AMD exec says AI will not replace traditional HPC
Scott Tease, AMD's VP for HPC and sovereign AI, says most HPC research groups now agree that AI will not replace traditional high-performance computing. High-precision science that relies on FP64 will still be needed even as AI use grows.
Tease says AI is good at low-precision, error-tolerant work such as model training, inference and fast screening. But climate modeling, nuclear energy, materials, fluid dynamics, chemistry and national security still need FP64-level precision to keep complex simulations reliable and repeatable.
He argues future research computing should be a mixed system that connects CPUs, GPUs, AI accelerators, supercomputers and quantum computers, with software deciding which tasks go where. AMD and IBM are running a pilot at Oak Ridge National Laboratory. They are connecting AMD's classic computing stack, the Umbra HPC system and IBM's 129-qubit Quantum Nighthawk quantum computer to the Oak Ridge Leadership Computing Facility. The goal is to explore quantum error correction, AI-assisted science and traditional simulation together.
Comments
Loading comments…