NVIDIA A10G vs NVIDIA L4, AI & Machine Learning Comparison
NVIDIA A10G wins 159 of 186 benchmarks, averaging 53.3% faster.
Both cards were measured first-party on our bench, same suite, same test rig.
What the numbers say
The gap is widest in EzAudio XL, where NVIDIA A10G leads by 87% (0.71 vs 0.38 x realtime); the closest fight is Sana 1.6B (2% apart); VRAM decides part of this one: NVIDIA A10G runs 172 of our 12 AI workloads while the other card runs 161, models that don't fit score zero.
Benchmark results head-to-head
| Benchmark | NVIDIA A10G | NVIDIA L4 | Difference |
|---|---|---|---|
| SmolLM2-135M tok/s | 680.37 | 652.15 | +4% |
| gemma-3-270m tok/s | 597.49 | 493.31 | +21% |
| SmolLM2-360M tok/s | 491.77 | 440.21 | +12% |
| Qwen1.5-0.5B tok/s | 542.19 | 431.41 | +26% |
| Qwen2 0.5B tok/s | 496.98 | 396.94 | +25% |
| Qwen2.5-0.5B tok/s | 508.78 | 399.4 | +27% |
| Qwen2.5-Coder-0.5B tok/s | 513.36 | 397.1 | +29% |
| Qwen3 0.6B tok/s | 451.38 | 357.79 | +26% |
| Qwen3-0.6B tok/s | 446.74 | 356.93 | +25% |
| Llama 3.2 1B tok/s | 400.6 | 259.48 | +54% |
| gemma-3-1b tok/s | 284.78 | 216.27 | +32% |
| LFM2.5-1.2B tok/s | 429.96 | 278.53 | +54% |
| DeepSeek-R1 Distill 1.5B tok/s | 265.75 | 191.38 | +39% |
| Qwen2-1.5B tok/s | 265.45 | 189.18 | +40% |
| Qwen2.5-1.5B tok/s | 265.7 | 191.39 | +39% |
| Qwen2.5-Coder-1.5B tok/s | 264.79 | 184.6 | +43% |
| Qwen3 1.7B tok/s | 267.08 | 177.5 | +50% |
| Qwen3-1.7B tok/s | 262.49 | 176.08 | +49% |
| MiniCPM5 2B tok/s | 201.2 | 138.25 | +46% |
| gemma-2-2b tok/s | 175.03 | 114.17 | +53% |
| gemma-2-2b-it-abliterated tok/s | 174.56 | 113 | +54% |
| LFM2.5 2.6B tok/s | 201.87 | 128.88 | +57% |
| AI21-Jamba-Reasoning-3B tok/s | 166.29 | 107.59 | +55% |
| Granite 4.1 3B tok/s | 147.45 | 96.18 | +53% |
| Hermes-3-Llama-3.2-3B tok/s | 169.23 | 106.14 | +59% |
| Llama 3.2 3B tok/s | 170.47 | 107.1 | +59% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 167.63 | 101.16 | +66% |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| Qwen2.5-3B tok/s | 168.23 | 110.84 | +52% |
| Qwen2.5-Coder-3B tok/s | 168 | 109.42 | +54% |
| SmolLM3 3B tok/s | 172.15 | 112.07 | +54% |
| SmolLM3-3B tok/s | 171.66 | 110.77 | +55% |
| Phi-4 Mini 3.8B tok/s | 144.42 | 88.48 | +63% |
| Gemma 3 4B tok/s | 124.85 | 82.03 | +52% |
| Nemotron 3 Nano 4B tok/s | 137.33 | 79.8 | +72% |
| Qwen3-4B tok/s | 129.87 | 84 | +55% |
| Qwen3-4B-Instruct-2507 tok/s | 129.9 | 83.13 | +56% |
| Qwen3-4B-Instruct-2507 tok/s | 130.83 | 83.03 | +58% |
| Qwen3-4B-Thinking-2507 tok/s | 129.99 | 83.12 | +56% |
| phi-2 tok/s | 172.45 | 114.99 | +50% |
| Dolphin X1 Trinity Nano 6B tok/s | 171.23 | 166.6 | +3% |
| Phi-3.5-mini tok/s | 145.97 | 90.56 | +61% |
| Phi-4-mini tok/s | 144.27 | 87.21 | +65% |
| DeepSeek Coder 7B Instruct v1.5 tok/s | 99.14 | 57.06 | +74% |
| DeepSeek-R1 Distill 7B tok/s | 89.79 | 53.27 | +69% |
| Llama-2-7B tok/s | 96.69 | 56.06 | +72% |
| Mistral 7B v0.3 tok/s | 92.82 | 54.02 | +72% |
| Mistral-7B-Instruct-v0.1 tok/s | 92.45 | 53.6 | +72% |
| Mistral-7B-Instruct-v0.2 tok/s | 92.74 | 53.31 | +74% |
| Mistral-7B-Instruct-v0.3 tok/s | 92.45 | 53.66 | +72% |
