Part 2/9:
Deep Seek V3 showcases an astonishing leap in AI model performance while maintaining a fraction of the training costs typically associated with such advanced technology. Various sources suggest that traditional AI models might require extensive computational resources—up to 16,000 GPUs for training. Conversely, Deep Seek V3 only utilized 2,048 GPUs for two months at a mere cost of $6 million.
While other models like LLaMA 3, which boasts 405 billion parameters, consumed vast GPU resources—30 million hours—Deep Seek V3 achieved similar or superior results at only 2.8 GPU hours. This extraordinary efficiency positions Deep Seek V3 squarely in the competitive landscape of AI language models.