

NVIDIA A100 40GB SXM4 wins 10 of 16 benchmarks, averaging 3.3% faster.
Both cards' numbers are anchored estimates calibrated against our measured cards, pending first-party measurement. Treat small gaps as ties.
The gap is widest in Stable Diffusion XL, where NVIDIA A100 40GB SXM4 leads by 11% (16.8 vs 18.576 images/min); the closest fight is Qwen3 8B (0% apart); VRAM decides part of this one: NVIDIA A100 40GB PCIe runs 135 of our 12 AI workloads while the other card runs 18, models that don't fit score zero.
| Benchmark | NVIDIA A100 40GB PCIe | NVIDIA A100 40GB SXM4 | Difference |
|---|---|---|---|
| Qwen3-4B tok/s | 197.27 | 193.14 | +2% |
| Llama-3.1-8B tok/s | 154.22 | 157.26 | -2% |
| Qwen3 8B tok/s | 146.86 | 146.21 | 0% |
| Gemma 4 12B tok/s | 90.61 | 87.14 | +4% |
| Qwen2.5-Coder-14B tok/s | 84.11 | 86.2 | -2% |
| Qwen3 14B tok/s | 87.51 | 90.01 | -3% |
| Qwen3 30B A3B tok/s | 182.96 | 169.1 | +8% |
| Qwen3-32B tok/s | 41.5 | 43.8 | -5% |
| Llama 3.3 70B tok/s | 0 | 0 | n/a |
| Stable Diffusion XL images/min | 16.8 | 18.576 | -10% |
| FLUX.1 Kontext dev images/min | 1.843 | 1.993 | -8% |
| FLUX.1 dev images/min | 3.964 | 4.264 | -7% |
| Qwen-Image-Edit images/min | 0 | 0 | n/a |
| Z-Image Turbo images/min | 9.3 | 9.975 | -7% |
| LTX-Video (distilled) frames/s | 8.51 | 8.95 | -5% |
| Wan 2.2 5B (720p) frames/s | 0.61 | 0.66 | -8% |
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A100 40GB PCIe is faster on 0 of 8; NVIDIA A100 40GB SXM4 on 8.
| Workflow | NVIDIA A100 40GB PCIe | NVIDIA A100 40GB SXM4 | Difference | Cost per run |
|---|---|---|---|---|
| 24-frame storyboard | 3.2 min | 3 min | NVIDIA A100 40GB SXM4 1.05x faster | $0.027 vs $0.020 |
| 60-second AI short film | 4.2 min | 4 min | NVIDIA A100 40GB SXM4 1.05x faster | $0.035 vs $0.027 |
| 6-panel comic page | 5.1 min | 4.8 min | NVIDIA A100 40GB SXM4 1.07x faster | $0.043 vs $0.032 |
| Character sheet, 12 poses | 6.8 min | 6.3 min | NVIDIA A100 40GB SXM4 1.08x faster | $0.057 vs $0.042 |
| Full codebase review | 11.9 min | 11.7 min | NVIDIA A100 40GB SXM4 1.02x faster | $0.101 vs $0.078 |
| Short social clips | 14.7 min | 13.7 min | NVIDIA A100 40GB SXM4 1.08x faster | $0.125 vs $0.092 |
| Product photo shoot | 24.1 min | 22.2 min | NVIDIA A100 40GB SXM4 1.08x faster | $0.204 vs $0.149 |
| Photo restoration batch | 54.3 min | 50.2 min | NVIDIA A100 40GB SXM4 1.08x faster | $0.459 vs $0.337 |
Renting by the hour, NVIDIA A100 40GB SXM4 finishes 8 of 8 cheaper. The quicker card is not automatically the cheaper way to get the work done.
| Card | Per hour |
|---|---|
| NVIDIA A100 40GB PCIe | $0.508 |
| NVIDIA A100 40GB SXM4 | $0.403 |
NVIDIA A100 40GB SXM4 is 1.26x cheaper per hour. Cheapest on-demand rate we see across RunPod and Vast.
| NVIDIA A100 40GB PCIe | NVIDIA A100 40GB SXM4 | |
|---|---|---|
| VRAM | 40GB | 40GB |
| Transistors | 54,200M | 54,200M |
| Die size | 826 mm² | 826 mm² |
| Process node | 7 nm | 7 nm |
| Transistor density | 65.6 M/mm² | 65.6 M/mm² |
| Architecture | Ampere | Ampere |
| Memory bandwidth | 1555 GB/s | 1555 GB/s |
| Boost clock | 1,410 MHz | 1,410 MHz |
| TDP | 250 W | 400 W |
| Launch MSRP | $10,000 | $12,000 |
| Release | 2020-06-22 | 2020-05-14 |
NVIDIA A100 40GB PCIe full review · NVIDIA A100 40GB SXM4 full review · All AI & Machine Learning rankings