Can gemma 4 31B it run on the L40S?
Yes. At 8K context the best fit is Q8_0, using about 33.3 GB of the 45.6 GB usable on the L40S. Estimated speed ceiling: ~26 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) | 13.2 GB Fits | 13.9 GB Fits | 17.6 GB Fits | 32.6 GB Fits | 74 |
| 3-bit (Q3_K_M) • | 16.0 GB Fits | 16.7 GB Fits | 20.4 GB Fits | 35.4 GB Fits | 59 |
| 4-bit (Q4_K_M) • | 19.4 GB Fits | 20.0 GB Fits | 23.7 GB Fits | 38.7 GB Fits | 47 |
| 5-bit (Q5_K_M) • | 22.5 GB Fits | 23.1 GB Fits | 26.9 GB Fits | 41.9 GB Tight | 40 |
| 6-bit (Q6_K) • | 25.8 GB Fits | 26.4 GB Fits | 30.2 GB Fits | 45.2 GB Tight | 34 |
| 8-bit (Q8_0) • | 32.7 GB Fits | 33.3 GB Fits | 37.1 GB Fits | 52.1 GB No | 26 |
| 16-bit (F16/BF16) • | 59.5 GB No | 60.1 GB No | 63.9 GB No | 78.9 GB No | — |
• real GGUF file size from unsloth/gemma-4-31B-it-GGUF. Speed is a bandwidth ceiling (864 GB/s), not a benchmark. Methodology.
More
gemma 4 31B it 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 (google/gemma-4-31B-it), NVIDIA specs. Data as of 2026-11-17.