Commit d190aa
2026-04-12 14:01:32 cloudy-jc: Add GPU vs CPU wiki page with YouTube embed| /dev/null .. GPU vs CPU.md | |
| @@ 0,0 1,36 @@ | |
| + | # GPU vs CPU |
| + | |
| + | *A companion page to [[Workshop 1 Hardware and Networks]].* |
| + | |
| + | When you cracked open a laptop in Workshop 1, you spotted the CPU — the brain of the computer. It fetches instructions, does the maths, and keeps everything running. A modern CPU has a handful of very fast, very clever cores (usually 4 to 16). It can juggle lots of different jobs, but it tackles them roughly one at a time per core. |
| + | |
| + | The GPU sits on a separate chip, often right next to the CPU. It was originally built to push pixels onto your screen — millions of tiny colour calculations every frame. To do that, it doesn't need a few brilliant cores; it needs **thousands of simple ones** all working at once. Each individual core is slower than a CPU core, but when thousands fire together the total throughput is enormous. |
| + | |
| + | Here's an analogy: imagine you need to deliver 1 000 parcels across town. A CPU is like one very fast truck — it zooms between addresses, but it still visits them one by one. A GPU is like 1 000 cyclists who each grab a parcel and pedal off at the same time. No single cyclist is as fast as the truck, but the whole fleet finishes the job far sooner. |
| + | |
| + | This is exactly why AI researchers love GPUs. Training an AI model means doing millions of simple maths operations (multiply, add, multiply, add — over and over). A CPU would grind through them in a long queue. A GPU spreads them across thousands of cores and finishes in a fraction of the time. |
| + | |
| + | You've already seen the CPU inside a laptop. The GPU sits right next to it — and after watching the video below, you'll never forget what makes them different. |
| + | |
| + | --- |
| + | |
| + | ## The Video |
| + | |
| + | {{Video|src=https://www.youtube.com/watch?v=8_ZTvG1WQxM}} |
| + | |
| + | In this clip, the *Mythbusters* team up with NVIDIA to demonstrate the difference using paintball guns. One gun fires one ball at a time (the CPU). A massive wall of guns fires thousands at once (the GPU). The result: the Mona Lisa, painted in about 80 milliseconds. |
| + | |
| + | --- |
| + | |
| + | ## What to Watch For |
| + | |
| + | - **The single-gun pass vs the wall of guns** — notice how the single gun is accurate but painfully slow, while the wall completes the whole image almost instantly. That's the core trade-off: few fast cores vs many parallel ones. |
| + | - **How the final image comes together** — the GPU rig doesn't paint one row at a time. Every part of the picture appears at once. That's parallelism in action. |
| + | - **The size of the rig** — look at how many barrels are in the GPU wall. A real GPU has even more cores than that (modern ones have over 10 000). |
| + | - **Why accuracy per core matters less** — each paintball lands roughly where it should, not perfectly. GPUs work the same way: individually simple, collectively powerful. |
| + | |
| + | --- |
| + | |
| + | See also: [[Workshop 1 Hardware and Networks]] |
| + | |
| + | *Last updated: April 2026.* |