The majority of CPUs and GPUs available today are tailored for dense computing, particularly in the realm of AI, relying on sheer computational power rather than the nuanced approach that our brains employ. Embracing sparsity is a surefire way to enhance performance, but traditional CPUs often struggle to excel in this domain. This is where Numenta steps in, wielding its unique expertise.
This innovative startup seeks to unlock the efficiency gains of sparse computing in AI models by integrating its “secret sauce” into general-purpose CPUs, as opposed to designing chips exclusively for AI-centric workloads.
While Numenta’s breakthrough can be harnessed on both CPUs and GPUs, the startup initially opted for Intel Xeon CPUs and applied Advanced Vector Extensions (AVX)-512 along with Advanced Matrix Extensions (AMX). Intel’s chips were the logical choice due to their widespread availability. These extensions to the x86 architecture introduce additional instruction sets that empower CPUs to tackle more demanding tasks efficiently.
