NVIDIA A10G vs NVIDIA H100 80GB HBM3, AI & Machine Learning Comparison
NVIDIA H100 80GB HBM3 wins 152 of 155 benchmarks, averaging 179.4% faster.
Both cards were measured first-party on our bench, same suite, same test rig.
What the numbers say
The gap is widest in Wan 2.2 5B (720p), where NVIDIA H100 80GB HBM3 leads by 639% (0.18 vs 1.33 frames/s); the closest fight is Swin2SR 4x Upscaler (14% apart); VRAM decides part of this one: NVIDIA H100 80GB HBM3 runs 156 of our 12 AI workloads while the other card runs 137, models that don't fit score zero.
Benchmark results head-to-head
| Benchmark | NVIDIA A10G | NVIDIA H100 80GB HBM3 | Difference |
|---|---|---|---|
| Qwen3-4B tok/s | 129.87 | 310.26 | -58% |
| Llama-3.1-8B tok/s | 86.62 | 261.83 | -67% |
| Qwen2.5-Coder-14B tok/s | 47.07 | 144.84 | -68% |
| Qwen3-32B tok/s | 22.04 | 74.07 | -70% |
| Llama 3.3 70B tok/s | 0 | 41 | n/a |
| Stable Diffusion XL images/min | 6.38 | 34.58 | -82% |
| Z-Image Turbo images/min | 3.45 | 22.125 | -84% |
| FLUX.1 dev images/min | 0 | 9.064 | n/a |
| FLUX.1 Kontext dev images/min | 0 | 4.264 | n/a |
| Qwen-Image-Edit images/min | 0 | 3.68 | n/a |
| Wan 2.2 5B (720p) frames/s | 0.18 | 1.33 | -86% |
| Llama 3.2 1B tok/s | 400.6 | 880.64 | -55% |
| Qwen3 0.6B tok/s | 451.38 | 713.53 | -37% |
| Qwen3 1.7B tok/s | 267.08 | 556.48 | -52% |
| DeepSeek-R1 Distill 1.5B tok/s | 265.75 | 537.32 | -51% |
| Gemma 3 4B tok/s | 124.85 | 289.61 | -57% |
| Llama 3.2 3B tok/s | 170.47 | 422.87 | -60% |
| Mistral 7B v0.3 tok/s | 92.82 | 275.85 | -66% |
| Phi-4 Mini 3.8B tok/s | 144.42 | 388.86 | -63% |
| Qwen2.5-Coder 7B tok/s | 90.05 | 263.91 | -66% |
| SmolLM3 3B tok/s | 172.15 | 397.99 | -57% |
| DeepSeek-R1 Distill Llama 8B tok/s | 86.83 | 261.15 | -67% |
| DeepSeek-R1 Distill 14B tok/s | 46.96 | 144.82 | -68% |
| DeepSeek-R1 Distill 7B tok/s | 89.79 | 264.42 | -66% |
| Gemma 3 12B tok/s | 52.07 | 150.58 | -65% |
| Gemma 4 12B tok/s | 52.67 | 148.17 | -64% |
| Mistral Small 24B tok/s | 30.94 | 105.62 | -71% |
| Phi-4 14B tok/s | 48.18 | 165.17 | -71% |
| Qwen3 14B tok/s | 48.12 | 151.16 | -68% |
| Qwen3 30B A3B tok/s | 140.12 | 283.78 | -51% |
| Qwen3 8B tok/s | 84.09 | 244.22 | -66% |
| TRELLIS Image-to-3D assets/hour | 293.6 | 864 | -66% |
| Codestral 22B tok/s | 32.09 | 110.05 | -71% |
| DeepSeek-R1 Distill 32B tok/s | 21.85 | 75.8 | -71% |
| Devstral Small 24B tok/s | 30.98 | 107.79 | -71% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 21.12 | 76.76 | -72% |
| Dolphin Mistral 24B Venice tok/s | 30.95 | 109.13 | -72% |
| Dolphin X1 8B tok/s | 86.61 | 266 | -67% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 86.54 | 266.19 | -67% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 30.9 | 107.83 | -71% |
| Gemma 3 27B tok/s | 24.8 | 84.77 | -71% |
