WorkThe story, in brief
AI and efficiency
Algorithmic progress is outpacing Moore's Law by 4x. Here's what that means for your AI roadmap.
The KeyNews take
Why it matters
OpenAI's analysis reveals that algorithmic efficiency gains have delivered 44x compute reduction since 2012—far exceeding hardware improvements alone. This challenges conventional wisdom about the compute ceiling and reframes how teams should think about training cost trajectory.
The key facts
5 to knowCompute required to train neural nets to ImageNet performance halves every 16 months since 2012
44x total compute reduction since 2012 vs. 11x from Moore's Law alone
Algorithmic progress accounts for ~75% of efficiency gains (4x multiplier advantage over hardware)
Published May 2020 by OpenAI
Focused on high-investment AI tasks with sustained R&D
Go to the source
OpenAI Blogopenai.com
Publisher excerpt: We’re releasing an analysis showing that since 2012 the amount of compute needed to train a neural net to the same performance on ImageNet classification has been decreasing by a factor of 2 every 16 months. Compared to 2012, it now takes 44 times less compute to train a neural network to the level…