Google Keynote Charts ML Drivers At Hot Chips 23
HPCWire, Thursday, August 31,2023
Google scientists Jeff Dean and Amin Vahdat delivered a fascinating tour of major ML hardware and software design trends in their joint Hot Chips 23 opening keynote this week. The pair tackled the rise of sparsity over density, efforts at adaptive communications, the development of better power consumption and system performance metrics, and AI-driven accelerated chip design, among other topics.
Dean and Vahdat left few stones unturned, although no closely-guarded Google news was revealed. Some of the latter seems likely to emerge at the Google Cloud Next 23 event, which starts today in San Francisco. Nevertheless, the broad brushstrokes from Dean and Vahdat emphasized the need for speeding up progress as demand for ML compute driven by many factors, including of course, the soaring size (number of parameters) of models outpaces infrastructure capability.