| Qwen2.5-7B tok/s | 89.68 | 53.32 | +68% |
| Qwen2.5-Coder 7B tok/s | 90.05 | 53.26 | +69% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 91.41 | 53.02 | +72% |
| Qwen2.5-VL 7B Instruct tok/s | 97.64 | 52.57 | +86% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 87.84 | 48.82 | +80% |
| DeepSeek-R1 Distill Llama 8B tok/s | 86.83 | 50.39 | +72% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 83.22 | 49.07 | +70% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 86.54 | 50.24 | +72% |
| Dolphin X1 8B tok/s | 86.61 | 50.12 | +73% |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 83.26 | 48.72 | +71% |
| L3-8B-Stheno-v3.2 tok/s | 86.8 | 50.07 | +73% |
| LFM2.5-8B-A1B tok/s | 268.04 | 170.2 | +57% |
| Llama 3 8B tok/s | 87.2 | 50.35 | +73% |
| Llama-3.1-8B tok/s | 86.62 | 50.45 | +72% |
| Meta-Llama-3.1-8B tok/s | 86.69 | 50.36 | +72% |
| Qwen3 8B tok/s | 84.09 | 48.95 | +72% |
| Qwen3-8B tok/s | 83.04 | 48.34 | +72% |
| dolphin-2.9-llama3-8b tok/s | 87.98 | 50.04 | +76% |
| Nemotron Nano 9B v2 tok/s | 64.8 | 35.8 | +81% |
| Ornith 1.5 9B tok/s | 75.38 | 43.76 | +72% |
| Ornith-1.0-9B tok/s | 73.78 | 43.35 | +70% |
| gemma-2-9b tok/s | 57.37 | 33.8 | +70% |
| Gemma 3 12B tok/s | 52.07 | 31.01 | +68% |
| Gemma 3 12B (Q3_K_M) tok/s | 46.29 | 32.29 | +43% |
| Gemma 4 12B tok/s | 52.67 | 31.49 | +67% |
| NemoMix-Unleashed-12B tok/s | 57.52 | 33.02 | +74% |
| DeepSeek-R1 Distill 14B tok/s | 46.96 | 27.39 | +71% |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 41.72 | 27.87 | +50% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 47 | 27.3 | +72% |
| Hermes-4-14B tok/s | 47.8 | 27.48 | +74% |
| Phi-4 14B tok/s | 48.18 | 27.24 | +77% |
| Phi-4 14B (Q3_K_M) tok/s | 43.56 | 27.86 | +56% |
| Qwen2.5-14B tok/s | 47.02 | 27.37 | +72% |
| Qwen2.5-Coder-14B tok/s | 47.07 | 27.46 | +71% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 46.91 | 27.29 | +72% |
| Qwen3 14B tok/s | 48.12 | 27.61 | +74% |
| Qwen3-14B tok/s | 47.8 | 27.48 | +74% |
| Uncensored tok/s | 47.45 | 27.29 | +74% |
| StarCoder2 15B tok/s | 42.19 | 24.4 | +73% |
| Mistral-Nemo-Instruct-2407 tok/s | 57 | 33.01 | +73% |
| gpt-oss-20b tok/s | 144.61 | 91.78 | +58% |
| Codestral 22B tok/s | 32.09 | 18.14 | +77% |
| Codestral 22B (Q3_K_M) tok/s | 28.01 | 17.96 | +56% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 97.48 | 70.87 | +38% |
| DeepSeek-Coder-V2-Lite tok/s | 160.13 | 110.11 | +45% |
| Cydonia-24B-v4.3 tok/s | 31.04 | 17.3 | +79% |
| Devstral Small 24B tok/s | 30.98 | 17.34 | +79% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 30.9 | 17.32 | +78% |
| Dolphin Mistral 24B Venice tok/s | 30.95 | 17.34 | +78% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 31 | 17.2 | +80% |
| Mistral Small 24B tok/s | 30.94 | 17.33 | +79% |
| Mistral Small 24B (Q3_K_M) tok/s | 26.8 | 16.62 | +61% |
| Gemma 4 26B A4B tok/s | 101.91 | 68.51 | +49% |
| Gemma 3 27B tok/s | 24.8 | 14.18 | +75% |
| Qwen3.6 27B tok/s | 24.86 | 14.2 | +75% |
| Qwen3.8 27B tok/s | 24.51 | 13.91 | +76% |
| Nemotron 3.5 Lightning 30B A3B tok/s | 0 | 0 | n/a |
| Nemotron-3-Nano-30B-A3B tok/s | 0 | 0 | n/a |
