NVIDIA A100 80GB SXM4 vs NVIDIA H100 80GB HBM3, AI & Machine Learning Comparison
NVIDIA H100 80GB HBM3 wins 154 of 157 benchmarks, averaging 68.9% 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 H100 80GB HBM3 leads by 234% (0.74 vs 2.47 x realtime); the closest fight is Depth Anything V2 Large (5% apart).
Benchmark results head-to-head
| Benchmark | NVIDIA A100 80GB SXM4 | NVIDIA H100 80GB HBM3 | Difference |
|---|---|---|---|
| Qwen3 4B tok/s | 197 | 310.26 | -37% |
| Llama 3.1 8B tok/s | 162.57 | 261.83 | -38% |
| Qwen2.5-Coder 14B tok/s | 89.45 | 144.84 | -38% |
| Qwen3 32B tok/s | 45.53 | 74.07 | -39% |
| Llama 3.3 70B tok/s | 24.4 | 41 | -40% |
| Stable Diffusion XL images/min | 16.36 | 34.58 | -53% |
| Z-Image Turbo images/min | 9.975 | 22.125 | -55% |
| FLUX.1 dev images/min | 4.264 | 9.064 | -53% |
| FLUX.1 Kontext dev images/min | 1.993 | 4.264 | -53% |
| Qwen-Image-Edit images/min | 1.64 | 3.68 | -55% |
| LTX-Video (distilled) frames/s | 8.95 | 17.47 | -49% |
| Wan 2.2 5B (720p) frames/s | 0.66 | 1.33 | -50% |
| DeepSeek-R1 Distill Llama 8B tok/s | 159.37 | 261.15 | -39% |
| DeepSeek-R1 Distill 1.5B tok/s | 327.66 | 537.32 | -39% |
| DeepSeek-R1 Distill 14B tok/s | 87 | 144.82 | -40% |
| DeepSeek-R1 Distill 7B tok/s | 162.96 | 264.42 | -38% |
| Gemma 3 12B tok/s | 91.45 | 150.58 | -39% |
| Gemma 3 4B tok/s | 176.94 | 289.61 | -39% |
| Gemma 4 12B tok/s | 92.43 | 148.17 | -38% |
| Llama 3.2 1B tok/s | 525.32 | 880.64 | -40% |
| Llama 3.2 3B tok/s | 264.25 | 422.87 | -38% |
| Mistral 7B v0.3 tok/s | 171.92 | 275.85 | -38% |
| Mistral Small 24B tok/s | 61.38 | 105.62 | -42% |
| Phi-4 14B tok/s | 95.22 | 165.17 | -42% |
| Phi-4 Mini 3.8B tok/s | 239.17 | 388.86 | -38% |
| Qwen2.5-Coder 7B tok/s | 164.6 | 263.91 | -38% |
| Qwen3 0.6B tok/s | 428.39 | 713.53 | -40% |
| Qwen3 1.7B tok/s | 343.58 | 556.48 | -38% |
| Qwen3 14B tok/s | 90.64 | 151.16 | -40% |
| Qwen3 30B A3B tok/s | 177.32 | 283.78 | -38% |
| Qwen3 8B tok/s | 149.96 | 244.22 | -39% |
| SmolLM3 3B tok/s | 249.76 | 397.99 | -37% |
| TRELLIS Image-to-3D assets/hour | 488.7 | 864 | -43% |
| Codestral 22B tok/s | 63.26 | 110.05 | -43% |
| DeepSeek-R1 Distill 32B tok/s | 43.11 | 75.8 | -43% |
| Devstral Small 24B tok/s | 61.63 | 107.79 | -43% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 43.47 | 76.76 | -43% |
| Dolphin Mistral 24B Venice tok/s | 62.72 | 109.13 | -43% |
| Dolphin X1 8B tok/s | 158.62 | 266 | -40% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 158.63 | 266.19 | -40% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 61.17 | 107.83 | -43% |
| Gemma 3 27B tok/s | 48.43 | 84.77 | -43% |
| Qwen2.5-Coder 32B tok/s | 43.36 | 75.81 | -43% |
| Qwen3-Coder 30B A3B tok/s | 182.26 | 296.22 | -38% |
| QwQ 32B tok/s | 43.27 | 75.82 | -43% |
| StarCoder2 15B tok/s | 79.22 | 134.92 | -41% |
| FLUX.1 Schnell images/min | 28.39 | 58.36 | -51% |
| Z-Image Turbo (1024px) images/min | 18.65 | 39.42 | -53% |
| Krea 2 Turbo images/min | 3.35 | 6.85 | -51% |
| TRELLIS.2 Image-to-3D (1536³ max quality) assets/hour | 32.7 | 62.7 | -48% |
