AMD updates efficiency roadmap: 30x targeted for AI and HPC by 2025

AMD today announced an update to its efficiency target, announcing that the company aims to make its EPYC and Instinct line products 30 times more efficient in 2025 than products launched under those lines in 2020 to address the trend of exponentially increasing power consumption in data centers.

AMD updates efficiency roadmap: 30x targeted for AI and HPC by 2025

Accelerated Computing Nodes are the most powerful and advanced computing systems in the world, used for scientific research and simulations on large supercomputers. They provide the computing power used by scientists in fields as diverse as materials science, climate prediction, genomics, drug discovery, and alternative energy.

Accelerated nodes are also an essential part of training neural networks for artificial intelligence, which are currently used for activities such as speech recognition, language translation, and recommender systems, and for which there are many promising applications in the next decade. This 30x target will save billions of kilowatt-hours of electricity by 2025 by reducing the energy these systems need to perform a single computation by 97% over five years.

Related:   Intel Core i5-11400 is 10% faster than i5-10400

AMD predicts that efficiency gains will be 150% higher than the rest of the industry, making AMD’s products significantly more efficient than their competitors. It’s important to note that this prediction was made based on improvements between 2015 and 2020, so it may not be accurate, but even without comparing it to the competition, it would be an impressive feat for the Reds to see their components achieve a 30x increase in efficiency in just five years.

Thanks to these changes, AMD expects to significantly reduce the increase in data center power consumption, which is currently growing exponentially, by 2032, so that if this goal is reached, it will be almost linear with a very small slope.

It will be interesting to see if AMD can achieve this goal and what improvements it implements at both the architecture and process levels, as it will not be easy to achieve such high improvements with the node upgrades planned for the next few years.