NVIDIA A100 40GB SXM4 vs NVIDIA A10G, AI & Machine Learning Comparison
NVIDIA A100 40GB SXM4 wins 122 of 149 benchmarks, averaging 61.1% faster.
NVIDIA A100 40GB SXM4's numbers are anchored estimates calibrated against our measured cards, pending first-party measurement. Treat small gaps as ties.
What the numbers say
The gap is widest in Wan 2.2 5B (720p), where NVIDIA A100 40GB SXM4 leads by 267% (0.66 vs 0.18 frames/s); the closest fight is Qwen2.5-0.5B (2% apart); VRAM decides part of this one: NVIDIA A100 40GB SXM4 runs 137 of our 12 AI workloads while the other card runs 135, models that don't fit score zero.
Benchmark results head-to-head
| Benchmark | NVIDIA A100 40GB SXM4 | NVIDIA A10G | Difference |
|---|---|---|---|
| Qwen3-4B tok/s | 193.14 | 129.87 | +49% |
| Llama-3.1-8B tok/s | 157.26 | 86.62 | +82% |
| Qwen2.5-Coder-14B tok/s | 86.2 | 47.07 | +83% |
| Qwen3-32B tok/s | 43.8 | 22.04 | +99% |
| Llama-3.3-70B tok/s | 0 | 0 | n/a |
| Stable Diffusion XL images/min | 18.576 | 6.38 | +191% |
| Z-Image Turbo images/min | 9.975 | 3.45 | +189% |
| FLUX.1 dev images/min | 4.264 | 0 | n/a |
| FLUX.1 Kontext dev images/min | 1.993 | 0 | n/a |
| Qwen-Image-Edit images/min | 0 | 0 | n/a |
| Wan 2.2 5B (720p) frames/s | 0.66 | 0.18 | +267% |
| DeepSeek-R1 Distill Llama 8B tok/s | 157.16 | 86.83 | +81% |
| DeepSeek-R1 Distill 1.5B tok/s | 320.78 | 265.75 | +21% |
| DeepSeek-R1 Distill 14B tok/s | 86.39 | 46.96 | +84% |
| DeepSeek-R1 Distill 7B tok/s | 159.09 | 89.79 | +77% |
| Gemma 3 12B tok/s | 88.97 | 52.07 | +71% |
| Gemma 3 4B tok/s | 174.14 | 124.85 | +39% |
| Gemma 4 12B tok/s | 87.14 | 52.67 | +65% |
| Llama 3.2 1B tok/s | 532.08 | 400.6 | +33% |
| Llama 3.2 3B tok/s | 261.24 | 170.47 | +53% |
| Mistral 7B v0.3 tok/s | 165.9 | 92.82 | +79% |
| Mistral Small 24B tok/s | 62.42 | 30.94 | +102% |
| Phi-4 14B tok/s | 98.65 | 48.18 | +105% |
| Phi-4 Mini 3.8B tok/s | 236.71 | 144.42 | +64% |
| Qwen2.5-Coder 7B tok/s | 159.9 | 90.05 | +78% |
| Qwen3 0.6B tok/s | 417.88 | 451.38 | -7% |
| Qwen3 1.7B tok/s | 340.76 | 267.08 | +28% |
| Qwen3 14B tok/s | 90.01 | 48.12 | +87% |
| Qwen3 30B A3B tok/s | 169.1 | 140.12 | +21% |
| Qwen3 8B tok/s | 146.21 | 84.09 | +74% |
| SmolLM3 3B tok/s | 246.25 | 172.15 | +43% |
| Codestral 22B tok/s | 63.74 | 32.09 | +99% |
| DeepSeek-R1 Distill 32B tok/s | 43.13 | 21.85 | +97% |
| Devstral Small 24B tok/s | 62.4 | 30.98 | +101% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 42.97 | 21.12 | +103% |
| Dolphin Mistral 24B Venice tok/s | 62.46 | 30.95 | +102% |
