DeepSeek-V3 required 2.788 million NVIDIA H800 GPU-hours for its full training.

Format Online paper / PDF

"DeepSeek-V3 required 2.788 million NVIDIA H800 GPU-hours for its full training."

DeepSeek's technical report states that full training required 2.788 million H800 GPU-hours.

Source title DeepSeek-V3 Technical Report

Source authors DeepSeek-AI

Source year 2024

Confidence high

Importance high