TOPS: how AI power in chips is measured

TOPS (trillions of operations per second) is the number of trillions of simple calculations per second that a neural processor or other AI accelerator can perform. Manufacturers use this number to show off the AI power of smartphones and laptops.

TOPS is like the top speed on a speedometer

In plain words

TOPS is like the top speed on a car's speedometer. 250 km/h looks impressive, but you rarely drive that fast, and a lot depends on the road, tyres and driver. TOPS is the same: a peak theoretical figure, while real AI speed depends on the software, memory and exactly how it was measured.

How it works

A neural processor multiplies and adds numbers - the basis of how neural networks work.
The manufacturer calculates how many such operations the chip can do per second at its peak.
One trillion operations per second = 1 TOPS.
The figure is often given for short numbers (INT8 or INT4), which makes it bigger.
The real speed of AI features also depends on memory and app optimisation.

AI power in laptops and smartphones

Where you'll see it

TOPS appear in the specs of chips for smartphones, tablets and laptops (for "AI PCs" a threshold of 40 TOPS for the NPU is often mentioned). They also appear in descriptions of accelerators for robots, cameras and cars.

Does it matter when buying

For a laptop with Windows AI features, yes - there are minimum NPU requirements. For a smartphone, less so, because what matters more is which AI features you actually get. Compare TOPS only between chips from the same manufacturer at the same precision. Between brands, I would rely on tests of specific features.

Myths

Myth: More TOPS always means faster AI features. Reality: Not always: a lot depends on memory, software and optimisation.
Myth: All manufacturers count TOPS the same way. Reality: No, the number precision and method can differ.

FAQ

How many TOPS do modern phones have?

Flagship NPUs reach dozens of TOPS, but the figures are often calculated differently.

What do INT8 and INT4 next to TOPS mean?

The bit width of the numbers. With shorter numbers, the chip performs more operations.

Are TOPS and TFLOPS the same?

No. TFLOPS usually refers to floating-point numbers (GPUs), TOPS to integers (NPUs).