Ask HN: Domain Specific Hardware – next big wave? Intel's Spectre/Meltdown bugs happened because of a desire to keep up with Moore's Law, IMO. Because of physical limitations the optimizations to improve CPU performance also decreased security. Industries reached a tipping point where security couldn't be ignored any longer for increased clock speeds. Spectre/Meltdown were a wake up call. Intel isn't necessarily a bad actor. They were tasked with creating a general purpose ISA to meet the needs of all customers while improving the performance of their chip-sets in all domains using their hardware.
Since a general purpose hardware and ISA isn't going to see much improvement from here on out, there needs to be a shift in the way we make improvements. Intel tried keeping up with yearly CPU speed increases but opened security holes. Now we as an industry are taking a step back to re-evaluate priorities and find new ways to progress. Here is my "Ask HN": Since security is in the spotlight now (rightfully so), will we start from a more secure open source instruction set (RISC-V perhaps)? If we've reached a wall in terms of general purpose computing is the following chain going to be the next wave/market for computing once cloud/big-data slow down? Identify a domain -> identify a type of problem in the domain -> build a language to optimize for said problem type -> build an instruction set to optimize for said language -> and finally, build hardware to optimize said instruction set |
No comments yet