H200 NVL
NVIDIA
141GB VRAM · 4890 GB/s bandwidth · 241.3 FP16 TFLOPS · 600W TDP
The H200 NVL has 141GB of VRAM with 4890 GB/s memory bandwidth and 241.3 TFLOPS FP16 compute. At Q4 quantization, it can comfortably run Gemma 3 4B (1760 tok/s), Qwen 2.5 7B (926 tok/s), Llama 3.1 8B (876 tok/s). Models larger than ~240B parameters won't fit even at Q4. Electricity cost is approximately $65/month at 600W TDP.
What LLMs can you run?
| Model | Params | Q4 Weight | Fit | Decode |
|---|---|---|---|---|
| Gemma 3 4B | 4.0B | 2 GB | comfortable | 1760 tok/s |
| Qwen 2.5 7B | 7.6B | 4 GB | comfortable | 926 tok/s |
| Llama 3.1 8B | 8.0B | 4 GB | comfortable | 876 tok/s |
| Mistral Small 24B | 24.0B | 12 GB | comfortable | 293 tok/s |
| Gemma 3 27B | 27.4B | 14 GB | comfortable | 256 tok/s |
| Qwen 2.5 Coder 32B | 32.5B | 16 GB | comfortable | 216 tok/s |
| Llama 3.3 70B | 70.6B | 35 GB | comfortable | 99 tok/s |
| Qwen 2.5 72B | 72.7B | 36 GB | comfortable | 96 tok/s |
| Llama 3.1 405B | 405B | 202 GB | won't fit | |
| DeepSeek R1 671B | 671B | 336 GB | won't fit |
Similar GPUs
| GPU | VRAM | BW | TFLOPS | TDP |
|---|---|---|---|---|
| H200 SXM 141 GB | 141GB | 4890 GB/s | 267.6 | 700W |
| B300 | 144GB | 4099 GB/s | 1231.8 | 1400W |
| Radeon Instinct MI250 | 128GB | 3280 GB/s | 362.1 | 500W |
| Radeon Instinct MI250X | 128GB | 3280 GB/s | 383.0 | 500W |
| Radeon Instinct MI300 | 128GB | 6550 GB/s | 383.0 | 600W |
Compare with another GPU
Select another GPU to compare specs and model performance side by side.