NVIDIA A10G vs NVIDIA H200, AI & Machine Learning Comparison
NVIDIA H200 wins 147 of 149 benchmarks, averaging 188.6% 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 H200 leads by 683% (0.18 vs 1.41 frames/s); the closest fight is Swin2SR 4x Upscaler (16% apart); VRAM decides part of this one: NVIDIA H200 runs 150 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 H200 | Difference |
|---|---|---|---|
| Qwen3-4B tok/s | 129.87 | 318.84 | -59% |
| Llama-3.1-8B tok/s | 86.62 | 268.31 | -68% |
| Qwen2.5-Coder-14B tok/s | 47.07 | 148.43 | -68% |
| Qwen3-32B tok/s | 22.04 | 76.58 | -71% |
| Llama 3.3 70B tok/s | 0 | 42.66 | n/a |
| Stable Diffusion XL images/min | 6.38 | 37.16 | -83% |
| Z-Image Turbo images/min | 3.45 | 23.175 | -85% |
| FLUX.1 dev images/min | 0 | 9.514 | n/a |
| FLUX.1 Kontext dev images/min | 0 | 4.393 | n/a |
| Qwen-Image-Edit images/min | 0 | 3.76 | n/a |
| Wan 2.2 5B (720p) frames/s | 0.18 | 1.41 | -87% |
| Llama 3.2 1B tok/s | 400.6 | 875.53 | -54% |
| Qwen3 0.6B tok/s | 451.38 | 718.67 | -37% |
| Qwen3 1.7B tok/s | 267.08 | 563.52 | -53% |
| DeepSeek-R1 Distill 1.5B tok/s | 265.75 | 542.95 | -51% |
| Gemma 3 4B tok/s | 124.85 | 291.76 | -57% |
| Llama 3.2 3B tok/s | 170.47 | 424.61 | -60% |
| Mistral 7B v0.3 tok/s | 92.82 | 278.5 | -67% |
| Phi-4 Mini 3.8B tok/s | 144.42 | 392.38 | -63% |
| Qwen2.5-Coder 7B tok/s | 90.05 | 266.36 | -66% |
| SmolLM3 3B tok/s | 172.15 | 402.24 | -57% |
| DeepSeek-R1 Distill Llama 8B tok/s | 86.83 | 265.25 | -67% |
| DeepSeek-R1 Distill 14B tok/s | 46.96 | 146.82 | -68% |
| DeepSeek-R1 Distill 7B tok/s | 89.79 | 267.1 | -66% |
| Gemma 3 12B tok/s | 52.07 | 153.26 | -66% |
| Gemma 4 12B tok/s | 52.67 | 151.05 | -65% |
| Mistral Small 24B tok/s | 30.94 | 107.84 | -71% |
| Phi-4 14B tok/s | 48.18 | 168.2 | -71% |
| Qwen3 14B tok/s | 48.12 | 154.51 | -69% |
| Qwen3 30B A3B tok/s | 140.12 | 292.11 | -52% |
| Qwen3 8B tok/s | 84.09 | 247.98 | -66% |
| TRELLIS Image-to-3D assets/hour | 293.6 | 818.8 | -64% |
| Codestral 22B tok/s | 32.09 | 111.34 | -71% |
| DeepSeek-R1 Distill 32B tok/s | 21.85 | 75.8 | -71% |
| Devstral Small 24B tok/s | 30.98 | 109.05 | -72% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 21.12 | 76.7 | -72% |
| Dolphin Mistral 24B Venice tok/s | 30.95 | 109.14 | -72% |
| Dolphin X1 8B tok/s | 86.61 | 268.36 | -68% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 86.54 | 267.84 | -68% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 30.9 | 109.06 | -72% |
| Gemma 3 27B tok/s | 24.8 | 84.75 | -71% |
| Qwen2.5-Coder 32B tok/s | 21.92 | 75.77 | -71% |
| Qwen3-Coder 30B A3B tok/s | 142.77 | 297.9 | -52% |
| QwQ 32B tok/s | 21.87 | 75.77 | -71% |
| StarCoder2 15B tok/s | 42.19 | 136.29 | -69% |
| Z-Image Turbo (1024px) images/min | 6.32 | 40.74 | -84% |
| BiRefNet images/min | 432.33 | 1457.73 | -70% |
| Depth Anything V2 Large images/min | 740.74 | 1040.81 | -29% |
| Depth Anything V2 Small images/min | 941.18 | 1198.58 | -21% |
| SAM ViT-Base images/min | 398.72 | 1469.34 | -73% |
| SAM ViT-Huge images/min | 84.22 | 325.19 | -74% |
| Swin2SR 4x Upscaler images/min | 22.31 | 25.85 | -14% |
| Kokoro TTS 82M x realtime | 101.11 | 170.82 | -41% |
| MusicGen Small x realtime | 1.2 | 1.87 | -36% |
| Whisper large-v3 x realtime | 86.07 | 163.78 | -47% |
| Dolphin X1 Trinity Nano 6B tok/s | 171.23 | 259.81 | -34% |
| gpt-oss-20b tok/s | 144.61 | 354.23 | -59% |
