A group of Chinese researchers published a paper describing how they used AI to automate the creation of a CPU about as fast as an Intel i486SX (via Tom’s Hardware) in what they called a “foundational step towards building self-evolving machines.” The Intel i486SX was an early-90s entry-level enterprise CPU.

Left: The outputs of booting up the Linux operating system. Right: The performance of generated CPU (i.e., CPU-AI) is compared against commercial CPUs on the Dhrystone benchmark, and CPU-AI performs comparably to the human-designed Intel 80486SX CPU.

  • @SheeEttinEnglish
    8•1 year ago

    In addition, the generated Boolean function is almost zero tolerance to inaccuracy, otherwise, CPUs will be malfunctioned and cause a huge amount of loss. A recent case in 2017 is that the Intel Atom C2000 bug affected many famous vendors, among which Cisco prepared 125M dollars to replace related products32. […] In this article, we report a RISC-V CPU automatically designed by a new AI approach, which generates large-scale Boolean function with almost 100% validation accuracy (e.g., > 99.99999999999% as Intel)

    Oh good, that was going to be my question. A new CPU is all well and good, but if you don’t validate it to prove it doesn’t have some weird bug, then it’s worthless.