DeepSeek-R1 (671B MoE) VRAM Calculator
Official DeepSeek-R1 (671B MoE) model by DeepSeek. Calculate hardware limits, context VRAM usage, and local inference requirements.
LLM (Language Model)Developer: DeepSeek
Recommended GPU: 8x H100 / Mac Studio M2 Ultra 192GB (CPU Offload)
12 GB
8,192 tokens
Estimated Total VRAM
410.5GB
VRAM Usage Ratio100% (410.5 / 12 GB)
Memory Allocation Breakdown
Model Weights378.7 GB
KV Cache30.5 GB
CUDA Runtime1.3 GB
⚠️ CUDA Out of Memory Warning
+398.5 GB exceeds your GPU limit. This configuration will trigger CUDA OOM or heavy memory swapping.