"GPUs [...] by no means limited to machine learning"
Indeed. With a good chunk of contemporary AI being essentially "a dumb person's idea of intelligence" (esp. LLMs), it's good to see Synopsys sedately focusing its GPU use, 4-to-1, to parallel acceleration of conventional workloads (linked under "touted"): PrimeSim™/SPICE, Proteus™ OPC/ILT litho, Sentaurus™ TCAD, and QuantumATK®, with 10x to 100x speedups, vs "Synopsys.ai Copilot", imho!