NVIDIA A100 80GB SXM4 vs NVIDIA L40S, AI & Machine Learning Comparison
NVIDIA A100 80GB SXM4 wins 85 of 154 benchmarks, averaging 3.1% faster.
Both cards were measured first-party on our bench, same suite, same test rig.
What the numbers say
The gap is widest in MusicGen Small, where NVIDIA L40S leads by 228% (0.74 vs 2.43 x realtime); the closest fight is Phi-3.5-mini (0% apart); VRAM decides part of this one: NVIDIA A100 80GB SXM4 runs 155 of our 12 AI workloads while the other card runs 152, models that don't fit score zero.
Benchmark results head-to-head
| Benchmark | NVIDIA A100 80GB SXM4 | NVIDIA L40S | Difference |
|---|---|---|---|
| Qwen3 4B tok/s | 197 | 212.05 | -7% |
| Llama 3.1 8B tok/s | 162.57 | 135.51 | +20% |
| Qwen2.5-Coder 14B tok/s | 89.45 | 74.6 | +20% |
| Qwen3 32B tok/s | 45.53 | 34.44 | +32% |
| Llama 3.3 70B tok/s | 24.4 | 16.49 | +48% |
| Stable Diffusion XL images/min | 16.36 | 17.02 | -4% |
| Z-Image Turbo images/min | 9.975 | 8.325 | +20% |
| FLUX.1 dev images/min | 4.264 | 3.836 | +11% |
| FLUX.1 Kontext dev images/min | 1.993 | 1.693 | +18% |
| Qwen-Image-Edit images/min | 1.64 | 1.06 | +55% |
| LTX-Video (distilled) frames/s | 8.95 | 8.01 | +12% |
| Wan 2.2 5B (720p) frames/s | 0.66 | 0.49 | +35% |
| DeepSeek-R1 Distill Llama 8B tok/s | 159.37 | 135.55 | +18% |
| DeepSeek-R1 Distill 1.5B tok/s | 327.66 | 426.38 | -23% |
| DeepSeek-R1 Distill 14B tok/s | 87 | 74.6 | +17% |
| DeepSeek-R1 Distill 7B tok/s | 162.96 | 143.97 | +13% |
| Gemma 3 12B tok/s | 91.45 | 80.92 | +13% |
| Gemma 3 4B tok/s | 176.94 | 198.87 | -11% |
| Gemma 4 12B tok/s | 92.43 | 80.82 | +14% |
| Llama 3.2 1B tok/s | 525.32 | 618.87 | -15% |
| Llama 3.2 3B tok/s | 264.25 | 270.85 | -2% |
| Mistral 7B v0.3 tok/s | 171.92 | 144.74 | +19% |
| Mistral Small 24B tok/s | 61.38 | 48.43 | +27% |
| Phi-4 14B tok/s | 95.22 | 74.98 | +27% |
| Phi-4 Mini 3.8B tok/s | 239.17 | 228.93 | +4% |
| Qwen2.5-Coder 7B tok/s | 164.6 | 143.93 | +14% |
| Qwen3 0.6B tok/s | 428.39 | 655.75 | -35% |
| Qwen3 1.7B tok/s | 343.58 | 405.52 | -15% |
| Qwen3 14B tok/s | 90.64 | 75.12 | +21% |
| Qwen3 30B A3B tok/s | 177.32 | 213.66 | -17% |
| Qwen3 8B tok/s | 149.96 | 130.27 | +15% |
| SmolLM3 3B tok/s | 249.76 | 272.88 | -8% |
| TRELLIS Image-to-3D assets/hour | 488.7 | 776.1 | -37% |
| Codestral 22B tok/s | 63.26 | 50.06 | +26% |
| DeepSeek-R1 Distill 32B tok/s | 43.11 | 34.6 | +25% |
| Devstral Small 24B tok/s | 61.63 | 48.43 | +27% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 43.47 | 33.08 | +31% |
| Dolphin Mistral 24B Venice tok/s | 62.72 | 48.43 | +30% |
| Dolphin X1 8B tok/s | 158.62 | 135.48 | +17% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 158.63 | 135.55 | +17% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 61.17 | 48.4 | +26% |
| Gemma 3 27B tok/s | 48.43 | 38.48 | +26% |
| Qwen2.5-Coder 32B tok/s | 43.36 | 34.59 | +25% |
| Qwen3-Coder 30B A3B tok/s | 182.26 | 219.04 | -17% |
| QwQ 32B tok/s | 43.27 | 34.59 | +25% |
| StarCoder2 15B tok/s | 79.22 | 66.88 | +18% |
| FLUX.1 Schnell images/min | 28.39 | 26.24 | +8% |
