Comment by achierius

Just because CPU performance is increasing faster than DRAM speeds doesn't mean that CPU performance is "free" while memory is "expensive". One thing that you're ignoring is the impact of caches and prefetching logic which have significantly improved the performance of memory-bound workloads in the last 5-10 years. DRAM might be slow, but if you avoid going out to DRAM...

More broadly, it 100% depends on your workload. You'd be surprised at how many workloads are compute-bound, even today: LLM inference might be memory bound (thus how it's possible to get remotely good performance on a CPU), but training, esp. prefill, is very much not. And once you get out of LLMs, I'd say that most applications of ML tend to be compute bound.