CPUs and GPUs: a brief history#
Chips are the brain of the computer. They, and the billions of little transistors that we’ve somehow figured out to put on them, are responsible for executing your code fast and reliably.
The CPU#
The most popular type of chip is the CPU, or central processing unit. It’s the workhorse of most computers, and is probably working to display this post that you’re reading as we speak. And they’ve been getting faster and faster and faster, thanks to an age old observation called Moore’s Law that processing power seems to double every 2 years. The more transistors we fit on the chip, the faster it is.
To give you some more context to how far we’ve come: the first CPU ever developed was the Intel 4004, released in 1971. It has 2,300 transistors, and could perform 60,000 operations per second. The Apple M4 chip – today’s state of the art – has…wait for it… 28 BILLION transistors. Someone smarter than me can check my math, but I believe this is roughly a 12 million times improvement.
But it’s not just transistors – we’ve also figured out how to get more chips to fit into a computer. Laptops shipping today have 10 cores, which you can roughly translate to 10 chips. We’ve also gotten better at making these chips work together to get tasks done.
When you combine these two dynamics – more transistors and more chips working better together – you get much faster hardware. Readers graced by the sweetness of age will recall how incredibly slow computers of even 10 years ago were compared to the snappy experiences we have with modern laptops and servers.
The GPU#
Powerful as the CPU is, it’s not perfect – like me, it has things it’s good at (writing) and things it’s not so good at (walking past a Shake Shack). CPUs are designed to do complex, intertwined computations in order. Think of it like a really long, convoluted checklist of things that have to happen in a specific sequence: that’s what most code looks like, and that’s what CPUs excel at. And for most of the tasks a computer needs to do, like swimming aimlessly through a sea of Chrome tabs, that’s exactly what you want.
But sometimes you need your computer to do the opposite: tons and tons and tons of very simple operations, but all at the same time. This is where GPUs come in.
I think a cooking analogy is appropriate here. Imagine you’re cooking some old school Pasta Genovese for your wife (or cellmates) and following a complex recipe. The recipe is sequential, and each step depends on the previous one: having an extra 5 people to help won’t really speed things up that much.
But what about individual tasks within a recipe? Imagine you’re cooking for your whole family (or cell block). You need to pass about 100 onions through a mandoline – and that would go a lot faster if you had 5 people with 5 mandolines instead of 1 with 1. The recipe sequence is still something you can’t really hand off, but some of the individual parts benefit with scale.
This onion-mandoline effect, as I’ve coined it, is called parallel computing. Instead of having a task done all by one worker, breaking it down into different parts and having multiple workers work on it at once. This is what GPUs excel at: doing pretty simple things, but at massive scale and speed. And GPUs have experienced the same kind of scaling curve as CPUs: the first NVIDIA GeForce in 1999 had 17 million transistors, while today’s state of the art Blackwell has 208 billion transistors.
The first application for GPUs was mostly gaming and simulators. They’re sitting in every major game console (PS5, anyone?) and any serious gamer building their own rig is mostly worried about getting the best GPU possible. Why? Because rendering 3D game worlds and video is exactly what GPUs are good at: doing simple operations (displaying pixels), but at hyper speed.
But around 2010, a new class of GPU enjoyers started to show up…
