

NVIDIA L40 wins 10 of 11 benchmarks, averaging 67.9% faster.
NVIDIA L40's numbers are anchored estimates calibrated against our measured cards, pending first-party measurement. Treat small gaps as ties.
The gap is widest in Wan 2.2 5B (720p), where NVIDIA L40 leads by 106% (0.18 vs 0.37 frames/s); the closest fight is Qwen3-32B (55% apart); VRAM decides part of this one: NVIDIA L40 runs 137 of our 12 AI workloads while the other card runs 11, models that don't fit score zero.
| Benchmark | NVIDIA A10G | NVIDIA L40 | Difference |
|---|---|---|---|
| Qwen3-4B tok/s | 129.87 | 211.33 | -39% |
| Llama-3.1-8B tok/s | 86.62 | 135.12 | -36% |
| Qwen2.5-Coder-14B tok/s | 47.07 | 74.27 | -37% |
| Qwen3-32B tok/s | 22.04 | 34.08 | -35% |
| Llama 3.3 70B tok/s | 0 | 0 | n/a |
| Stable Diffusion XL images/min | 6.38 | 11.2 | -43% |
| Z-Image Turbo images/min | 3.45 | 5.625 | -39% |
| FLUX.1 dev images/min | 0 | 2.464 | n/a |
| FLUX.1 Kontext dev images/min | 0 | 1.114 | n/a |
| Qwen-Image-Edit images/min | 0 | 0.48 | n/a |
| Wan 2.2 5B (720p) frames/s | 0.18 | 0.37 | -51% |
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A10G is faster on 0 of 3; NVIDIA L40 on 3.
| Workflow | NVIDIA A10G | NVIDIA L40 | Difference |
|---|---|---|---|
| 24-frame storyboard | 9 min | 7 min | NVIDIA L40 1.28x faster |
| Full codebase review | 21.3 min | 13.5 min | NVIDIA L40 1.58x faster |
| 10 short social clips | 50.7 min | 28.2 min | NVIDIA L40 1.80x faster |
| NVIDIA A10G | NVIDIA L40 | |
|---|---|---|
| VRAM | 24GB | 48GB |
| Architecture | Ampere | Ada Lovelace |
| Memory bandwidth | 600 GB/s | 864 GB/s |
| Boost clock | 1,710 MHz | 2,490 MHz |
| TDP | 150 W | 300 W |
| Launch MSRP | $2,800 | $7,000 |
| Release | 2021-11-01 | 2022-10-13 |
NVIDIA A10G full review · NVIDIA L40 full review · All AI & Machine Learning rankings