Moore’s Law Outstripped by AI Training Advances

December 6, 2021

The time taken to train AI has decreased at a rate that dramatically outpaces predictions made using Moore’s Law, which posits that transistor density (and therefore processing speeds) doubles every two years. Since the first iteration of the MLPerf benchmarks in 2018, systems by Nvidia have seen a 20-fold reduction in training speed. Some training times have been halved since the last results in June 2021.

According to IEEE spectrum, “The increase in transistor density would account for a little more than doubling of performance between the early version of the MLPerf benchmarks and those from June 2021. But improvements to software as well as processor and computer architecture produced a 6.8-11-fold speedup for the best benchmark results.”

Read the entire article by clicking here.

(Source: IEEE Spectrum, December 2nd, 2021)

Share This Story!