| Qwen2.5-Coder 32B tok/s | 21.92 | 75.81 | -71% |
| Qwen3-Coder 30B A3B tok/s | 142.77 | 296.22 | -52% |
| QwQ 32B tok/s | 21.87 | 75.82 | -71% |
| StarCoder2 15B tok/s | 42.19 | 134.92 | -69% |
| Z-Image Turbo (1024px) images/min | 6.32 | 39.42 | -84% |
| BiRefNet images/min | 432.33 | 1310.97 | -67% |
| Depth Anything V2 Large images/min | 740.74 | 918.71 | -19% |
| Depth Anything V2 Small images/min | 941.18 | 1081.54 | -13% |
| SAM ViT-Base images/min | 398.72 | 1431.27 | -72% |
| SAM ViT-Huge images/min | 84.22 | 320.64 | -74% |
| Swin2SR 4x Upscaler images/min | 22.31 | 25.48 | -12% |
| Kokoro TTS 82M x realtime | 101.11 | 222.02 | -54% |
| MusicGen Small x realtime | 1.2 | 2.47 | -51% |
| Whisper large-v3 x realtime | 86.07 | 181.47 | -53% |
| Florence-2 Base images/min | 210.79 | 249.62 | -16% |
| Florence-2 Large images/min | 117.94 | 156.15 | -24% |
| Dolphin X1 Trinity Nano 6B tok/s | 171.23 | 247.94 | -31% |
| gpt-oss-20b tok/s | 144.61 | 346.8 | -58% |
| Olmo-3.1-32B-Think tok/s | 22.13 | 75.53 | -71% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 31 | 105.63 | -71% |
| DeepSeek-Coder-V2-Lite tok/s | 160.13 | 309.14 | -48% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 83.22 | 245.26 | -66% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 47 | 145.15 | -68% |
| gemma-2-2b-it-abliterated tok/s | 174.56 | 392.65 | -56% |
| gemma-2-2b tok/s | 175.03 | 392.26 | -55% |
| gemma-2-9b tok/s | 57.37 | 174.7 | -67% |
| gemma-3-1b tok/s | 284.78 | 504.17 | -44% |
| gemma-3-270m tok/s | 597.49 | 961.29 | -38% |
| GLM-4.7-Flash tok/s | 103.24 | 186.05 | -45% |
| SmolLM3-3B tok/s | 171.66 | 399.14 | -57% |
| KAT-Coder-V2.5-Dev tok/s | 113.6 | 231.83 | -51% |
| L3-8B-Stheno-v3.2 tok/s | 86.8 | 261.83 | -67% |
| LFM2.5-1.2B tok/s | 429.96 | 926.61 | -54% |
| Llama-2-7B tok/s | 96.69 | 284.62 | -66% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 167.63 | 424.25 | -60% |
| Meta-Llama-3.1-8B tok/s | 86.69 | 261.7 | -67% |
| Phi-4-mini tok/s | 144.27 | 381.52 | -62% |
| Mistral-7B-Instruct-v0.1 tok/s | 92.45 | 276.29 | -67% |
| Mistral-7B-Instruct-v0.2 tok/s | 92.74 | 276.73 | -66% |
| Mistral-7B-Instruct-v0.3 tok/s | 92.45 | 276.61 | -67% |
| Ornith-1.0-35B tok/s | 104.9 | 221.65 | -53% |
| Ornith-1.0-9B tok/s | 73.78 | 216.89 | -66% |
| Phi-3.5-mini tok/s | 145.97 | 345.07 | -58% |
| Qwen-AgentWorld-35B-A3B tok/s | 104.36 | 219.1 | -52% |
| Qwen3-0.6B tok/s | 446.74 | 710.97 | -37% |
| Qwen3-1.7B tok/s | 262.49 | 557.09 | -53% |
| Qwen3-14B tok/s | 47.8 | 152.2 | -69% |
| Qwen3-8B tok/s | 83.04 | 244.2 | -66% |
| Qwen2.5-0.5B tok/s | 508.78 | 890 | -43% |
| Qwen2.5-1.5B tok/s | 265.7 | 537.24 | -51% |
| Qwen2.5-14B tok/s | 47.02 | 144.7 | -68% |
| Qwen2.5-32B tok/s | 21.98 | 73.5 | -70% |
