How GPU clock speed affects peak FLOPs

Peak GPU FLOPs depend directly on clock speed alongside parallelism, so frequency remains a first-order performance variable

you're in an ml performance interview at openai and they ask, "how does gpu clock speed affect peak flops?" you respond, "gpus are massively parallel, so clock speed barely matters." wrong! here's how you answer: