Sürekli güncellenen araştırma, hızla değişen yapay zekâ donanımını haritalıyor

Since 2018, researchers at MIT Lincoln Laboratory’s Supercomputing Center have tracked commercial AI accelerators through the Lincoln AI Computing Survey, known as LAICS. These specialised systems speed up machine learning and other parallel tasks, including molecular modelling and fluid-dynamics simulations. The survey compares publicly available figures for peak performance and peak power across CPUs, GPUs, ASICs, FPGAs and dataflow accelerators, whose flexibility and strengths differ by workload. Its first paper covered 57 accelerators; the latest examined more than 120. Each edition also investigates a new issue, such as how smaller transistors and lower numerical precision have contributed to performance gains, or how architectural choices alter a system. Collecting comparable data is difficult because some manufacturers keep specifications private, while new startups and designs continue to appear. By maintaining a neutral, current overview, the team helps research sponsors and computing specialists make better acquisition choices, including decisions about future systems at the centre itself.