| BiRefNet images/min | 916.67 | 1310.97 | -30% |
| Depth Anything V2 Large images/min | 876.29 | 918.71 | -5% |
| Depth Anything V2 Small images/min | 925.15 | 1081.54 | -14% |
| SAM ViT-Base images/min | 744.79 | 1431.27 | -48% |
| SAM ViT-Huge images/min | 145.7 | 320.64 | -55% |
| Swin2SR 4x Upscaler images/min | 28.72 | 25.48 | +13% |
| Florence-2 Base images/min | 126.4 | 249.62 | -49% |
| Florence-2 Large images/min | 70.94 | 156.15 | -55% |
| Qwen2.5 1.5B LoRA train tok/s | 8252.8 | 16034.9 | -49% |
| Qwen2.5 7B LoRA train tok/s | 3900.6 | 8514.3 | -54% |
| SmolLM2 1.7B LoRA train tok/s | 9750.8 | 18910.3 | -48% |
| TinyLlama 1.1B LoRA train tok/s | 9553.3 | 17522 | -45% |
| Stable Video Diffusion XT clips/min | 1.536 | 2.882 | -47% |
| Wan 2.2 TI2V-5B (image to video) clips/min | 2.156 | 4.126 | -48% |
| CogVideoX-5B I2V clips/min | 0.468 | 0.879 | -47% |
| Qwen2.5 1.5B served serve tok/s | 4564.7 | 6794.8 | -33% |
| Qwen2.5 7B served serve tok/s | 2187.2 | 3601.1 | -39% |
| SmolLM2 1.7B served serve tok/s | 4351.4 | 6509.9 | -33% |
| TinyLlama 1.1B served serve tok/s | 5701.5 | 8336.7 | -32% |
| DeepSeek-R1-Distill-Llama-70B tok/s | 22.99 | 41.17 | -44% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 43.48 | 73.48 | -41% |
| Kokoro TTS 82M x realtime | 110.17 | 222.02 | -50% |
| Llama-3.3-70B-Instruct-abliterated tok/s | 22.92 | 41.15 | -44% |
| Meta-Llama-3.1-70B tok/s | 22.92 | 41.19 | -44% |
| MusicGen Small x realtime | 0.74 | 2.47 | -70% |
| Nemotron-3-Nano-30B-A3B tok/s | 204.06 | 316.51 | -36% |
| Whisper large-v3 x realtime | 73.83 | 181.47 | -59% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 31.99 | 58.05 | -45% |
| AI21-Jamba-Reasoning-3B tok/s | 239.23 | 363.09 | -34% |
| Olmo-3.1-32B-Think tok/s | 46.49 | 75.53 | -38% |
| Codestral 22B (Q3_K_M) tok/s | 49.52 | 87.21 | -43% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 65.77 | 105.63 | -38% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 164.46 | 262.76 | -37% |
| DeepSeek-Coder-V2-Lite tok/s | 213.91 | 309.14 | -31% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 152.76 | 245.26 | -38% |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 67.13 | 118.43 | -43% |
| dolphin-2.9-llama3-8b tok/s | 164.09 | 262.47 | -37% |
| Dolphin X1 Trinity Nano 6B tok/s | 158.41 | 247.94 | -36% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 90.35 | 145.15 | -38% |
| gemma-2-2b-it-abliterated tok/s | 244.41 | 392.65 | -38% |
| gemma-2-2b tok/s | 244.52 | 392.26 | -38% |
| gemma-2-9b tok/s | 108.68 | 174.7 | -38% |
| Gemma 3 12B (Q3_K_M) tok/s | 71.88 | 125.92 | -43% |
| gemma-3-1b tok/s | 324.08 | 504.17 | -36% |
| gemma-3-270m tok/s | 617.58 | 961.29 | -36% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 112.75 | 168.41 | -33% |
| GLM-4.7-Flash tok/s | 124.08 | 186.05 | -33% |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 152.29 | 245 | -38% |
| gpt-oss-20b tok/s | 212.63 | 346.8 | -39% |
| Hermes-3-Llama-3.2-3B tok/s | 269.13 | 423.72 | -36% |
| Hermes-4-70B tok/s | 24.47 | 41.2 | -41% |
