Actually, to fully utilize NVME performance, one really need to try to avoid OS overhead by leveraging AsyncIO such as IO_Uring. In fact, 4KB page works quite well if you can issue enough outstanding requests. See a paper from the link below by the TUM folks.
It's such a pity to see how much respect Apple has lost due to its lack of real technological ambition. Just a few years ago, Apple was admired for pushing boundaries—now it feels like history is repeating itself.
The article makes two points—one about software performance, the other about development performance. On the topic of software performance, I think the best explanation I got came from the great Prof. Charles Leiserson’s MIT course on performance engineering: performance is a currency. It’s something we spend to gain other qualities—safety, rich features, user-friendliness, and so on. At the end of the day, a computer can executes about 100 simple primitve instructions. The reason we have a software industry is because of speed.
These days, it feels like “AI” basically just means neural network-based models—especially large autoregressive ones. Even convolutional neural networks probably don’t count as “real AI” anymore in most people’s eyes. Funny how things change. Not long ago, search algorithms like A* were considered the cutting edge of AI.
https://dl.acm.org/doi/abs/10.14778/3598581.3598584