NVIDIA H100 80GB HBM3 vs NVIDIA T4, AI & Machine Learning Comparison
NVIDIA H100 80GB HBM3 wins 146 of 149 benchmarks, averaging 494.5% faster.
Both cards were measured first-party on our bench, same suite, same test rig.
What the numbers say
The gap is widest in Stable Diffusion XL, where NVIDIA H100 80GB HBM3 leads by 1365% (34.58 vs 2.36 images/min); the closest fight is Qwen2.5 1.5B + 0.5B draft (24% apart); VRAM decides part of this one: NVIDIA H100 80GB HBM3 runs 156 of our 12 AI workloads while the other card runs 111, models that don't fit score zero.
Benchmark results head-to-head
| Benchmark | NVIDIA H100 80GB HBM3 | NVIDIA T4 | Difference |
|---|---|---|---|
| Qwen3 4B tok/s | 310.26 | 67.71 | +358% |
| Llama 3.1 8B tok/s | 261.83 | 35.01 | +648% |
| Qwen2.5-Coder 14B tok/s | 144.84 | 19.76 | +633% |
| Qwen3 32B tok/s | 74.07 | 0 | n/a |
| Llama 3.3 70B tok/s | 41 | 0 | n/a |
| Stable Diffusion XL images/min | 34.58 | 2.36 | +1365% |
| FLUX.1 dev images/min | 9.064 | 0 | n/a |
| FLUX.1 Kontext dev images/min | 4.264 | 0 | n/a |
| Qwen-Image-Edit images/min | 3.68 | 0 | n/a |
| Wan 2.2 5B (720p) frames/s | 1.33 | 0 | n/a |
| DeepSeek-R1 Distill Llama 8B tok/s | 261.15 | 36.36 | +618% |
| DeepSeek-R1 Distill 1.5B tok/s | 537.32 | 148.72 | +261% |
| DeepSeek-R1 Distill 7B tok/s | 264.42 | 37.58 | +604% |
| Gemma 3 12B tok/s | 150.58 | 23.12 | +551% |
| Gemma 3 4B tok/s | 289.61 | 65.12 | +345% |
| Llama 3.2 1B tok/s | 880.64 | 207.71 | +324% |
| Llama 3.2 3B tok/s | 422.87 | 85.94 | +392% |
| Mistral 7B v0.3 tok/s | 275.85 | 40.01 | +589% |
| Phi-4 Mini 3.8B tok/s | 388.86 | 64.15 | +506% |
| Qwen2.5-Coder 7B tok/s | 263.91 | 37.54 | +603% |
| Qwen3 0.6B tok/s | 713.53 | 263.47 | +171% |
| Qwen3 1.7B tok/s | 556.48 | 139.92 | +298% |
| Qwen3 14B tok/s | 151.16 | 19.31 | +683% |
| Qwen3 8B tok/s | 244.22 | 36.51 | +569% |
| SmolLM3 3B tok/s | 397.99 | 86.08 | +362% |
| DeepSeek-R1 Distill 14B tok/s | 144.82 | 19.21 | +654% |
| Gemma 4 12B tok/s | 148.17 | 24.03 | +517% |
| Mistral Small 24B tok/s | 105.62 | 11.33 | +832% |
| Phi-4 14B tok/s | 165.17 | 17.46 | +846% |
| Qwen3 30B A3B tok/s | 283.78 | 0 | n/a |
| Codestral 22B tok/s | 110.05 | 12.59 | +774% |
| DeepSeek-R1 Distill 32B tok/s | 75.8 | 0 | n/a |
| Devstral Small 24B tok/s | 107.79 | 11.4 | +846% |
| Dolphin 2.9.1 Yi 1.5 34B tok/s | 76.76 | 0 | n/a |
| Dolphin Mistral 24B Venice tok/s | 109.13 | 11.36 | +861% |
| Dolphin X1 8B tok/s | 266 | 35.66 | +646% |
| Dolphin 3.0 Llama 3.1 8B tok/s | 266.19 | 35.68 | +646% |
| Dolphin 3.0 R1 Mistral 24B tok/s | 107.83 | 11.32 | +853% |
