3 ms·
> Consequently, the x86 processors in today’s PCs may still appear to be executing software-compatible CISC instructions, but, as soon as those instructions cro
by mtgx 9y ago
> Consequently, the x86 processors in today’s PCs may still appear to be executing software-compatible CISC instructions, but, as soon as those instructions cross over from external RAM into the processor, an instruction chopper/shredder slices and dices x86 machine instructions into simpler “micro-ops” (Intel-Speak for RISC instructions) that are then scheduled and executed on multiple RISC execution pipelines. Today’s x86 processors got faster by evolving into RISC machines.
Going by that and the graph, then we can conclude that Intel saw the rapid 90's and early 2,000's gains because it was converting its chips into RISC chips?
Also, that paragraph is basically saying that Intel's architecture has an extra layer of abstraction - so now we actually see that there is indeed an "x86 bloat" and why ARM chips seem to be so much more efficient (assuming all else, including process node is equal). It also looks like Intel may have made a "mistake" going with CISC decades ago, and it tried to rectify that in the 90's.
- russdill 9y agoEarly 2000s was just cranking frequency higher and higher with longer and longer pipelines. Initially this gave higher performance, but towards the end it mainly just gave higher frequencies. Intel finally gave up on the frequency wars (longer pipelines) and instead concentrated on wider datapath/pipelines. See: Netburst vs Core. I really feel like the changeover needs it's own act.
- blattimwind 9y agoWhich is not true, since ARM processors - the faster ones anyway - use micro-ops as well. Arguably µops are not RISC, either, unless you consider "very wide instruction word whose bits map to control lines" RISC.
- Narishma 9y agoARM isn't exactly your typical RISC either.