| SmolLM3-3B tok/s | 254.6 | 399.14 | -36% |
| Qwen3-Coder-Next-abliterated tok/s | 119.56 | 190.32 | -37% |
| KAT-Coder-V2.5-Dev tok/s | 146.38 | 231.83 | -37% |
| L3-8B-Stheno-v3.2 tok/s | 164.51 | 261.83 | -37% |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| LFM2.5-1.2B tok/s | 576.95 | 926.61 | -38% |
| LFM2.5-8B-A1B tok/s | 369.01 | 581.98 | -37% |
| Llama-2-7B tok/s | 180.09 | 284.62 | -37% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 269.13 | 424.25 | -37% |
| Meta-Llama-3.1-8B tok/s | 164.24 | 261.7 | -37% |
| Phi-4-mini tok/s | 243.38 | 381.52 | -36% |
| Mistral-7B-Instruct-v0.1 tok/s | 173.87 | 276.29 | -37% |
| Mistral-7B-Instruct-v0.2 tok/s | 173.8 | 276.73 | -37% |
| Mistral-7B-Instruct-v0.3 tok/s | 173.27 | 276.61 | -37% |
| Mistral-Nemo-Instruct-2407 tok/s | 111.07 | 177.38 | -37% |
| Mistral Small 24B (Q3_K_M) tok/s | 46.63 | 84 | -44% |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| NemoMix-Unleashed-12B tok/s | 110.99 | 177.28 | -37% |
| Hermes-4-14B tok/s | 94.48 | 152 | -38% |
| Ornith-1.0-35B tok/s | 138.8 | 221.65 | -37% |
| Ornith-1.0-9B tok/s | 134.67 | 216.89 | -38% |
| phi-2 tok/s | 235.55 | 340.09 | -31% |
| Phi-3.5-mini tok/s | 229.46 | 345.07 | -34% |
| Phi-4 14B (Q3_K_M) tok/s | 82.68 | 138.87 | -40% |
| Qwen-AgentWorld-35B-A3B tok/s | 139.69 | 219.1 | -36% |
| Qwen3-0.6B tok/s | 432.66 | 710.97 | -39% |
| Qwen3-1.7B tok/s | 350.4 | 557.09 | -37% |
| Qwen3-14B tok/s | 94.35 | 152.2 | -38% |
| Qwen3-30B-A3B tok/s | 178.14 | 285.44 | -38% |
| Qwen3-4B-Instruct-2507 tok/s | 199.39 | 310.58 | -36% |
| Qwen3-8B tok/s | 152.8 | 244.2 | -37% |
| Qwen3-Coder-Next tok/s | 118.62 | 188.21 | -37% |
| Qwen3-Next-80B-A3B-Thinking tok/s | 118.93 | 188.14 | -37% |
| Qwen1.5-0.5B tok/s | 540.32 | 847.02 | -36% |
| Qwen2-1.5B tok/s | 333.79 | 536.52 | -38% |
| Qwen2.5-0.5B tok/s | 571.49 | 890 | -36% |
| Qwen2.5-1.5B tok/s | 335.29 | 537.24 | -38% |
| Uncensored tok/s | 89.43 | 144.92 | -38% |
| Qwen2.5-14B tok/s | 90.46 | 144.7 | -37% |
| Qwen2.5-32B tok/s | 45.6 | 73.5 | -38% |
| Qwen2.5-3B tok/s | 248.47 | 395.52 | -37% |
| Qwen2.5-72B tok/s | 23.95 | 40.78 | -41% |
| Qwen2.5-7B tok/s | 166.92 | 264.17 | -37% |
| Qwen2.5-Coder-0.5B tok/s | 562.59 | 892.81 | -37% |
| Qwen2.5-Coder-1.5B tok/s | 333.87 | 538.44 | -38% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 90.28 | 145.01 | -38% |
| Qwen2.5-Coder-3B tok/s | 248.7 | 394 | -37% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 166.84 | 264.24 | -37% |
| Qwen3 30B A3B (Q3_K_M) tok/s | 146.72 | 242.6 | -40% |
| Qwen3-4B-Instruct-2507 tok/s | 199.65 | 311.47 | -36% |
| Qwen3-4B-Thinking-2507 tok/s | 199.71 | 310.84 | -36% |
| Qwen3-Coder-Next tok/s | 120.71 | 193.79 | -38% |
| Qwen3-Next-80B-A3B-Thinking tok/s | 120.98 | 191.08 | -37% |
| Qwen3-Next-80B-A3B tok/s | 117.44 | 186.22 | -37% |
| SmolLM2-135M tok/s | 554.44 | 898.53 | -38% |
| Cydonia-24B-v4.3 tok/s | 65.82 | 105.49 | -38% |
Whole-job comparison
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A100 80GB SXM4 is faster on 2 of 17; NVIDIA H100 80GB HBM3 on 15.