| Gemma 3 27B tok/s | 84.77 | 0 | n/a |
| Qwen2.5-Coder 32B tok/s | 75.81 | 0 | n/a |
| Qwen3-Coder 30B A3B tok/s | 296.22 | 0 | n/a |
| QwQ 32B tok/s | 75.82 | 0 | n/a |
| StarCoder2 15B tok/s | 134.92 | 15.66 | +762% |
| FLUX.1 Schnell images/min | 58.36 | 0 | n/a |
| BiRefNet images/min | 1310.97 | 183.48 | +615% |
| Depth Anything V2 Large images/min | 918.71 | 402.34 | +128% |
| Depth Anything V2 Small images/min | 1081.54 | 612.25 | +77% |
| SAM ViT-Base images/min | 1431.27 | 180.73 | +692% |
| SAM ViT-Huge images/min | 320.64 | 34.91 | +818% |
| Swin2SR 4x Upscaler images/min | 25.48 | 10.53 | +142% |
| Kokoro TTS 82M x realtime | 222.02 | 42.88 | +418% |
| MusicGen Small x realtime | 2.47 | 0.92 | +168% |
| Whisper large-v3 x realtime | 181.47 | 44.36 | +309% |
| Florence-2 Base images/min | 249.62 | 148.61 | +68% |
| Florence-2 Large images/min | 156.15 | 83.43 | +87% |
| Qwen2.5 1.5B LoRA train tok/s | 16034.9 | 1434.4 | +1018% |
| Qwen2.5 7B LoRA train tok/s | 8514.3 | 0 | n/a |
| SmolLM2 1.7B LoRA train tok/s | 18910.3 | 1561.9 | +1111% |
| TinyLlama 1.1B LoRA train tok/s | 17522 | 1520.1 | +1053% |
| Qwen2.5 1.5B served serve tok/s | 6794.8 | 1337.1 | +408% |
| Qwen2.5 7B served serve tok/s | 3601.1 | 0 | n/a |
| SmolLM2 1.7B served serve tok/s | 6509.9 | 1222.5 | +433% |
| TinyLlama 1.1B served serve tok/s | 8336.7 | 1897.8 | +339% |
| Qwen2.5 1.5B + 0.5B draft x vs solo | 0.725 | 0.9 | -19% |
| AI21-Jamba-Reasoning-3B tok/s | 363.09 | 86.91 | +318% |
| Olmo-3.1-32B-Think tok/s | 75.53 | 0 | n/a |
| Codestral 22B (Q3_K_M) tok/s | 87.21 | 11.3 | +672% |
| Dolphin-Mistral-24B-Venice-Edition tok/s | 105.63 | 11.33 | +832% |
| DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored tok/s | 262.76 | 35.73 | +635% |
| DeepSeek-Coder-V2-Lite tok/s | 309.14 | 83.49 | +270% |
| DeepSeek-R1-0528-Qwen3-8B tok/s | 245.26 | 37.7 | +551% |
| DeepSeek-R1-Distill-Llama-70B tok/s | 41.17 | 0 | n/a |
| DeepSeek-R1 Distill 14B (Q3_K_M) tok/s | 118.43 | 16.86 | +602% |
| DeepSeek-R1-Distill-Qwen-32B-abliterated tok/s | 73.48 | 0 | n/a |
| dolphin-2.9-llama3-8b tok/s | 262.47 | 35.1 | +648% |
| Dolphin X1 Trinity Nano 6B tok/s | 247.94 | 114.21 | +117% |
| EVA-Qwen2.5-14B-v0.2 tok/s | 145.15 | 19.92 | +629% |
| gemma-2-2b-it-abliterated tok/s | 392.65 | 91.91 | +327% |
| gemma-2-2b tok/s | 392.26 | 92.13 | +326% |
| gemma-2-9b tok/s | 174.7 | 28.62 | +510% |
| Gemma 3 12B (Q3_K_M) tok/s | 125.92 | 19.79 | +536% |
| gemma-3-1b tok/s | 504.17 | 156.88 | +221% |
| gemma-3-270m tok/s | 961.29 | 362.8 | +165% |
