Scott Tease, AMD's Corporate Vice President for High Performance Computing and Sovereign AI, stated that AI will not replace traditional high-performance computing, and that FP64 high-precision computing remains indispensable in scientific research. AI is better suited for low-precision, highly fault-tolerant data tasks and is primarily used for model training, inference, and rapid data screening. In contrast, frontier scientific research scenarios such as climate simulation and nuclear energy research impose rigid requirements on the reliability and reproducibility of computational results, relying on FP64 double-precision computing power—areas where AI cannot substitute. He believes the future direction of scientific research computing lies in building heterogeneous computing systems that integrate various types of computing power collaboratively. Currently, AMD is conducting a pilot with IBM at Oak Ridge National Laboratory to test multi-technology integration solutions, exploring implementation paths that combine quantum error correction, AI-assisted scientific research, and traditional numerical simulation, while tapping into the potential of heterogeneous computing power in frontier scientific research.
