (est., 8K)
2-bit (Q2_K)2.6 GB Fits2.7 GB Fits3.5 GB Fits6.5 GB Fits1968
3-bit (Q3_K_M) •3.1 GB Fits3.2 GB Fits4.0 GB Fits7.0 GB Fits1481
4-bit (Q4_K_M) •3.5 GB Fits3.6 GB Fits4.4 GB Fits7.4 GB Fits1237
5-bit (Q5_K_M) •3.9 GB Fits4.0 GB Fits4.7 GB Fits7.7 GB Fits1089
6-bit (Q6_K) •4.2 GB Fits4.3 GB Fits5.1 GB Fits8.1 GB Fits967
8-bit (Q8_0) •5.2 GB Fits5.3 GB Fits6.0 GB Fits9.0 GB Fits747
16-bit (F16/BF16) •8.8 GB Fits9.0 GB Fits9.7 GB Fits12.7 GB Fits398

• real GGUF file size from prithivMLmods/LightOnOCR-3-4B-GGUF. Speed is a bandwidth ceiling (3350 GB/s), not a benchmark. Methodology.

More

LightOnOCR 3 4B 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 (lightonai/LightOnOCR-3-4B), NVIDIA specs. Data as of 2026-11-17.