| Z-Image Turbo (1024px) images/min | 18.65 | 18.2 | +2% |
| BiRefNet images/min | 916.67 | 863.81 | +6% |
| Depth Anything V2 Large images/min | 876.29 | 959.33 | -9% |
| Depth Anything V2 Small images/min | 925.15 | 970.61 | -5% |
| SAM ViT-Base images/min | 744.79 | 998.79 | -25% |
| SAM ViT-Huge images/min | 145.7 | 238.08 | -39% |
| Swin2SR 4x Upscaler images/min | 28.72 | 34.14 | -16% |
| Florence-2 Base images/min | 126.4 | 184.2 | -31% |
| Florence-2 Large images/min | 70.94 | 106.54 | -33% |
| Qwen2.5 1.5B LoRA train tok/s | 8252.8 | 12120.1 | -32% |
| Qwen2.5 7B LoRA train tok/s | 3900.6 | 3736 | +4% |
| SmolLM2 1.7B LoRA train tok/s | 9750.8 | 12052.1 | -19% |
| TinyLlama 1.1B LoRA train tok/s | 9553.3 | 16408.1 | -42% |
| Stable Video Diffusion XT clips/min | 1.536 | 1.166 | +32% |
| Wan 2.2 TI2V-5B (image to video) clips/min | 2.156 | 1.606 | +34% |
| CogVideoX-5B I2V clips/min | 0.468 | 0.383 | +22% |
| Qwen2.5 1.5B served serve tok/s | 4564.7 | 4686.8 | -3% |
| Qwen2.5 7B served serve tok/s | 2187.2 | 1406.8 | +55% |
| SmolLM2 1.7B served serve tok/s | 4351.4 | 3525.9 | +23% |
| TinyLlama 1.1B served serve tok/s | 5701.5 | 6278.4 | -9% |
| DeepSeek-R1-Distill-Llama-70B tok/s | 22.99 | 16.49 | +39% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 43.48 | 34.43 | +26% |
| Kokoro TTS 82M x realtime | 110.17 | 243.99 | -55% |
| Llama-3.3-70B-Instruct-abliterated tok/s | 22.92 | 16.49 | +39% |
| Meta-Llama-3.1-70B tok/s | 22.92 | 16.49 | +39% |
| MusicGen Small x realtime | 0.74 | 2.43 | -70% |
| Nemotron-3-Nano-30B-A3B tok/s | 204.06 | 190.96 | +7% |
| Whisper large-v3 x realtime | 73.83 | 193.2 | -62% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 31.99 | 41.06 | -22% |
| AI21-Jamba-Reasoning-3B tok/s | 239.23 | 258.41 | -7% |
| Olmo-3.1-32B-Think tok/s | 46.49 | 34.11 | +36% |
| Codestral 22B (Q3_K_M) tok/s | 49.52 | 59.53 | -17% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 65.77 | 48.43 | +36% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 164.46 | 134.78 | +22% |
| DeepSeek-Coder-V2-Lite tok/s | 213.91 | 257.64 | -17% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 152.76 | 130.23 | +17% |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 67.13 | 85.85 | -22% |
| dolphin-2.9-llama3-8b tok/s | 164.09 | 134.94 | +22% |
| Dolphin X1 Trinity Nano 6B tok/s | 158.41 | 246.45 | -36% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 90.35 | 74.63 | +21% |
| gemma-2-2b-it-abliterated tok/s | 244.41 | 278.14 | -12% |
| gemma-2-2b tok/s | 244.52 | 278.1 | -12% |
| gemma-2-9b tok/s | 108.68 | 88.87 | +22% |
| Gemma 3 12B (Q3_K_M) tok/s | 71.88 | 93.05 | -23% |
| gemma-3-1b tok/s | 324.08 | 426.49 | -24% |
| gemma-3-270m tok/s | 617.58 | 822.52 | -25% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 112.75 | 146.51 | -23% |
| GLM-4.7-Flash tok/s | 124.08 | 157.89 | -21% |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 152.29 | 130.25 | +17% |
| gpt-oss-20b tok/s | 212.63 | 233.82 | -9% |
| Hermes-3-Llama-3.2-3B tok/s | 269.13 | 265.87 | +1% |
