What Does '3nm' or '8nm' Actually Mean? Chip Process Nodes, Explained

Phone, console and GPU chips are sold on numbers like 8nm, 4nm and 3nm, but those numbers stopped being real measurements decades ago. Here's what a process node is, which one is inside your PS5, Switch 2 or graphics card, and why Samsung's new 8LPU+ has people talking about a cheaper Switch 2.

Zerocade · Updated · 4 min read
ShareWhatsAppX
Zerocade graphic reading What is a process node?
Image: Zerocade

The short version

  • A process node (like 8nm or 3nm) is the name of a chip factory's manufacturing recipe.
  • The number isn't a real size any more. It's a brand name that roughly means "newer and denser", and each company names its own.
  • A newer node usually lets a chip do the same work using less power, with more transistors in the same space. It also costs more per chip.
  • Samsung's upgraded 8LPU+ recipe reportedly cuts power by 15%. That's why rumours have tied it to a cheaper Switch 2, though Nintendo hasn't announced anything.

Where the number came from

Decades ago, a node's number really was a measurement, roughly the length of a transistor's gate, the switch at the heart of every chip. As nodes shrank from micrometres to 500nm, 250nm and onwards, the name and the physical size moved together.

That stopped being true in the mid-1990s. Transistor designs changed shape, most dramatically when they went from flat to 3D "fin" designs (FinFET) in the 2010s. Today the number is a marketing label, and it doesn't match any single part of the chip. No feature on a "3nm" chip is 3 nanometres long.

So companies' numbers don't line up with each other. Intel's old "10nm" process, for example, was about as dense as TSMC's "7nm", which is partly why Intel renamed its later nodes "Intel 7", "Intel 4" and so on. Comparing two chips by nanometres alone can mislead you.

Which node is inside your hardware

Zerocade table of which process node is used in popular gaming chips Image: Zerocade

Your hardware is tied to these recipes. Nvidia's RTX 30 series GPUs were made on Samsung's 8nm-class process. The Switch 2's Nvidia T239 chip uses Samsung's 8N too: Nintendo never said so, but a die shot published by Chinese tech channel Geekerwan, analysed by TweakTown, matched it to that process, with an 8-core Arm CPU and a 1,536-core Ampere GPU. The launch PS5 used TSMC's 7nm, and the RTX 40 and RTX 50 series use TSMC's custom 4N.

Phones run a generation or two ahead. Apple's A17 Pro in the iPhone 15 Pro was the first phone chip on TSMC 3nm, and 2nm parts are starting to arrive in 2026.

The pattern is useful: consoles and graphics cards often sit on an older, cheaper node than the newest phones. A console has to be made in tens of millions at a fixed price for years, so a mature node with good yields often wins over the newest one.

What actually improves

A better gauge is density: how many transistors fit in a square millimetre. TSMC's 5nm-class N5 packs around 171 million transistors per mm².

More transistors in less space usually means three things:

  • Less power for the same work, so better battery life and less heat. This matters most in handhelds and phones.
  • More performance in the same power budget, because the chip can run faster or fit more cores before it gets too hot.
  • Smaller chips, so more of them per silicon wafer, which can lower cost per chip if the wafer price doesn't rise too much.

Our guide on thermal throttling explains why the first point decides how a handheld performs after ten minutes of play.

Samsung 8LPU+ and the Switch 2 Lite rumour

On October 23, Korea's Electronic Times (ETNews) reported that Samsung is preparing an upgraded 8nm recipe called 8LPU+, the second generation of the 8LPU process it introduced in 2018.

Zerocade cards showing Samsung 8LPU+ reported performance, power and area claims Image: Zerocade

Compared with 8LPU, Samsung reportedly claims about 12% more performance, 15% lower power and 8% less logic area, with an unveiling planned before the end of the year. The report pitches it at cars, robotics and edge AI chips. 8nm is Samsung's last major node that doesn't need EUV, the expensive extreme-ultraviolet lithography machines its newer nodes use, which keeps it cheap.

A patent-watching account on Bluesky then linked it to a possible cheaper Switch 2 Lite, and Notebookcheck and Android Headlines picked that up. That link is speculation. The original report doesn't mention Nintendo, and Nintendo hasn't announced a Lite or OLED model. It's plausible on paper, because moving a chip to an improved version of the same node is far simpler than a full redesign. Nintendo did exactly that with the original Switch, whose 2019 revision used a more efficient Tegra X1 chip and gained hours of battery life. But plausible isn't the same as reported.

With the Switch 2 listed at $499.99 (about ₹47,500) in the US, a cheaper model would be welcome. Nintendo still doesn't sell the Switch 2 officially in India, so Indian buyers rely on imports either way.

The catch

A new node isn't automatically better for you. Cutting-edge recipes cost far more per wafer, which is one reason chip prices keep rising (see our post on Intel, Qualcomm and AMD price increases). A mature, cheaper node can make more sense for a budget console or phone.

Next time a spec sheet says "4nm", treat it as a family name, not a measurement. Two chips on the same node can still behave very differently. For one reason why, see our guide on chip binning.

Found this useful?
Send it to a friend who's been asking.
ShareWhatsAppX
Whodunit
Who did it? Place the suspects and catch the murderer.
Solve today's case →

Keep reading