Some Comments on Moore’s Law and Super Exponential Growth in Supercomputer Performance
Some prior comments on supercomputing repeated prompted by need in ChatGPT and other Large Language Models, autonomous vehicles and other Digital Twins applications, including the MetaVerse, etc. where huge amounts of supercomputer ppower are required. NVIDIA will be at the forefront.
Some Comments on Moore’s Law
Moore’s Law was coined in the early 1960’s by then-Intel CEO Gordon Moore to describe his forecast that the number of transistors in a computer chip would double every two years. An exponential curve fit (using linear regression to the logs of the counts) which I did to the historical number of transistors on an Intel CPU grew from 2,300 transistors per chip in 1971 to 8 billion transistors per chip in 2017, growing in the 46 year period from 1971 to 2017 at a CAGR of 39.1%, doubling every 2.1 years. See graph below:
Super Exponential Growth in Supercomputer Performance
However, NVIDIA CEO Jensen Huang, in a supercomputer conference earlier this year (Teradec), pointed out that supercomputer performance was growing at a super exponential growth rate. He noted that in the fifteen prior years supercomputer performance had grown 10 trillion-fold, a CAGR of 635%! At this CAGR, supercomputer performance grows over seven-fold “every year”! This is really amazing! Huang announced a GTC that NVIDIA is planning on using this super exponential growth rates in supercomputer performance to create a new supercomputer to study and forecast Climate change by creating an Omniverse Digital Twin of the entire earth. These are very interesting times!
Cheers,
Sully |