| Olmo-3.1-32B-Think tok/s | 22.13 | 78.46 | -72% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 31 | 109.2 | -72% |
| DeepSeek-Coder-V2-Lite tok/s | 160.13 | 312.1 | -49% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 83.22 | 250.61 | -67% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 47 | 148.74 | -68% |
| gemma-2-2b-it-abliterated tok/s | 174.56 | 397.43 | -56% |
| gemma-2-2b tok/s | 175.03 | 398.48 | -56% |
| gemma-2-9b tok/s | 57.37 | 180.47 | -68% |
| gemma-3-1b tok/s | 284.78 | 513.58 | -45% |
| gemma-3-270m tok/s | 597.49 | 967.35 | -38% |
| GLM-4.7-Flash tok/s | 103.24 | 188.57 | -45% |
| SmolLM3-3B tok/s | 171.66 | 404.27 | -58% |
| KAT-Coder-V2.5-Dev tok/s | 113.6 | 236.31 | -52% |
| L3-8B-Stheno-v3.2 tok/s | 86.8 | 267.92 | -68% |
| LFM2.5-1.2B tok/s | 429.96 | 946.53 | -55% |
| Llama-2-7B tok/s | 96.69 | 290.81 | -67% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 167.63 | 429.78 | -61% |
| Meta-Llama-3.1-8B tok/s | 86.69 | 262.93 | -67% |
| Phi-4-mini tok/s | 144.27 | 395.3 | -64% |
| Mistral-7B-Instruct-v0.1 tok/s | 92.45 | 282.64 | -67% |
| Mistral-7B-Instruct-v0.2 tok/s | 92.74 | 282.32 | -67% |
| Mistral-7B-Instruct-v0.3 tok/s | 92.45 | 282.39 | -67% |
| Ornith-1.0-35B tok/s | 104.9 | 208.51 | -50% |
| Ornith-1.0-9B tok/s | 73.78 | 222.94 | -67% |
| Phi-3.5-mini tok/s | 145.97 | 348.14 | -58% |
| Qwen-AgentWorld-35B-A3B tok/s | 104.36 | 225.31 | -54% |
| Qwen3-0.6B tok/s | 446.74 | 724.67 | -38% |
| Qwen3-1.7B tok/s | 262.49 | 568.85 | -54% |
| Qwen3-14B tok/s | 47.8 | 156.3 | -69% |
| Qwen3-8B tok/s | 83.04 | 251.04 | -67% |
| Qwen2.5-0.5B tok/s | 508.78 | 914.58 | -44% |
| Qwen2.5-1.5B tok/s | 265.7 | 549.36 | -52% |
| Qwen2.5-14B tok/s | 47.02 | 148.76 | -68% |
| Qwen2.5-32B tok/s | 21.98 | 75.76 | -71% |
| Qwen2.5-3B tok/s | 168.23 | 400.48 | -58% |
| Qwen2.5-7B tok/s | 89.68 | 265.15 | -66% |
| Qwen2.5-Coder-1.5B tok/s | 264.79 | 545.54 | -51% |
| Qwen2.5-Coder-3B tok/s | 168 | 401.11 | -58% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 97.48 | 171.45 | -43% |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 83.26 | 250.88 | -67% |
| LFM2.5-8B-A1B tok/s | 268.04 | 583.96 | -54% |
| Mistral-Nemo-Instruct-2407 tok/s | 57 | 181.3 | -69% |
| Hermes-4-14B tok/s | 47.8 | 156.26 | -69% |
| phi-2 tok/s | 172.45 | 347.08 | -50% |
| Qwen3-30B-A3B tok/s | 140.14 | 293.47 | -52% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 46.91 | 148.48 | -68% |
| Qwen3-4B-Instruct-2507 tok/s | 129.9 | 318.27 | -59% |
| Qwen3-4B-Thinking-2507 tok/s | 129.99 | 318.94 | -59% |
| SmolLM2-135M tok/s | 680.37 | 905.38 | -25% |
| Cydonia-24B-v4.3 tok/s | 31.04 | 109.2 | -72% |
| Qwen2.5 1.5B LoRA train tok/s | 4214.4 | 14931.7 | -72% |
| Qwen2.5 7B LoRA train tok/s | 1239 | 8855.4 | -86% |
| SmolLM2 1.7B LoRA train tok/s | 3668.3 | 17650.4 | -79% |
| TinyLlama 1.1B LoRA train tok/s | 5499.6 | 16222.4 | -66% |
| Qwen2.5 1.5B served serve tok/s | 2702.5 | 7541.6 | -64% |
| Qwen2.5 7B served serve tok/s | 858.7 | 4394.8 | -80% |
| SmolLM2 1.7B served serve tok/s | 2242.7 | 7107.8 | -68% |
| TinyLlama 1.1B served serve tok/s | 3558.2 | 9137.7 | -61% |
| FLUX.1 Schnell images/min | 0 | 60.51 | n/a |
| AI21-Jamba-Reasoning-3B tok/s | 166.29 | 370.66 | -55% |
| Codestral 22B (Q3_K_M) tok/s | 28.01 | 88.03 | -68% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 87.84 | 267.82 | -67% |
