

NVIDIA A40 wins 7 of 8 benchmarks, averaging 10.5% faster.
NVIDIA A40's numbers are anchored estimates calibrated against our measured cards, pending first-party measurement. Treat small gaps as ties.
The gap is widest in Llama 3.1 8B, where NVIDIA A40 leads by 13% (93.62 vs 106.18 tok/s); the closest fight is Qwen3 4B (5% apart); VRAM decides part of this one: NVIDIA A40 runs 8 of our 12 AI workloads while the other card runs 3, models that don't fit score zero.
| Benchmark | NVIDIA GeForce RTX 4070 Super | NVIDIA A40 | Difference |
|---|---|---|---|
| Qwen3 4B tok/s | 152.84 | 160.79 | -5% |
| Llama 3.1 8B tok/s | 93.62 | 106.18 | -12% |
| Qwen2.5-Coder 14B tok/s | 51.23 | 57.85 | -11% |
| Qwen3 32B tok/s | 0 | 26.89 | n/a |
| Llama 3.3 70B tok/s | 0 | 0 | n/a |
| Z-Image Turbo images/min | 0 | 3.225 | n/a |
| FLUX.1 dev images/min | 0 | 1.95 | n/a |
| Wan 2.2 5B (720p) frames/s | 0 | 0.31 | n/a |
How long each card takes to finish a complete pipeline, not just one model. NVIDIA GeForce RTX 4070 Super is faster on 0 of 1; NVIDIA A40 on 1.
| Workflow | NVIDIA GeForce RTX 4070 Super | NVIDIA A40 | Difference |
|---|---|---|---|
| Full codebase review | 19.5 min | 17.3 min | NVIDIA A40 1.13x faster |
| NVIDIA GeForce RTX 4070 Super | NVIDIA A40 | |
|---|---|---|
| VRAM | 12GB | 48GB |
| Architecture | Ada Lovelace | Ampere |
| Memory bandwidth | 504 GB/s | 696 GB/s |
| Boost clock | 2,475 MHz | 1,740 MHz |
| TDP | 220 W | 300 W |
| Launch MSRP | $599 | $5,000 |
| Release | 2024-01-17 | 2020-10-05 |
NVIDIA GeForce RTX 4070 Super full review · NVIDIA A40 full review · All AI & Machine Learning rankings