What Is a QPU? CPU vs GPU vs QPU, Explained Simply

QPU stands for quantum processing unit. It is the part of a quantum computer that holds the qubits. A qubit is the quantum version of a bit. A normal bit is always 0 or 1. A qubit carries two numbers that set how likely you are to read 0 or 1 when you look at it. You can learn the details in qubit vs bit.

The name copies CPU (central processing unit) and GPU (graphics processing unit). That is on purpose. All three are chips that do work. But they do very different kinds of work, and they are good at very different jobs.

Why should you care? Because most confusion about quantum computing starts here. People picture a QPU as a super CPU that will make every app faster. It will not. Once you see what a QPU really is, the news gets much easier to judge.

Start with the two chips you already own.

A CPU is the boss of your computer. It has a few strong cores. Each core can do almost any task: open a file, run a game loop, check your email. Think of a head chef. One chef can cook any dish, one step after another.

A GPU has thousands of small, simple cores. They all do the same kind of math at once, on lots of numbers. Think of a big team of kitchen helpers. Each helper can only chop. But a thousand helpers chop a mountain of onions very fast. That is why GPUs are great for graphics and for training AI models.

Here is the key point. A GPU does not replace the CPU. The CPU hands the GPU a big chunk of math, waits, and takes the result back. The GPU is a helper chip, often called an accelerator.

A QPU is also a helper chip. But it does not do faster math on ordinary numbers. It does a different kind of math.

A QPU runs a quantum circuit. That is a short list of steps called gates. Each gate nudges the qubits. Some gates make qubits depend on each other, which is called entanglement. At the end, you measure the qubits. Measuring turns each qubit into a plain 0 or 1.

Back to the kitchen. The QPU is not a faster chef or more helpers. It is more like a strange oven that only bakes certain recipes. For those recipes, it can do things no normal oven can. For everything else, it is useless. And it cannot even read the recipe by itself. Someone else has to load it.

Unlike a real oven, a QPU does not give the same result each time. The answer comes out as chances. So you run the same circuit many times. Each run is called a shot. Then you count what came out.

The last row matters. You do not buy a QPU card for your PC. You send a circuit over the internet to a QPU that sits in a special room. Our catalog of quantum processors lists which ones you can reach and how.

The circuit above puts one qubit in an even state. The H gate (the Hadamard gate) sets its chance of reading 0 to one half, and its chance of reading 1 to one half. You can see why in superposition.

Do the arithmetic. Half of 1,024 is 512. So about 512 shots should read 0 and about 512 should read 1. One shot alone tells you almost nothing. It is like one coin toss. Only the count across many shots shows you the answer.

This is normal QPU life. A real program is a circuit plus a number of shots. The result is a table of counts. A normal computer then reads that table and decides what it means.

The chip is only the middle of a long chain. A normal computer does the rest.

1. Your code builds a circuit.
2. A compiler rewrites it into the few gates the chip can really do. Each chip has its own gate set and its own wiring. See native gates and transpilation.
3. Control electronics turn each gate into timed pulses, such as microwave or laser pulses.
4. The qubits react to those pulses.
5. Readout turns each qubit back into a plain 0 or 1.
6. A normal computer collects the counts and sends them back to you.

So a QPU is never a stand-alone computer. It is one part of a system that is mostly classical. Our course chapter What Is a QPU? walks one job through every step.

QPUs are built in very different ways. Some use tiny loops of metal cooled close to absolute zero. Some use single charged atoms held in place by electric fields. Some use atoms held by lasers. Some use light. We explain each one in types of quantum computers.

Here are a few real records from our data. Every figure below is reported by the vendor.

• Google Willow: 105 qubits, announced 2024-12-09. It is a research chip with no public cloud access.
• IBM ibm_boston (a Heron r3 chip): 156 qubits, open to paid IBM plans since January 2026.
• Quantinuum Helios: 98 qubits made from trapped barium ions.
• Quandela Ascella: 6 qubits made of light (photons), on a public cloud.

Some QPUs need deep cold. For example, Google's paper on its Sycamore chip says it ran below 20 thousandths of a degree above absolute zero (20 mK). D-Wave lists one of its Advantage machines at 15.4 mK. Others do not. Xanadu's paper on its Aurora system says its light-based parts run at room temperature.

Not by itself. Qubits make mistakes. Every gate has a small chance of going wrong. A chip with many noisy qubits can do less useful work than a chip with fewer, cleaner qubits.

Here is a small worked example. It uses round numbers, not any real chip. Say each gate fails 1 time in 100. The chance that one gate works is 0.99. The chance that 100 gates in a row all work is 0.99 multiplied by itself 100 times. That is about 0.37. So only about 37 of every 100 runs would be fully clean.

That is why a good spec sheet shows error rates next to the qubit count. Our guide on reading hardware specs shows what to look for. The qubit count and two-qubit fidelity pages explain each number. We never rank QPUs with one score, because no single number is fair across such different machines.

Yes, and it is easier than most people think. Several QPUs are open to the public over the cloud. For example, Amazon Braket listed 8 QPUs from 5 hardware makers when we checked in 2026. Some need a paid plan. Some have a free tier with limited time.

But you do not need a real QPU to start. Small circuits can be simulated on a normal computer. Our Lab does this in your browser. A simulator gives you clean, perfect results. That makes it the best place to learn, before you meet the noise of real hardware.

A fair warning: simulation gets hard fast. Each extra qubit doubles the memory a simulator needs. That limit is one real reason QPUs exist at all.

• A CPU does any task, one step at a time.
• A GPU does the same simple math on many numbers at once.
• A QPU runs quantum circuits on qubits. It helps only with some problems, and it always needs a normal computer beside it.
• QPU results are chances, so you run many shots and count.
• Qubit count alone does not tell you how good a QPU is.

Next, try building a circuit in the Lab, or browse every QPU we track with its sources.