| DeepSeek-R1-Distill-Llama-70B tok/s | 0 | 42.76 | n/a |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 41.72 | 119.66 | -65% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 22.14 | 75.71 | -71% |
| dolphin-2.9-llama3-8b tok/s | 87.98 | 267.96 | -67% |
| Gemma 3 12B (Q3_K_M) tok/s | 46.29 | 127.04 | -64% |
| Hermes-3-Llama-3.2-3B tok/s | 169.23 | 430.79 | -61% |
| Hermes-4-70B tok/s | 0 | 42.77 | n/a |
| Qwen3-Coder-Next-abliterated tok/s | 0 | 197.14 | n/a |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| Llama-3.3-70B-Instruct-abliterated tok/s | 0 | 42.75 | n/a |
| Meta-Llama-3.1-70B tok/s | 0 | 42.72 | n/a |
| Mistral Small 24B (Q3_K_M) tok/s | 26.8 | 85.1 | -69% |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| NemoMix-Unleashed-12B tok/s | 57.52 | 181.53 | -68% |
| Nemotron-3-Nano-30B-A3B tok/s | 0 | 327.48 | n/a |
| Phi-4 14B (Q3_K_M) tok/s | 43.56 | 139.97 | -69% |
| Qwen3-4B-Instruct-2507 tok/s | 130.83 | 318.67 | -59% |
| Qwen3-Coder-Next tok/s | 0 | 195.21 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 194.82 | n/a |
| Qwen1.5-0.5B tok/s | 542.19 | 855.22 | -37% |
| Qwen2-1.5B tok/s | 265.45 | 546.47 | -51% |
| Uncensored tok/s | 47.45 | 148.7 | -68% |
| Qwen2.5-72B tok/s | 0 | 43.4 | n/a |
| Qwen2.5-Coder-0.5B tok/s | 513.36 | 920.36 | -44% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 18.78 | 58.87 | -68% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 91.41 | 270.92 | -66% |
| Qwen3-Coder-Next tok/s | 0 | 196.65 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 201.12 | n/a |
| Qwen3-Next-80B-A3B tok/s | 0 | 192.52 | n/a |
| Qwen3 30B A3B (Q3_K_M) tok/s | 120.38 | 246.35 | -51% |
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 H200 on 8.
| Workflow | NVIDIA A10G | NVIDIA H200 | Difference |
|---|---|---|---|
| 50-image depth pass | 7 s | 6 s | NVIDIA H200 1.07x faster |
| 30-minute podcast pass | 77 s | 46 s | NVIDIA H200 1.68x faster |
| 500-image masking run | 6 min | 1.7 min | NVIDIA H200 3.60x faster |
| 20-asset 3D game kit | 7.6 min | 2.4 min | NVIDIA H200 3.15x faster |
| 24-frame storyboard | 9 min | 3.9 min | NVIDIA H200 2.29x faster |
| 200-product catalogue cutout | 9.6 min | 8 min | NVIDIA H200 1.20x faster |
| Full codebase review | 21.3 min | 6.7 min | NVIDIA H200 3.16x faster |
| 10 short social clips | 50.7 min | 10 min | NVIDIA H200 5.07x faster |
Specifications compared
| NVIDIA A10G | NVIDIA H200 | |
|---|---|---|
| VRAM | 24GB | 141GB |
| Architecture | Ampere | Hopper |
| Memory bandwidth | 600 GB/s | 4800 GB/s |
| Boost clock | 1,710 MHz | 1,980 MHz |
| TDP | 150 W | 700 W |
| Launch MSRP | $2,800 | $31,000 |
| Release | 2021-11-01 | 2024-03-18 |
FAQ
- Which is better for ai & machine learning: NVIDIA A10G or NVIDIA H200?
- NVIDIA H200 performs better for ai & machine learning, winning 147 of 149 benchmarks in our suite with an average 188.6% advantage.
- What are the main hardware differences between NVIDIA A10G and NVIDIA H200?
- NVIDIA A10G has 24GB VRAM and a 150W TDP, while NVIDIA H200 has 141GB VRAM and a 700W TDP.
- Does VRAM matter more than speed between NVIDIA A10G and NVIDIA H200?
- For AI, yes, NVIDIA H200 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 H200?
- Wan 2.2 5B (720p): NVIDIA H200 leads by roughly 123% (0.18 vs 1.41 frames/s) in our testing.
NVIDIA A10G full review · NVIDIA H200 full review · All AI & Machine Learning rankings