The Unspoken Pact Between Software and Hardware Engineers
Description
A two-part meme that contrasts the attitudes of software and hardware engineers. The top section contains text that reads, "Software engineers: it's just too expensive to optimize our code, also computers will be faster anyway in a few years". Below this, it says, "Hardware engineers:". The bottom section features two identical images of the Monkey Puppet meme, where the puppet looks away with a nervous, sidelong glance. This meme humorously depicts the pressure software developers place on hardware engineers. It suggests that software teams often defer performance optimization, relying on the relentless advance of hardware (Moore's Law) to solve their problems, which causes anxiety for the hardware engineers who have to deliver these advancements
Comments
7Comment deleted
Software engineers see Moore's Law as a promise of a free lunch, while hardware engineers know they're the ones stuck in the kitchen working double shifts to cook it
Hardware engineer’s quiet scream: “I’m dark-siliconing half the die to stay under 200 W and you’re shipping a 400 MB Electron ‘settings’ page - maybe run a profiler before you sacrifice another transistor in my honor?”
Hardware engineers watching software eat 90% of the performance gains they spent five years achieving with a new architecture, knowing the next JavaScript framework will somehow make it even worse
Hardware engineers watching software devs justify O(n³) algorithms with 'but CPUs double in speed every 18 months' is like watching someone set money on fire while explaining that inflation will make it worthless anyway. Meanwhile, they're over here squeezing every clock cycle because physics doesn't care about your sprint velocity, and transistors aren't getting meaningfully faster - just more numerous and power-hungry
Software roadmap: wait for GHz; hardware reality: DVFS, dark silicon, and the memory wall - maybe try vectorizing that O(n^2) hot path instead
“We’ll optimize later; hardware will get faster” - hardware engineer side-eye: Dennard scaling is dead, the memory wall is alive, and your O(n^2) won’t vectorize
Hardware fabs billions into 2nm nodes so SWEs can confidently ship that unchecked N+1 query