| Dolphin X1 8B tok/s | 156.3 | 86.61 | +80% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 156.55 | 86.54 | +81% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 62.4 | 30.9 | +102% |
| Gemma 3 27B tok/s | 47.47 | 24.8 | +91% |
| Qwen2.5-Coder 32B tok/s | 43.14 | 21.92 | +97% |
| Qwen3-Coder 30B A3B tok/s | 173.47 | 142.77 | +22% |
| QwQ 32B tok/s | 43.02 | 21.87 | +97% |
| StarCoder2 15B tok/s | 80.25 | 42.19 | +90% |
| FLUX.1 Schnell images/min | 28.44 | 0 | n/a |
| Z-Image Turbo (1024px) images/min | 18.7 | 6.32 | +196% |
| BiRefNet images/min | 611.56 | 432.33 | +41% |
| Depth Anything V2 Large images/min | 536.74 | 740.74 | -28% |
| Depth Anything V2 Small images/min | 613.42 | 941.18 | -35% |
| SAM ViT-Base images/min | 696.77 | 398.72 | +75% |
| SAM ViT-Huge images/min | 146.81 | 84.22 | +74% |
| Swin2SR 4x Upscaler images/min | 17.25 | 22.31 | -23% |
| Kokoro TTS 82M x realtime | 142.03 | 101.11 | +40% |
| MusicGen Small x realtime | 1.08 | 1.2 | -10% |
| Whisper large-v3 x realtime | 102.47 | 86.07 | +19% |
| Dolphin X1 Trinity Nano 6B tok/s | 153.27 | 171.23 | -10% |
| gpt-oss-20b tok/s | 205.1 | 144.61 | +42% |
| Olmo-3.1-32B-Think tok/s | 43.67 | 22.13 | +97% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 62.56 | 31 | +102% |
| DeepSeek-Coder-V2-Lite tok/s | 205.42 | 160.13 | +28% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 145.84 | 83.22 | +75% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 86.48 | 47 | +84% |
| Qwen2.5 1.5B LoRA train tok/s | 6349.7 | 4214.4 | +51% |
| Qwen2.5 7B LoRA train tok/s | 3670.8 | 1239 | +196% |
| SmolLM2 1.7B LoRA train tok/s | 6938.8 | 3668.3 | +89% |
| TinyLlama 1.1B LoRA train tok/s | 7471 | 5499.6 | +36% |
| gemma-2-2b-it-abliterated tok/s | 232.27 | 174.56 | +33% |
| gemma-2-2b tok/s | 234.68 | 175.03 | +34% |
| gemma-2-9b tok/s | 101.97 | 57.37 | +78% |
| gemma-3-1b tok/s | 310.69 | 284.78 | +9% |
| gemma-3-270m tok/s | 574.24 | 597.49 | -4% |
| GLM-4.7-Flash tok/s | 119.1 | 103.24 | +15% |
| SmolLM3-3B tok/s | 247.66 | 171.66 | +44% |
| KAT-Coder-V2.5-Dev tok/s | 137.31 | 113.6 | +21% |
| LFM2.5-1.2B tok/s | 572.69 | 429.96 | +33% |
| Llama-2-7B tok/s | 172.53 | 96.69 | +78% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 260.83 | 167.63 | +56% |
| Meta-Llama-3.1-8B tok/s | 156.68 | 86.69 | +81% |
| Phi-4-mini tok/s | 237.38 | 144.27 | +65% |
| Mistral-7B-Instruct-v0.1 tok/s | 166.39 | 92.45 | +80% |
| Mistral-7B-Instruct-v0.2 tok/s | 166.8 | 92.74 | +80% |
| Mistral-7B-Instruct-v0.3 tok/s | 167.06 | 92.45 | +81% |
| Ornith-1.0-35B tok/s | 131.77 | 104.9 | +26% |
| Ornith-1.0-9B tok/s | 126.79 | 73.78 | +72% |
| Phi-3.5-mini tok/s | 222.53 | 145.97 | +52% |
| Qwen-AgentWorld-35B-A3B tok/s | 131.16 | 104.36 | +26% |