| Qwen2.5-3B tok/s | 168.23 | 395.52 | -57% |
| Qwen2.5-7B tok/s | 89.68 | 264.17 | -66% |
| Qwen2.5-Coder-1.5B tok/s | 264.79 | 538.44 | -51% |
| Qwen2.5-Coder-3B tok/s | 168 | 394 | -57% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 97.48 | 168.41 | -42% |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 83.26 | 245 | -66% |
| LFM2.5-8B-A1B tok/s | 268.04 | 581.98 | -54% |
| Mistral-Nemo-Instruct-2407 tok/s | 57 | 177.38 | -68% |
| Hermes-4-14B tok/s | 47.8 | 152 | -69% |
| phi-2 tok/s | 172.45 | 340.09 | -49% |
| Qwen3-30B-A3B tok/s | 140.14 | 285.44 | -51% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 46.91 | 145.01 | -68% |
| Qwen3-4B-Instruct-2507 tok/s | 129.9 | 311.47 | -58% |
| Qwen3-4B-Thinking-2507 tok/s | 129.99 | 310.84 | -58% |
| SmolLM2-135M tok/s | 680.37 | 898.53 | -24% |
| Cydonia-24B-v4.3 tok/s | 31.04 | 105.49 | -71% |
| CogVideoX-5B I2V clips/min | 0 | 0.879 | n/a |
| Qwen2.5 1.5B LoRA train tok/s | 4214.4 | 16034.9 | -74% |
| Qwen2.5 7B LoRA train tok/s | 1239 | 8514.3 | -85% |
| SmolLM2 1.7B LoRA train tok/s | 3668.3 | 18910.3 | -81% |
| TinyLlama 1.1B LoRA train tok/s | 5499.6 | 17522 | -69% |
| Stable Video Diffusion XT clips/min | 0 | 2.882 | n/a |
| Wan 2.2 TI2V-5B (image to video) clips/min | 0 | 4.126 | n/a |
| Qwen2.5 1.5B served serve tok/s | 2702.5 | 6794.8 | -60% |
| Qwen2.5 7B served serve tok/s | 858.7 | 3601.1 | -76% |
| SmolLM2 1.7B served serve tok/s | 2242.7 | 6509.9 | -66% |
| TinyLlama 1.1B served serve tok/s | 3558.2 | 8336.7 | -57% |
| FLUX.1 Schnell images/min | 0 | 58.36 | n/a |
| Qwen2.5 1.5B + 0.5B draft x vs solo | 0.919 | 0.725 | +27% |
| AI21-Jamba-Reasoning-3B tok/s | 166.29 | 363.09 | -54% |
| Codestral 22B (Q3_K_M) tok/s | 28.01 | 87.21 | -68% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 87.84 | 262.76 | -67% |
| DeepSeek-R1-Distill-Llama-70B tok/s | 0 | 41.17 | n/a |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 41.72 | 118.43 | -65% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 22.14 | 73.48 | -70% |
| dolphin-2.9-llama3-8b tok/s | 87.98 | 262.47 | -66% |
| Gemma 3 12B (Q3_K_M) tok/s | 46.29 | 125.92 | -63% |
| Hermes-3-Llama-3.2-3B tok/s | 169.23 | 423.72 | -60% |
| Hermes-4-70B tok/s | 0 | 41.2 | n/a |
| Qwen3-Coder-Next-abliterated tok/s | 0 | 190.32 | n/a |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| Llama-3.3-70B-Instruct-abliterated tok/s | 0 | 41.15 | n/a |
| Meta-Llama-3.1-70B tok/s | 0 | 41.19 | n/a |
| Mistral Small 24B (Q3_K_M) tok/s | 26.8 | 84 | -68% |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| NemoMix-Unleashed-12B tok/s | 57.52 | 177.28 | -68% |
| Nemotron-3-Nano-30B-A3B tok/s | 0 | 316.51 | n/a |
| Phi-4 14B (Q3_K_M) tok/s | 43.56 | 138.87 | -69% |
| Qwen3-4B-Instruct-2507 tok/s | 130.83 | 310.58 | -58% |