| GLM-4.7-Flash-REAP-23B-A3B tok/s | 168.41 | 53.83 | +213% |
| GLM-4.7-Flash tok/s | 186.05 | 0 | n/a |
| Josiefied-Qwen3-8B-abliterated-v1 tok/s | 245 | 35.62 | +588% |
| gpt-oss-20b tok/s | 346.8 | 63.63 | +445% |
| Hermes-3-Llama-3.2-3B tok/s | 423.72 | 84.99 | +399% |
| Hermes-4-70B tok/s | 41.2 | 0 | n/a |
| SmolLM3-3B tok/s | 399.14 | 86.86 | +360% |
| Qwen3-Coder-Next-abliterated tok/s | 190.32 | 0 | n/a |
| KAT-Coder-V2.5-Dev tok/s | 231.83 | 0 | n/a |
| L3-8B-Stheno-v3.2 tok/s | 261.83 | 36.28 | +622% |
| Laguna-XS-2.1 tok/s | 0 | 0 | n/a |
| LFM2.5-1.2B tok/s | 926.61 | 221.66 | +318% |
| LFM2.5-8B-A1B tok/s | 581.98 | 139.03 | +319% |
| Llama-2-7B tok/s | 284.62 | 42.98 | +562% |
| Llama-3.2-3B-Instruct-uncensored tok/s | 424.25 | 86.29 | +392% |
| Llama-3.3-70B-Instruct-abliterated tok/s | 41.15 | 0 | n/a |
| Meta-Llama-3.1-70B tok/s | 41.19 | 0 | n/a |
| Meta-Llama-3.1-8B tok/s | 261.7 | 37.8 | +592% |
| Phi-4-mini tok/s | 381.52 | 66 | +478% |
| Mistral-7B-Instruct-v0.1 tok/s | 276.29 | 40.63 | +580% |
| Mistral-7B-Instruct-v0.2 tok/s | 276.73 | 41.31 | +570% |
| Mistral-7B-Instruct-v0.3 tok/s | 276.61 | 39.87 | +594% |
| Mistral-Nemo-Instruct-2407 tok/s | 177.38 | 22.98 | +672% |
| Mistral Small 24B (Q3_K_M) tok/s | 84 | 10.13 | +729% |
| Nanbeige4.2-3B tok/s | 0 | 0 | n/a |
| NemoMix-Unleashed-12B tok/s | 177.28 | 23.19 | +664% |
| Nemotron-3-Nano-30B-A3B tok/s | 316.51 | 0 | n/a |
| Hermes-4-14B tok/s | 152 | 19.02 | +699% |
| Ornith-1.0-35B tok/s | 221.65 | 0 | n/a |
| Ornith-1.0-9B tok/s | 216.89 | 32.93 | +559% |
| phi-2 tok/s | 340.09 | 89.6 | +280% |
| Phi-3.5-mini tok/s | 345.07 | 72.07 | +379% |
| Phi-4 14B (Q3_K_M) tok/s | 138.87 | 16.04 | +766% |
| Qwen-AgentWorld-35B-A3B tok/s | 219.1 | 0 | n/a |
| Qwen3-0.6B tok/s | 710.97 | 259.91 | +174% |
| Qwen3-1.7B tok/s | 557.09 | 135.21 | +312% |
| Qwen3-14B tok/s | 152.2 | 19.44 | +683% |
| Qwen3-30B-A3B tok/s | 285.44 | 0 | n/a |
| Qwen3-4B-Instruct-2507 tok/s | 310.58 | 66.89 | +364% |
| Qwen3-8B tok/s | 244.2 | 37.37 | +553% |
| Qwen3-Coder-Next tok/s | 188.21 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 188.14 | 0 | n/a |
| Qwen1.5-0.5B tok/s | 847.02 | 328.29 | +158% |
| Qwen2-1.5B tok/s | 536.52 | 147.68 | +263% |
| Qwen2.5-0.5B tok/s | 890 | 295.47 | +201% |
| Qwen2.5-1.5B tok/s | 537.24 | 148.14 | +263% |
| Uncensored tok/s | 144.92 | 18.94 | +665% |
| Qwen2.5-14B tok/s | 144.7 | 19.84 | +629% |
| Qwen2.5-32B tok/s | 73.5 | 0 | n/a |
| Qwen2.5-3B tok/s | 395.52 | 86.04 | +360% |
| Qwen2.5-72B tok/s | 40.78 | 0 | n/a |