| Qwen3 30B A3B tok/s | 140.12 | 96.15 | +46% |
| Qwen3 30B A3B (Q3_K_M) tok/s | 120.38 | 94.03 | +28% |
| Qwen3 30B A3B Instruct 2507 tok/s | 147.33 | 99.74 | +48% |
| Qwen3-30B-A3B tok/s | 140.14 | 95.02 | +47% |
| Qwen3-Coder 30B A3B tok/s | 142.77 | 99.14 | +44% |
| Gemma 4 31B tok/s | 22.67 | 12.84 | +77% |
| DeepSeek-R1 Distill 32B tok/s | 21.85 | 12.28 | +78% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 22.14 | 12.26 | +81% |
| Olmo-3.1-32B-Think tok/s | 22.13 | 12.27 | +80% |
| QwQ 32B tok/s | 21.87 | 12.28 | +78% |
| Qwen2.5-32B tok/s | 21.98 | 12.24 | +80% |
| Qwen2.5-Coder 32B tok/s | 21.92 | 12.29 | +78% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 18.78 | 12.05 | +56% |
| Qwen3-32B tok/s | 22.04 | 12.42 | +77% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 21.12 | 11.92 | +77% |
| Ornith 1.5 35B A3B tok/s | 123.77 | 84.69 | +46% |
| Ornith-1.0-35B tok/s | 104.9 | 70.31 | +49% |
| Qwen-AgentWorld-35B-A3B tok/s | 104.36 | 70.23 | +49% |
| Qwen3.6 35B A3B tok/s | 117.92 | 78.98 | +49% |
| GLM-4.7-Flash tok/s | 103.24 | 75.03 | +38% |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| KAT-Coder-V2.5-Dev tok/s | 113.6 | 79.01 | +44% |
| DeepSeek-R1-Distill-Llama-70B tok/s | 0 | 0 | n/a |
| Hermes-4-70B tok/s | 0 | 0 | n/a |
| Llama 3.3 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 |
| Qwen2.5-72B tok/s | 0 | 0 | n/a |
| Qwen3-Next-80B-A3B tok/s | 0 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 0 | 0 | n/a |
| Qwen3-Coder-Next tok/s | 0 | 0 | n/a |
| Qwen3-Coder-Next tok/s | 0 | 0 | n/a |
| Qwen3-Coder-Next-abliterated tok/s | 0 | 0 | n/a |
| gpt-oss-120b tok/s | 0 | 0 | n/a |
| BiRefNet images/min | 432.33 | 305.81 | +41% |
| Swin2SR 4x Upscaler images/min | 22.31 | 18.23 | +22% |
| Sana 1.6B images/min | 14.13 | 13.84 | +2% |
| Stable Diffusion XL images/min | 6.38 | 5.18 | +23% |
| PixArt-Sigma XL images/min | 9.93 | 8.08 | +23% |
| Stable Diffusion 3.5 Medium images/min | 4.11 | 2.96 | +39% |
| Z-Image Turbo images/min | 3.45 | 2.475 | +39% |
| Z-Image Turbo (1024px) images/min | 6.32 | 4.6 | +37% |
| AuraFlow v0.3 images/min | 1.39 | 1.02 | +36% |
| FLUX.1 Kontext dev images/min | 0 | 0 | n/a |
| FLUX.1 dev images/min | 0 | 0 | n/a |
| Stable Diffusion 3.5 Large images/min | 0 | 0 | n/a |
| FLUX.1 Schnell images/min | 0 | 0 | n/a |
| Qwen-Image-Edit images/min | 0 | 0 | n/a |
| Wan 2.2 5B (720p) frames/s | 0.18 | 0.15 | +20% |
| Wan 2.2 TI2V-5B (image to video) clips/min | 0 | 0 | n/a |
| Stable Video Diffusion XT clips/min | 0 | 0 | n/a |
| CogVideoX-5B I2V clips/min | 0 | 0 | n/a |
| TripoSR Image-to-3D assets/hour | 1250.7 | 967.5 | +29% |
| TripoSG Image-to-3D assets/hour | 127.3 | 83.2 | +53% |
| TRELLIS Image-to-3D assets/hour | 94 | 68.8 | +37% |
| TRELLIS.2 Image-to-3D assets/hour | 30.8 | 23.6 | +31% |
| Kokoro TTS 82M x realtime | 101.11 | 97.35 | +4% |
| MusicGen Small x realtime | 1.2 | 1.01 | +19% |
| EzAudio XL x realtime | 0.71 | 0.38 | +87% |
| Whisper large-v3 x realtime | 86.07 | 70.11 | +23% |
| ACE-Step 1.5 x realtime | 18.65 | 12.84 | +45% |
| MOSS-SoundEffect v2.0 x realtime | 0.53 | 0.37 | +43% |