| Qwen3-Coder-Next tok/s | 0 | 188.21 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 188.14 | n/a |
| Qwen1.5-0.5B tok/s | 542.19 | 847.02 | -36% |
| Qwen2-1.5B tok/s | 265.45 | 536.52 | -51% |
| Uncensored tok/s | 47.45 | 144.92 | -67% |
| Qwen2.5-72B tok/s | 0 | 40.78 | n/a |
| Qwen2.5-Coder-0.5B tok/s | 513.36 | 892.81 | -43% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 18.78 | 58.05 | -68% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 91.41 | 264.24 | -65% |
| Qwen3-Coder-Next tok/s | 0 | 193.79 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 191.08 | n/a |
| Qwen3-Next-80B-A3B tok/s | 0 | 186.22 | n/a |
| Qwen3 30B A3B (Q3_K_M) tok/s | 120.38 | 242.6 | -50% |
Whole-job comparison
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A10G is faster on 0 of 8; NVIDIA H100 80GB HBM3 on 8.
| Workflow | NVIDIA A10G | NVIDIA H100 80GB HBM3 | Difference |
|---|---|---|---|
| 50-image depth pass | 7 s | 5 s | NVIDIA H100 80GB HBM3 1.20x faster |
| 30-minute podcast pass | 77 s | 34 s | NVIDIA H100 80GB HBM3 2.25x faster |
| 500-image masking run | 6 min | 1.6 min | NVIDIA H100 80GB HBM3 3.72x faster |
| 20-asset 3D game kit | 7.6 min | 2.2 min | NVIDIA H100 80GB HBM3 3.49x faster |
| 24-frame storyboard | 9 min | 1.8 min | NVIDIA H100 80GB HBM3 4.93x faster |
| 200-product catalogue cutout | 9.6 min | 8.1 min | NVIDIA H100 80GB HBM3 1.17x faster |
| Full codebase review | 21.3 min | 6.9 min | NVIDIA H100 80GB HBM3 3.08x faster |
| 10 short social clips | 50.7 min | 8 min | NVIDIA H100 80GB HBM3 6.35x faster |
Specifications compared
| NVIDIA A10G | NVIDIA H100 80GB HBM3 | |
|---|---|---|
| VRAM | 24GB | 80GB |
| Architecture | Ampere | Hopper |
| Memory bandwidth | 600 GB/s | 3350 GB/s |
| Boost clock | 1,710 MHz | 1,980 MHz |
| TDP | 150 W | 700 W |
| Launch MSRP | $2,800 | $30,000 |
| Release | 2021-11-01 | 2022-09-20 |
FAQ
- Which is better for ai & machine learning: NVIDIA A10G or NVIDIA H100 80GB HBM3?
- NVIDIA H100 80GB HBM3 performs better for ai & machine learning, winning 152 of 155 benchmarks in our suite with an average 179.4% advantage.
- What are the main hardware differences between NVIDIA A10G and NVIDIA H100 80GB HBM3?
- NVIDIA A10G has 24GB VRAM and a 150W TDP, while NVIDIA H100 80GB HBM3 has 80GB VRAM and a 700W TDP.
- Does VRAM matter more than speed between NVIDIA A10G and NVIDIA H100 80GB HBM3?
- For AI, yes, NVIDIA H100 80GB HBM3 fits 21 more of our 12 workloads than NVIDIA A10G. 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 A10G and NVIDIA H100 80GB HBM3?
- Wan 2.2 5B (720p): NVIDIA H100 80GB HBM3 leads by roughly 115% (0.18 vs 1.33 frames/s) in our testing.
NVIDIA A10G full review · NVIDIA H100 80GB HBM3 full review · All AI & Machine Learning rankings