OP does not talk about data parallelism vs task parallelism at all. SIMD instructions and GPGPU has come a long way and while clock rates have somewhat stagnat

OP does not talk about data parallelism vs task parallelism at all. SIMD instructions and GPGPU has come a long way and while clock rates have somewhat stagnated, we can solve many problems on consumer hardware today that were impossible in Pentium 3 days (big ML models, augmented reality, live ray tracing, ...). It was a painful transition from existing computation models, but it has been a gigantic success imho.