DeepBrain Chain, established in November 2017, aims to create a highly scalable and distributed high-performance computing network utilizing blockchain technology, positioning itself as a vital infrastructure for the upcoming 5G and AI revolution. As society steps into the intelligence age, artificial intelligence is becoming intertwined with daily life, driven by the synergy of deep learning models, extensive data (from the Internet, sensors, and IoT), and robust computing resources (such as GPUs, FPGAs, and specialized chips).
The demand for computational power among individual deep learning models continues to soar. For instance, tasks like ImageNet image recognition require between 1 to 10 GPUs, while advanced models such as AlphaFold and BERT can need 100 to 200 GPUs. Training sophisticated models like GPT-3, which boasts 175 billion parameters, can take up to 1,000 GPUs and cost millions of dollars for a single training session. Furthermore, large multimodal pre-training models may require up to 2,000 GPUs.
The competition in artificial intelligence equates to a race for computing power, making it crucial to address the supply and demand challenges of this resource. DeepBrain Chain aspires to leverage blockchain to create a limitless and decentralized high-performance computing network, enhancing the affordability and efficiency of AI computational resources globally. This initiative aims to democratize access to AI capabilities and hasten the emergence of a truly intelligent era.