| Qwen3-0.6B tok/s | 408.39 | 446.74 | -9% |
| Qwen3-1.7B tok/s | 341.85 | 262.49 | +30% |
| Qwen3-14B tok/s | 90.07 | 47.8 | +88% |
| Qwen3-8B tok/s | 145.96 | 83.04 | +76% |
| Qwen2.5-0.5B tok/s | 517.09 | 508.78 | +2% |
| Qwen2.5-1.5B tok/s | 315.59 | 265.7 | +19% |
| Qwen2.5-14B tok/s | 86.07 | 47.02 | +83% |
| Qwen2.5-32B tok/s | 43.23 | 21.98 | +97% |
| Qwen2.5-3B tok/s | 240.48 | 168.23 | +43% |
| Qwen2.5-7B tok/s | 159.81 | 89.68 | +78% |
| Qwen2.5-Coder-1.5B tok/s | 322.96 | 264.79 | +22% |
| Qwen2.5-Coder-3B tok/s | 241.51 | 168 | +44% |
| AI21-Jamba-Reasoning-3B tok/s | 231.15 | 166.29 | +39% |
| Codestral 22B (Q3_K_M) tok/s | 48.37 | 28.01 | +73% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 156.62 | 87.84 | +78% |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 65.33 | 41.72 | +57% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 43.07 | 22.14 | +95% |
| dolphin-2.9-llama3-8b tok/s | 153.34 | 87.98 | +74% |
| Gemma 3 12B (Q3_K_M) tok/s | 69.7 | 46.29 | +51% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 109.1 | 97.48 | +12% |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 146.8 | 83.26 | +76% |
| Hermes-3-Llama-3.2-3B tok/s | 261 | 169.23 | +54% |
| L3-8B-Stheno-v3.2 tok/s | 156.55 | 86.8 | +80% |
| LFM2.5-8B-A1B tok/s | 351.47 | 268.04 | +31% |
| Mistral-Nemo-Instruct-2407 tok/s | 106.33 | 57 | +87% |
| Mistral Small 24B (Q3_K_M) tok/s | 45.34 | 26.8 | +69% |
| NemoMix-Unleashed-12B tok/s | 105.05 | 57.52 | +83% |
| Nemotron-3-Nano-30B-A3B tok/s | 192.44 | 0 | n/a |
| Hermes-4-14B tok/s | 89.97 | 47.8 | +88% |
| phi-2 tok/s | 228.1 | 172.45 | +32% |
| Phi-4 14B (Q3_K_M) tok/s | 80.71 | 43.56 | +85% |
| Qwen3-30B-A3B tok/s | 167.1 | 140.14 | +19% |
| Qwen3-4B-Instruct-2507 tok/s | 192.89 | 130.83 | +47% |
| Qwen1.5-0.5B tok/s | 497.73 | 542.19 | -8% |
| Qwen2-1.5B tok/s | 322.09 | 265.45 | +21% |
| Uncensored tok/s | 85.82 | 47.45 | +81% |
| Qwen2.5-Coder-0.5B tok/s | 542.36 | 513.36 | +6% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 86.34 | 46.91 | +84% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 31.04 | 18.78 | +65% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 159.76 | 91.41 | +75% |
| Qwen3 30B A3B (Q3_K_M) tok/s | 142.78 | 120.38 | +19% |
| Qwen3-4B-Instruct-2507 tok/s | 193.39 | 129.9 | +49% |
| Qwen3-4B-Thinking-2507 tok/s | 189.89 | 129.99 | +46% |
| SmolLM2-135M tok/s | 544.06 | 680.37 | -20% |
| Cydonia-24B-v4.3 tok/s | 62.41 | 31.04 | +101% |
| Qwen2.5 1.5B served serve tok/s | 4111.1 | 2702.5 | +52% |
| Qwen2.5 7B served serve tok/s | 1769.6 | 858.7 | +106% |
