A faster way to estimate AI power consumption
Reports and Proceedings
Updates every hour. Last Updated: 12-May-2026 18:16 ET (12-May-2026 22:16 GMT/UTC)
The EnergAIzer technique can predict how much power a certain AI workload will consume when run on a particular processor. This method could help data center operators and algorithm developers improve the sustainability of AI workloads.
A research team led by the Ningbo Institute of Materials Technology and Engineering (NIMTE), CAS, has achieved a breakthrough in spintronics by demonstrating that nonsymmorphic symmetry in hexagonal SrIrO3 protects topological Dirac semimetal states. This unique electronic structure leads to record-breaking charge-spin conversion efficiency, enabling magnetic switching with ultra-low power dissipation. The study establishes a robust and universal criterion for designing future energy-efficient spintronic devices.
A new ultra-efficient microchip developed at MIT enables resource-constrained wireless biomedical devices to implement post-quantum encryption protocols that can defend against cyberattacks from powerful quantum computers.
New research examines Finnish lower secondary special needs math instruction via a survey of teachers, focusing on topics taught, preparedness, and instructional practices. Teachers report high confidence overall, but gaps remain in data processing, statistics, and probability. While guided practice and feedback are common, high-impact strategies such as mastery learning, scaffolding, peer tutoring, and structured sequences are used inconsistently. Findings reveal a disconnect between preparation and classroom practice, highlighting the need for targeted training and support.
In the book, “Priority Technologies,” MIT faculty analyze how the U.S. can move ahead in multiple key industrial sectors — semiconductors, biotechnology, critical minerals, drones, quantum computing, and advanced manufacturing — to drive the economy and support national security.