Can DeepSeek OCR 2 run on the L40S?
Yes. At 8K context the best fit is F16, using about 6.6 GB of the 45.6 GB usable on the L40S. Estimated speed ceiling: ~417 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | Max tok/s (est., 8K) |
|---|---|---|---|
| 2-bit (Q2_K) | 2.1 GB Fits | 2.3 GB Fits | 1930 |
| 3-bit (Q3_K_M) | 2.4 GB Fits | 2.7 GB Fits | 1481 |
| 4-bit (Q4_K_M) • | 2.7 GB Fits | 2.9 GB Fits | 1257 |
| 5-bit (Q5_K_M) | 3.1 GB Fits | 3.4 GB Fits | 1016 |
| 6-bit (Q6_K) | 3.5 GB Fits | 3.7 GB Fits | 882 |
| 8-bit (Q8_0) • | 3.8 GB Fits | 4.0 GB Fits | 784 |
| 16-bit (F16/BF16) • | 6.4 GB Fits | 6.6 GB Fits | 417 |
• real GGUF file size from sabafallah/DeepSeek-OCR-2-GGUF. Speed is a bandwidth ceiling (864 GB/s), not a benchmark. Methodology.
More
DeepSeek OCR 2 on every GPU and Mac · Every model that runs on the L40S · Other models that fit 48 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (deepseek-ai/DeepSeek-OCR-2), NVIDIA specs. Data as of 2026-11-17.