| Workflow | NVIDIA A100 80GB SXM4 | NVIDIA H100 80GB HBM3 | Difference | Cost per run |
|---|---|---|---|---|
| 50-image depth pass | 5 s | 5 s | NVIDIA A100 80GB SXM4 1.10x faster | $0.002 vs $0.002 |
| 30-minute podcast pass | 67 s | 34 s | NVIDIA H100 80GB HBM3 1.96x faster | $0.026 vs $0.013 |
| 500-image masking run | 3.5 min | 1.6 min | NVIDIA H100 80GB HBM3 2.16x faster | $0.081 vs $0.036 |
| 24-frame storyboard | 3.7 min | 1.8 min | NVIDIA H100 80GB HBM3 2.02x faster | $0.085 vs $0.040 |
| 20-asset 3D game kit | 4 min | 2.2 min | NVIDIA H100 80GB HBM3 1.83x faster | $0.092 vs $0.049 |
| 60-second AI short film | 4.7 min | 2.5 min | NVIDIA H100 80GB HBM3 1.88x faster | $0.108 vs $0.055 |
| 60-second AI short film, narrated | 4.9 min | 2.6 min | NVIDIA H100 80GB HBM3 1.87x faster | $0.114 vs $0.059 |
| 6-panel comic page | 6 min | 3 min | NVIDIA H100 80GB HBM3 2.01x faster | $0.139 vs $0.066 |
| 200-product catalogue cutout | 7.3 min | 8.1 min | NVIDIA A100 80GB SXM4 1.12x faster | $0.169 vs $0.181 |
| Character sheet, 12 poses | 7.6 min | 3.7 min | NVIDIA H100 80GB HBM3 2.06x faster | $0.175 vs $0.082 |
| Full codebase review | 11.2 min | 6.9 min | NVIDIA H100 80GB HBM3 1.62x faster | $0.259 vs $0.154 |
| 10 short social clips | 15.1 min | 8 min | NVIDIA H100 80GB HBM3 1.89x faster | $0.350 vs $0.178 |
| 20 long-form articles | 19.1 min | 11.4 min | NVIDIA H100 80GB HBM3 1.68x faster | $0.443 vs $0.253 |
| 40-product photo shoot | 23.5 min | 11.1 min | NVIDIA H100 80GB HBM3 2.11x faster | $0.543 vs $0.248 |
| 40-product shoot, start to finish | 25 min | 12.9 min | NVIDIA H100 80GB HBM3 1.94x faster | $0.579 vs $0.287 |
| 100-photo restoration batch | 50.8 min | 23.8 min | NVIDIA H100 80GB HBM3 2.13x faster | $1.177 vs $0.531 |
| 100-photo restore and enlarge | 54.3 min | 27.8 min | NVIDIA H100 80GB HBM3 1.96x faster | $1.258 vs $0.618 |
Renting by the hour, NVIDIA H100 80GB HBM3 finishes 15 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 H100 80GB HBM3 | $1.336 |
NVIDIA H100 80GB HBM3 is 1.04x cheaper per hour. Cheapest on-demand rate we see across RunPod and Vast.
Specifications compared
| NVIDIA A100 80GB SXM4 | NVIDIA H100 80GB HBM3 | |
|---|---|---|
| VRAM | 80GB | 80GB |
| Architecture | Ampere | Hopper |
| Memory bandwidth | 2039 GB/s | 3350 GB/s |
| Boost clock | 1,410 MHz | 1,980 MHz |
| TDP | 400 W | 700 W |
| Launch MSRP | $17,000 | $30,000 |
| Release | 2020-11-16 | 2022-09-20 |
FAQ
- Which is better for ai & machine learning: NVIDIA A100 80GB SXM4 or NVIDIA H100 80GB HBM3?
- NVIDIA H100 80GB HBM3 performs better for ai & machine learning, winning 154 of 157 benchmarks in our suite with an average 68.9% advantage.
- What are the main hardware differences between NVIDIA A100 80GB SXM4 and NVIDIA H100 80GB HBM3?
- NVIDIA A100 80GB SXM4 has 80GB VRAM and a 400W TDP, while NVIDIA H100 80GB HBM3 has 80GB VRAM and a 700W TDP.
- Where is the biggest performance difference between NVIDIA A100 80GB SXM4 and NVIDIA H100 80GB HBM3?
- MusicGen Small: NVIDIA H100 80GB HBM3 leads by roughly 173% (0.74 vs 2.47 x realtime) in our testing.
NVIDIA A100 80GB SXM4 full review · NVIDIA H100 80GB HBM3 full review · All AI & Machine Learning rankings