The growing demand for high-performance computers that can support increasingly complex (and expensive) AI models has led engineers to explore new methods for expanding the computational capabilities of their machines.
Now Manya Ghobadi, an associate professor at MIT’s Department of Electrical Engineering and Computer Science (EECS) and a CSAIL member, and her colleagues have developed a solution: a system (dubbed "Lightning”) that connects photons (produced by lasers) to the electronic components (transistors and wires) of computers—creating hybrid photonic-electronic, reconfigurable network interface cards ("SmartNICs).
First photonic-computing prototype to serve real-time, machine-learning inference requests
The new system allows deep neural networks (machine-learning models that imitate how brains process information) to complete inference tasks (like image recognition and large language models generated in chatbots such as ChatGPT).
Machine-learning services completing inference-based tasks, like ChatGPT and BERT, currently require heavy computing resources and are expensive—some estimates show that ChatGPT requires $3 million or more per month to run.
They’re also environmentally detrimental. Instead, Lightning uses photons, which move faster than electrons do in wires while generating less heat.
The team is currently (Sept. 10–14) presenting their findings at the Association for Computing Machinery’s Special Interest Group on Data Communication (SIGCOMM).