| Qwen2.5-7B tok/s | 264.17 | 39.01 | +577% |
| Qwen2.5-Coder-0.5B tok/s | 892.81 | 294.96 | +203% |
| Qwen2.5-Coder-1.5B tok/s | 538.44 | 148.51 | +263% |
| Qwen2.5-Coder-14B-Instruct-abliterated tok/s | 145.01 | 18.86 | +669% |
| Qwen2.5-Coder 32B (Q3_K_M) tok/s | 58.05 | 0 | n/a |
| Qwen2.5-Coder-3B tok/s | 394 | 85.74 | +360% |
| Qwen2.5-Coder-7B-Instruct-abliterated tok/s | 264.24 | 37.31 | +608% |
| Qwen3 30B A3B (Q3_K_M) tok/s | 242.6 | 61.82 | +292% |
| Qwen3-4B-Instruct-2507 tok/s | 311.47 | 67.45 | +362% |
| Qwen3-4B-Thinking-2507 tok/s | 310.84 | 67 | +364% |
| Qwen3-Coder-Next tok/s | 193.79 | 0 | n/a |
| Qwen3-Next-80B-A3B-Thinking tok/s | 191.08 | 0 | n/a |
| Qwen3-Next-80B-A3B tok/s | 186.22 | 0 | n/a |
| SmolLM2-135M tok/s | 898.53 | 416.24 | +116% |
| Cydonia-24B-v4.3 tok/s | 105.49 | 11.21 | +841% |
Whole-job comparison
How long each card takes to finish a complete pipeline, not just one model. NVIDIA H100 80GB HBM3 is faster on 4 of 4; NVIDIA T4 on 0.
| Workflow | NVIDIA H100 80GB HBM3 | NVIDIA T4 | Difference |
|---|---|---|---|
| 50-image depth pass | 5 s | 13 s | NVIDIA H100 80GB HBM3 2.43x faster |
| 500-image masking run | 1.6 min | 14.4 min | NVIDIA H100 80GB HBM3 8.93x faster |
| Full codebase review | 6.9 min | 50.7 min | NVIDIA H100 80GB HBM3 7.34x faster |
| 200-product catalogue cutout | 8.1 min | 20.3 min | NVIDIA H100 80GB HBM3 2.49x faster |
Specifications compared
| NVIDIA H100 80GB HBM3 | NVIDIA T4 | |
|---|---|---|
| VRAM | 80GB | 16GB |
| Architecture | Hopper | Turing |
| Memory bandwidth | 3350 GB/s | 320 GB/s |
| Boost clock | 1,980 MHz | 1,590 MHz |
| TDP | 700 W | 70 W |
| Launch MSRP | $30,000 | $2,299 |
| Release | 2022-09-20 | 2018-09-13 |
FAQ
- Which is better for ai & machine learning: NVIDIA H100 80GB HBM3 or NVIDIA T4?
- NVIDIA H100 80GB HBM3 performs better for ai & machine learning, winning 146 of 149 benchmarks in our suite with an average 494.5% advantage.
- What are the main hardware differences between NVIDIA H100 80GB HBM3 and NVIDIA T4?
- NVIDIA H100 80GB HBM3 has 80GB VRAM and a 700W TDP, while NVIDIA T4 has 16GB VRAM and a 70W TDP.
- Does VRAM matter more than speed between NVIDIA H100 80GB HBM3 and NVIDIA T4?
- For AI, yes, NVIDIA H100 80GB HBM3 fits 42 more of our 12 workloads than NVIDIA T4. 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 H100 80GB HBM3 and NVIDIA T4?
- Stable Diffusion XL: NVIDIA H100 80GB HBM3 leads by roughly 1365% (34.58 vs 2.36 images/min) in our testing.
NVIDIA H100 80GB HBM3 full review · NVIDIA T4 full review · All AI & Machine Learning rankings