AI has pushed an explosion of latest quantity codecs—the methods through which numbers are represented digitally. Engineers are each attainable means to save lots of computation time and power, together with shortening the variety of bits used to symbolize knowledge. However what works for AI doesn’t essentially work for scientific computing, be it for computational physics, biology, fluid dynamics, or engineering simulations. IEEE Spectrum spoke with Laslo Hunhold, who not too long ago joined Barcelona-based Openchip as an AI engineer, about his efforts to develop a bespoke quantity format for scientific computing.
LASLO HUNHOLD
Laslo Hunhold is a senior AI accelerator engineer at Barcelona-based startup Openchip. He not too long ago accomplished a Ph.D. in laptop science from the College of Cologne, in Germany.
What makes quantity codecs fascinating to you?
Laslo Hunhold: I don’t know one other instance of a discipline that so few are excited by however has such a excessive impression. When you make a quantity format that’s 10 p.c extra [energy] environment friendly, it might translate to all purposes being 10 p.c extra environment friendly, and it can save you a number of power.
Why are there so many new quantity codecs?
Hunhold: For many years, laptop customers had it very easy. They may simply purchase new methods each few years, and they’d have efficiency advantages free of charge. However this hasn’t been the case for the final 10 years. In computer systems, you will have a sure variety of bits used to symbolize a single quantity, and for years the default was 64 bits. And for AI, corporations seen that they don’t want 64 bits for every quantity. So that they had a robust incentive to go right down to 16, 8, and even 2 bits [to save energy]. The issue is, the dominating customary for representing numbers in 64 bits shouldn’t be properly designed for decrease bit counts. So within the AI discipline, they got here up with new codecs that are extra tailor-made towards AI.
Why does AI want totally different quantity codecs than scientific computing?
Hunhold: Scientific computing wants excessive dynamic vary: You want very massive numbers, or very small numbers, and really excessive accuracy in each instances. The 64-bit customary has an extreme dynamic vary, and it’s many extra bits than you want more often than not. It’s totally different with AI. The numbers often comply with a particular distribution, and also you don’t want as a lot accuracy.
What makes a quantity format “good”?
Hunhold: You’ve gotten infinite numbers however solely finite bit representations. So it is advisable to determine the way you assign numbers. Crucial half is to symbolize numbers that you just’re truly going to make use of. As a result of when you symbolize a quantity that you just don’t use, you’ve wasted a illustration. The best factor to have a look at is the dynamic vary. The subsequent is distribution: How do you assign your bits to sure values? Do you will have a uniform distribution, or one thing else? There are infinite prospects.
What motivated you to introduce the takum quantity format?
Hunhold: Takums are based mostly on posits. In posits, the numbers that get used extra continuously might be represented with extra density. However posits don’t work for scientific computing, and it is a big situation. They’ve a excessive density for [numbers close to one], which is nice for AI, however the density falls off sharply when you take a look at bigger or smaller values. Individuals have been proposing dozens of quantity codecs in the previous couple of years, however takums are the one quantity format that’s truly tailor-made for scientific computing. I discovered the dynamic vary of values you employ in scientific computations, when you take a look at all of the fields, and designed takums such that while you take away bits, you don’t cut back that dynamic vary
This text seems within the March 2026 print situation as “Laslo Hunhold.”
From Your Website Articles
Associated Articles Across the Net