| SmolLM2 1.7B served serve tok/s | 3865.2 | 2242.7 | +72% |
| TinyLlama 1.1B served serve tok/s | 3221.3 | 3558.2 | -9% |
| Qwen2.5 1.5B + 0.5B draft x vs solo | 0.694 | 0.919 | -24% |
| Qwen3-Coder-Next-abliterated tok/s | 0 | 0 | n/a |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| Qwen3-Coder-Next tok/s | 0 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 0 | n/a |
| Qwen2.5-72B tok/s | 0 | 0 | n/a |
| Qwen3-Coder-Next tok/s | 0 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 0 | n/a |
| Qwen3-Next-80B-A3B tok/s | 0 | 0 | n/a |
| DeepSeek-R1-Distill-Llama-70B tok/s | 0 | 0 | n/a |
| Hermes-4-70B tok/s | 0 | 0 | n/a |
| Llama-3.3-70B-Instruct-abliterated tok/s | 0 | 0 | n/a |
| Meta-Llama-3.1-70B tok/s | 0 | 0 | n/a |
Whole-job comparison
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A100 40GB SXM4 is faster on 5 of 7; NVIDIA A10G on 2.
| Workflow | NVIDIA A100 40GB SXM4 | NVIDIA A10G | Difference |
|---|---|---|---|
| 50-image depth pass | 10 s | 7 s | NVIDIA A10G 1.46x faster |
| 30-minute podcast pass | 59 s | 77 s | NVIDIA A100 40GB SXM4 1.30x faster |
| 24-frame storyboard | 3 min | 9 min | NVIDIA A100 40GB SXM4 2.94x faster |
| 500-image masking run | 3.5 min | 6 min | NVIDIA A100 40GB SXM4 1.72x faster |
| Full codebase review | 11.7 min | 21.3 min | NVIDIA A100 40GB SXM4 1.83x faster |
| 200-product catalogue cutout | 12.2 min | 9.6 min | NVIDIA A10G 1.27x faster |
| 10 short social clips | 13.7 min | 50.7 min | NVIDIA A100 40GB SXM4 3.71x faster |
Specifications compared
| NVIDIA A100 40GB SXM4 | NVIDIA A10G | |
|---|---|---|
| VRAM | 40GB | 24GB |
| Architecture | Ampere | Ampere |
| Memory bandwidth | 1555 GB/s | 600 GB/s |
| Boost clock | 1,410 MHz | 1,710 MHz |
| TDP | 400 W | 150 W |
| Launch MSRP | $12,000 | $2,800 |
| Release | 2020-05-14 | 2021-11-01 |
FAQ
- Which is better for ai & machine learning: NVIDIA A100 40GB SXM4 or NVIDIA A10G?
- NVIDIA A100 40GB SXM4 performs better for ai & machine learning, winning 122 of 149 benchmarks in our suite with an average 61.1% advantage.
- What are the main hardware differences between NVIDIA A100 40GB SXM4 and NVIDIA A10G?
- NVIDIA A100 40GB SXM4 has 40GB VRAM and a 400W TDP, while NVIDIA A10G has 24GB VRAM and a 150W TDP.
- Does VRAM matter more than speed between NVIDIA A100 40GB SXM4 and NVIDIA A10G?
- For AI, yes, NVIDIA A100 40GB SXM4 fits 8 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 A100 40GB SXM4 and NVIDIA A10G?
- Wan 2.2 5B (720p): NVIDIA A100 40GB SXM4 leads by roughly 48% (0.66 vs 0.18 frames/s) in our testing.
NVIDIA A100 40GB SXM4 full review · NVIDIA A10G full review · All AI & Machine Learning rankings