Path: csiph.com!eternal-september.org!feeder.eternal-september.org!nntp.eternal-september.org!.POSTED!not-for-mail From: John Ames Newsgroups: comp.os.linux.misc Subject: CPU architecture (was Re: Mercury) Date: Thu, 2 Apr 2026 10:43:02 -0700 Organization: A place where nothing fits quite right Lines: 55 Message-ID: <20260402104302.00007623@gmail.com> References: <1rsoqz0.19zzbh71ebfb7bN%snipeco.2@gmail.com> <10qb9df$1inu5$14@dont-email.me> <1rsr591.1ssq8oh1dihjwuN%nospam@de-ster.demon.nl> <10qc4dc$1silm$10@dont-email.me> <0cqjskp5oprp9v1utu6t3q8u0urkpnjbvs@4ax.com> <97uq9mxp8h.ln2@Telcontar.valinor> <10qj2u6$63en$10@dont-email.me> <8f2cnUuvT4KC8lD0nZ2dnZfqnPidnZ2d@giganews.com> <8f2cnUqvT4JIKFD0nZ2dnZfqnPgAAAAA@giganews.com> <8f2cnUWvT4L8SFD0nZ2dnZfqnPidnZ2d@giganews.com> MIME-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Injection-Date: Thu, 02 Apr 2026 17:43:07 +0000 (UTC) Injection-Info: dont-email.me; posting-host="0efebc393f35cca95110794f8de8bb40"; logging-data="1376647"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX19THC/16+dFq2PayL8QwCGD0pGFQrgByVI=" Cancel-Lock: sha1:hzOOFH1jaIreYFnDvrDWhNDBL0Q= X-Newsreader: Claws Mail 4.3.0 (GTK 3.24.42; x86_64-w64-mingw32) Xref: csiph.com comp.os.linux.misc:85166 On Wed, 1 Apr 2026 22:32:57 -0400 c186282 wrote: > > Nobody can explain or understand a modern CPU from scratch. > > > > And I could not write the explanation of a simple CPU, either. I > > have forgotten many things. > > SOME people "get it" - but very few. It's like asking Mary-Lou how > the metal in her car engine is made/machined and why which parts are > as big/thick/shaped as they are. The engineering just gets deeper and > deeper. The basics aren't really that complex; a Platonic CPU is just a state machine sequencing the operation of an ALU and register file according to a rules table, with some of its outputs directing the flow from one state to another, and the clock latching the prior out-state to become the in-state for the next operation, like a two-stroke engine. while (!halt) { state = rulesTable[state]; } Where it gets complicated is in the scale (how big is the word size? How many functions do you break out into separate circuits vs. using a general-purpose ALU?) and the tricks we pull to maximize speed (how many sub-operations of a larger instruction can be done in parallel? How nimbly can you keep multiple cores from stepping on each others' toes? How effectively can you apply caching to reduce memory bottle- necks and keep data flowing into & out of the CPU? How many stages can you put in a pipeline before the cost of a missed branch prediction eats any gains in raw clock speed--trick question, the answer is "the Pentium 4...") *That's* what takes it from "tuning a two-stroke engine" territory to "optimizing a factory floor..." > Likely the last CPU gurus are now instructing AIs on how to carry on > the craft before they get too old and senile. The next-gen ... only > the AIs will "get it" - humans won't - and thus it becomes "magic" > like in the ancient days. That's certainly the dream of the C-suite - but the truth is that the bubble is already bursting. OpenAI just killed its most expensive feature, mere months after announcing a billion-dollar deal with Disney for it. Oracle slashed 10k jobs because they're up to their eyeballs in debt for their data-center build-out. Even private credit is starting to go "ehhh, no thanks" when approached for new data-center funding deals, and piles of already-announced projects are behind schedule, stalled, or not even *started.* I wouldn't put money on *anything* in this clownshow of a timeline, but I would not be at all surprised to see the whole house-o'-cards collapse before the end of next year, if not sooner.