Forscher der Tsinghua-Universität in China haben einen revolutionären neuen KI-Chip entwickelt, der zur Datenverarbeitung Licht statt Strom nutzt. Der „Taichi“ genannte Chip soll über 1.000-mal energieeffizienter sein als Nvidias Hochleistungs-GPU-Chip H100
https://interestingengineering.com/science/taichi-light-based-chip
2 Kommentare
Dont get too excited. Universities don’t have resources to develop a fully operational product, it takes billions. It’s probably a tiny demonstrator. But the idea to use photonics in AI is indeed great.
„Taichi’s modular design could be a key advantage in scaling up computing power to the levels needed for AGI. The Tsinghua team demonstrated a distributed Taichi network with 13.96 million artificial neurons, exceeding the 1.47 million neurons in the next largest competing photonic chip design.“
„Taichi achieved an energy efficiency of 160.82 trillion operations per watt (TOPS/W), dwarfing the previous 2.9 TOPS/W achieved by another team in 2022, LiveScience reports. This dramatic improvement in energy efficiency is crucial for the sustainability of AI computing, which has become an increasingly power-hungry industry.“
The team’s research was published in the journal Science [https://www.science.org/doi/10.1126/science.adl1203](https://www.science.org/doi/10.1126/science.adl1203)