To the best of my understanding, AMD/Nvidia/Intel each run their own forms of architecture (eg. AMDs RDNA) which are probably closest to RISC for simple instructions and SIMD/SIMT (single instruction, multiple data/threads) for more complex vector calculations.