| Hermes-4-70B tok/s | 24.47 | 16.45 | +49% |
| SmolLM3-3B tok/s | 254.6 | 272.96 | -7% |
| Qwen3-Coder-Next-abliterated tok/s | 119.56 | 0 | n/a |
| KAT-Coder-V2.5-Dev tok/s | 146.38 | 173.06 | -15% |
| L3-8B-Stheno-v3.2 tok/s | 164.51 | 135.53 | +21% |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| LFM2.5-1.2B tok/s | 576.95 | 661.54 | -13% |
| LFM2.5-8B-A1B tok/s | 369.01 | 394.48 | -6% |
| Llama-2-7B tok/s | 180.09 | 150.35 | +20% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 269.13 | 271.01 | -1% |
| Meta-Llama-3.1-8B tok/s | 164.24 | 135.46 | +21% |
| Phi-4-mini tok/s | 243.38 | 229.28 | +6% |
| Mistral-7B-Instruct-v0.1 tok/s | 173.87 | 144.82 | +20% |
| Mistral-7B-Instruct-v0.2 tok/s | 173.8 | 144.82 | +20% |
| Mistral-7B-Instruct-v0.3 tok/s | 173.27 | 144.68 | +20% |
| Mistral-Nemo-Instruct-2407 tok/s | 111.07 | 89.72 | +24% |
| Mistral Small 24B (Q3_K_M) tok/s | 46.63 | 58.21 | -20% |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| NemoMix-Unleashed-12B tok/s | 110.99 | 88.85 | +25% |
| Hermes-4-14B tok/s | 94.48 | 75.1 | +26% |
| Ornith-1.0-35B tok/s | 138.8 | 157.39 | -12% |
| Ornith-1.0-9B tok/s | 134.67 | 114.57 | +18% |
| phi-2 tok/s | 235.55 | 283.26 | -17% |
| Phi-3.5-mini tok/s | 229.46 | 229.06 | 0% |
| Phi-4 14B (Q3_K_M) tok/s | 82.68 | 89.18 | -7% |
| Qwen-AgentWorld-35B-A3B tok/s | 139.69 | 156.98 | -11% |
| Qwen3-0.6B tok/s | 432.66 | 656.26 | -34% |
| Qwen3-1.7B tok/s | 350.4 | 405.3 | -14% |
| Qwen3-14B tok/s | 94.35 | 75.01 | +26% |
| Qwen3-30B-A3B tok/s | 178.14 | 214.05 | -17% |
| Qwen3-4B-Instruct-2507 tok/s | 199.39 | 210.24 | -5% |
| Qwen3-8B tok/s | 152.8 | 130.28 | +17% |
| Qwen3-Coder-Next tok/s | 118.62 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 118.93 | 0 | n/a |
| Qwen1.5-0.5B tok/s | 540.32 | 774.78 | -30% |
| Qwen2-1.5B tok/s | 333.79 | 416.57 | -20% |
| Qwen2.5-0.5B tok/s | 571.49 | 738.06 | -23% |
| Qwen2.5-1.5B tok/s | 335.29 | 427.23 | -22% |
| Uncensored tok/s | 89.43 | 74.31 | +20% |
| Qwen2.5-14B tok/s | 90.46 | 74.62 | +21% |
| Qwen2.5-32B tok/s | 45.6 | 34.59 | +32% |
| Qwen2.5-3B tok/s | 248.47 | 270.04 | -8% |
| Qwen2.5-7B tok/s | 166.92 | 143.94 | +16% |
| Qwen2.5-Coder-0.5B tok/s | 562.59 | 715.81 | -21% |
| Qwen2.5-Coder-1.5B tok/s | 333.87 | 427.12 | -22% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 90.28 | 74.63 | +21% |
| Qwen2.5-Coder-3B tok/s | 248.7 | 269.9 | -8% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 166.84 | 142.97 | +17% |
| Qwen3 30B A3B (Q3_K_M) tok/s | 146.72 | 224.95 | -35% |
| Qwen3-4B-Instruct-2507 tok/s | 199.65 | 212.05 | -6% |
| Qwen3-4B-Thinking-2507 tok/s | 199.71 | 211.88 | -6% |
| Qwen3-Coder-Next tok/s | 120.71 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 120.98 | 0 | n/a |
| Qwen3-Next-80B-A3B tok/s | 117.44 | 0 | n/a |
| SmolLM2-135M tok/s | 554.44 | 916.56 | -40% |
| Cydonia-24B-v4.3 tok/s | 65.82 | 48.43 | +36% |
Whole-job comparison
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A100 80GB SXM4 is faster on 14 of 17; NVIDIA L40S on 3.