| Depth Anything V2 Small images/min | 941.18 | 916.66 | +3% |
| Florence-2 Base images/min | 210.79 | 174.65 | +21% |
| SAM ViT-Base images/min | 398.72 | 315.89 | +26% |
| Depth Anything V2 Large images/min | 740.74 | 770.37 | -4% |
| Florence-2 Large images/min | 117.94 | 93.77 | +26% |
| SAM ViT-Huge images/min | 84.22 | 70.46 | +20% |
| TinyLlama 1.1B LoRA train tok/s | 5499.6 | 4756 | +16% |
| TinyLlama 1.1B served serve tok/s | 3558.2 | 2584.3 | +38% |
| Qwen2.5 1.5B LoRA train tok/s | 4214.4 | 3720.9 | +13% |
| Qwen2.5 1.5B served serve tok/s | 2702.5 | 1827 | +48% |
| SmolLM2 1.7B LoRA train tok/s | 3668.3 | 3240.2 | +13% |
| SmolLM2 1.7B served serve tok/s | 2242.7 | 1524.9 | +47% |
| Qwen2.5 1.5B + 0.5B draft x vs solo | 0.919 | 0.968 | -5% |
| Qwen2.5 7B LoRA train tok/s | 1239 | 1063.9 | +16% |
| Qwen2.5 7B served serve tok/s | 858.7 | 469.9 | +83% |
Whole-job comparison
How long each card takes to finish a complete pipeline, not just one model. NVIDIA A10G is faster on 12 of 12; NVIDIA L4 on 0.
| Workflow | NVIDIA A10G | NVIDIA L4 | Difference |
|---|---|---|---|
| Depth pass on a batch | 7 s | 17 s | NVIDIA A10G 2.58x faster |
| Voiceovers from scripts | 28 s | 36 s | NVIDIA A10G 1.28x faster |
| Podcast episode pass | 77 s | 2 min | NVIDIA A10G 1.57x faster |
| Transcribe and subtitle videos | 3.6 min | 4.4 min | NVIDIA A10G 1.24x faster |
| Masking run | 6 min | 7.5 min | NVIDIA A10G 1.24x faster |
| Caption a training dataset | 8.5 min | 10.8 min | NVIDIA A10G 1.27x faster |
| 24-frame storyboard | 9 min | 12.2 min | NVIDIA A10G 1.36x faster |
| Product catalogue cutout | 9.6 min | 12.1 min | NVIDIA A10G 1.26x faster |
| 3D game asset kit | 16.9 min | 22.6 min | NVIDIA A10G 1.33x faster |
| Full codebase review | 21.3 min | 36.4 min | NVIDIA A10G 1.71x faster |
| Photos to 3D models | 50.7 min | 66.2 min | NVIDIA A10G 1.30x faster |
| Short social clips | 50.7 min | 60.3 min | NVIDIA A10G 1.19x faster |
Specifications compared
| NVIDIA A10G | NVIDIA L4 | |
|---|---|---|
| VRAM | 24GB | 24GB |
| Transistors | 28,300M | 35,800M |
| Die size | 628.4 mm² | 294.5 mm² |
| Process node | 8 nm | 4 nm |
| Transistor density | 45 M/mm² | 121.6 M/mm² |
| Architecture | Ampere | Ada Lovelace |
| Memory bandwidth | 600 GB/s | 300 GB/s |
| Boost clock | 1,710 MHz | 2,040 MHz |
| TDP | 150 W | 72 W |
| Launch MSRP | $2,800 | $2,500 |
| Release | 2021-11-01 | 2023-03-21 |
FAQ
- Which is better for ai & machine learning: NVIDIA A10G or NVIDIA L4?
- NVIDIA A10G performs better for ai & machine learning, winning 159 of 186 benchmarks in our suite with an average 53.3% advantage.
- What are the main hardware differences between NVIDIA A10G and NVIDIA L4?
- NVIDIA A10G has 24GB VRAM and a 150W TDP, while NVIDIA L4 has 24GB VRAM and a 72W TDP.
- Does VRAM matter more than speed between NVIDIA A10G and NVIDIA L4?
- For AI, yes, NVIDIA A10G fits 1 more of our 12 workloads than NVIDIA L4. 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 L4?
- EzAudio XL: NVIDIA A10G leads by roughly 33% (0.71 vs 0.38 x realtime) in our testing.
NVIDIA A10G full review · NVIDIA L4 full review · All AI & Machine Learning rankings