CPU “Ah Gan” Explains the Boot Process, Memory Hierarchy, Cache, and Pipelining

The piece opens with a personified CPU named “Ah Gan,” who likens its rapid nanosecond‑scale work to a sprint, noting that a human second contains a billion of its actions and that memory and disk are orders of magnitude slower.

Ah Gan describes waking up in a chassis, hearing the fan, and recalling the creator’s three rules: execute instructions, keep them only in memory, and start at address 0xFFFFFFF0. It contacts memory via the system bus, I/O bridge, and storage bus to fetch the first instruction, which turns out to be a jump to the BIOS.

The BIOS performs the power‑on self‑test, checks memory, disk, and other hardware, and then hands control to the interrupt vector table. Ah Gan looks up interrupt 0x19, which loads the first 512‑byte boot sector from the disk into memory at 0x0000:0x7C00 and transfers execution there.

From the boot sector, a series of sophisticated instructions load the operating system kernel from the hard disk into memory, allowing the OS to take over. Ah Gan notes that while it can execute about 1 ns per instruction, a disk read may take 16 ms, during which millions of instructions could have been processed.

To avoid waiting for the slow disk, Ah Gan suggests using Direct Memory Access (DMA) so the disk can transfer data directly into memory and signal completion.

The narrative then introduces the concept of program locality: memory locations that are accessed frequently tend to be near each other, and recent accesses are likely to be repeated. Ah Gan and memory discuss adding a cache, which stores recently used data and instructions, dramatically speeding up reads.

However, when the OS switches processes, the cache may be invalidated because the new program’s data does not share locality with the previous one, requiring the cache to be rebuilt.

Finally, Ah Gan reflects on its four “hands”: fetching instructions, decoding them, executing them, and writing results back. It observes that only one hand is busy at a time, leading to idle cycles. By adopting a pipeline—analogous to a car‑wash line—each hand can work on a different instruction simultaneously, improving throughput.

The story concludes with Ah Gan shutting down for the day, grateful for the cache and pipeline enhancements, and looking forward to the next day’s work.