| Workflow | NVIDIA A100 80GB SXM4 | NVIDIA L40S | Difference | Cost per run |
|---|---|---|---|---|
| 50-image depth pass | 5 s | 6 s | NVIDIA A100 80GB SXM4 1.22x faster | $0.002 vs $0.001 |
| 30-minute podcast pass | 67 s | 75 s | NVIDIA A100 80GB SXM4 1.12x faster | $0.026 vs $0.017 |
| 500-image masking run | 3.5 min | 2.2 min | NVIDIA L40S 1.62x faster | $0.081 vs $0.028 |
| 24-frame storyboard | 3.7 min | 4.1 min | NVIDIA A100 80GB SXM4 1.10x faster | $0.085 vs $0.053 |
| 20-asset 3D game kit | 4 min | 2.9 min | NVIDIA L40S 1.36x faster | $0.092 vs $0.039 |
| 60-second AI short film | 4.7 min | 4.9 min | NVIDIA A100 80GB SXM4 1.06x faster | $0.108 vs $0.065 |
| 60-second AI short film, narrated | 4.9 min | 5.1 min | NVIDIA A100 80GB SXM4 1.03x faster | $0.114 vs $0.067 |
| 6-panel comic page | 6 min | 6.5 min | NVIDIA A100 80GB SXM4 1.09x faster | $0.139 vs $0.086 |
| 200-product catalogue cutout | 7.3 min | 6.2 min | NVIDIA L40S 1.17x faster | $0.169 vs $0.082 |
| Character sheet, 12 poses | 7.6 min | 8.4 min | NVIDIA A100 80GB SXM4 1.11x faster | $0.175 vs $0.110 |
| Full codebase review | 11.2 min | 13.5 min | NVIDIA A100 80GB SXM4 1.20x faster | $0.259 vs $0.177 |
| 10 short social clips | 15.1 min | 19.1 min | NVIDIA A100 80GB SXM4 1.26x faster | $0.350 vs $0.251 |
| 20 long-form articles | 19.1 min | 28.5 min | NVIDIA A100 80GB SXM4 1.49x faster | $0.443 vs $0.375 |
| 40-product photo shoot | 23.5 min | 26.7 min | NVIDIA A100 80GB SXM4 1.14x faster | $0.543 vs $0.352 |
| 40-product shoot, start to finish | 25 min | 28 min | NVIDIA A100 80GB SXM4 1.12x faster | $0.579 vs $0.369 |
| 100-photo restoration batch | 50.8 min | 59.6 min | NVIDIA A100 80GB SXM4 1.17x faster | $1.177 vs $0.784 |
| 100-photo restore and enlarge | 54.3 min | 62.5 min | NVIDIA A100 80GB SXM4 1.15x faster | $1.258 vs $0.823 |
Renting by the hour, NVIDIA L40S finishes 17 of 17 cheaper. The quicker card is not automatically the cheaper way to get the work done.
Cost to rent
| Card | Per hour |
|---|---|
| NVIDIA A100 80GB SXM4 | $1.390 |
| NVIDIA L40S | $0.790 |
NVIDIA L40S is 1.76x cheaper per hour. Cheapest on-demand rate we see across RunPod and Vast.
Specifications compared
| NVIDIA A100 80GB SXM4 | NVIDIA L40S | |
|---|---|---|
| VRAM | 80GB | 48GB |
| Architecture | Ampere | Ada Lovelace |
| Memory bandwidth | 2039 GB/s | 864 GB/s |
| Boost clock | 1,410 MHz | 2,520 MHz |
| TDP | 400 W | 350 W |
| Launch MSRP | $17,000 | $7,500 |
| Release | 2020-11-16 | 2023-08-08 |
FAQ
- Which is better for ai & machine learning: NVIDIA A100 80GB SXM4 or NVIDIA L40S?
- NVIDIA A100 80GB SXM4 performs better for ai & machine learning, winning 85 of 154 benchmarks in our suite with an average 3.1% advantage.
- What are the main hardware differences between NVIDIA A100 80GB SXM4 and NVIDIA L40S?
- NVIDIA A100 80GB SXM4 has 80GB VRAM and a 400W TDP, while NVIDIA L40S has 48GB VRAM and a 350W TDP.
- Does VRAM matter more than speed between NVIDIA A100 80GB SXM4 and NVIDIA L40S?
- For AI, yes, NVIDIA A100 80GB SXM4 fits 6 more of our 12 workloads than NVIDIA L40S. A model that exceeds VRAM doesn't run slower, it doesn't run at all, so the card that fits the model wins that workload outright.
- Where is the biggest performance difference between NVIDIA A100 80GB SXM4 and NVIDIA L40S?
- MusicGen Small: NVIDIA L40S leads by roughly 169% (0.74 vs 2.43 x realtime) in our testing.
NVIDIA A100 80GB SXM4 full review · NVIDIA L40S full review · All AI & Machine Learning rankings