Chips: CPU, GPU, TPU, NPU... The guide to not lose your mind

Once upon a time, there was just the CPU, the computer's "brain" that did everything, from accounting to solitaire. Then 3D graphics arrived, and the poor CPU just couldn't keep up. Imagine an orchestra conductor who also has to play all the instruments: impossible.

That's where the GPU, the Graphics Processing Unit, comes in. Born to render virtual worlds at insane speeds, it turned out to be excellent at doing many simple calculations in parallel, the very foundation of artificial intelligence. NVIDIA, founded by Jensen Huang in 1993, became a giant thanks to its GPUs.

Artificial intelligence, as we know, has an insatiable hunger for computation. That's why specialized chips are needed, optimized for its particular demands. Otherwise, AI just drags along like a sloth.

Google, for example, quickly caught on and developed its Tensor Processing Units (TPU) in 2016 to accelerate machine learning workloads. These units are perfect for matrix operations, the daily bread of neural networks. Then there are NPUs, Neural Processing Units, often found directly in our smartphones. The Apple A17 Pro, released in 2023, includes an NPU to handle Face ID and other smart functions directly on the device.

But it's not over yet, because the tech world never stops. Some are thinking about networking, and others are dreaming of calculations that today are pure science fiction. Does it sound like too much? Maybe it is.

Data Processing Units (DPUs) are specialized processors for managing network and data in datacenters, freeing up CPUs from tedious, repetitive tasks. NVIDIA, for instance, acquired Mellanox in 2020, a leading DPU company, to strengthen its cloud computing position. And then there are QPUs, Quantum Processing Units, still mad scientist stuff, promising to solve problems currently impossible by leveraging quantum physics. IBM announced its first commercial quantum computer, the IBM Q System One, in 2019.