You want to achieve 30 times more efficiency in AI and HPC by 2025

AMD today announced an update to its efficiency target, revealing that it will seek its products launched under the EPYC and Instinct lines in 2025 to be 30 times more efficient than its products launched under those lines in 2020, breaking with the trend that increases the consumption of data centers exponentially.

Accelerated compute nodes are the most powerful and advanced computing systems in the world used for scientific research and simulations on large-scale supercomputers. These provide the computing power used by scientists across diverse fields including materials science, climate prediction, genomics, drug discovery, and alternative energy. Accelerated nodes are also integral to training neural networks for AI currently used for activities such as speech recognition, language translation, and recommendation systems, with various promising uses in the next decade. This 30X goal will save billions of kilowatt hours of electricity in 2025, reducing the energy required for these systems to complete a single calculation by 97% over these five years.

According to AMD’s predictions, its efficiency increase will be 150% higher than projected for the rest of the industry, so its products would be much more efficient than those of the competition. It is important to note that this projection was made based on the improvements that occurred between 2015 and 2020, so it may not be correct, although in any case, even without comparing against the competition, to achieve that its components reach an efficiency 30 times greater In just 5 years it will be an impressive achievement for the Reds.

Thanks to these changes, AMD expects that the increase in consumption of data centers will be significantly reduced, which currently have an exponential increase in consumption trend towards 2032, which will end up being almost linear with a very low slope of meeting this goal.

It will be interesting to see if AMD ends up meeting this goal, and what improvements it implements to achieve it both at the architectural and process levels, since it will not be easy to achieve such high improvements with the improvements in the nodes planned for the next few years. We will keep you updated on all the news, so stay tuned to HD Tecnología.

What do you think about this challenge from AMD? Will they be able to increase the efficiency of their EPYC processors and Instinct computer cards 30x in just 5 years?