Multiplying grids of numbers

How it works, step by step

1. Pick a row and column

   Take the top row [1, 2] from the left grid. Match it with the first column [5, 7] from the right grid.

   Two pairs to multiply: (1, 5) and (2, 7).

2. Multiply, then add

   Multiply to get 5 and 14. Add them: 1 × 5 + 2 × 7 = 19.

   The top-left answer is 19.

3. Fill the other spaces

   Use the same multiply-and-add rule for each space in the answer grid.

   The complete result is [[19, 22], [43, 50]].

In plain language

A matrix is a grid of numbers, and AI models do a lot of matrix multiplication. Tensor Cores are GPU parts built to speed up supported versions of this multiply-and-add work.

A way to picture it

Imagine a calculator made for one repeated spreadsheet task. It can help with that task, but the rest of the spreadsheet still needs other work.

A worked example

[[1, 2], [3, 4]] × [[5, 6], [7, 8]] = [[19, 22], [43, 50]]. For the top-left answer, multiply matching row and column numbers, then add: 1 × 5 + 2 × 7 = 19.

Keep in mind

The 2 × 2 grid teaches the math, not the exact size a Tensor Core handles at once. Not every AI calculation uses Tensor Cores. FP4 and FP32 store numbers differently, so their advertised calculation rates are not interchangeable or promised chat speeds.

What these words mean

Matrix

A rectangular grid of numbers.

Tensor Core

A GPU part designed for supported matrix calculations. It does not run every calculation in an AI model.

Accumulate

Add a result to a running total.

Tile

A small section of a larger grid. Real Tensor Core instructions use supported shapes, not necessarily our 2 × 2 example.

What is 1 × 5 + 2 × 7?

19. Multiply first: 1 × 5 = 5 and 2 × 7 = 14. Then add 5 + 14 = 19. This is one space in the answer grid.

Where this comes from

This explanation is checked against primary documentation. The small arithmetic examples are ours and are not hardware measurements or vendor benchmarks.

Read the primary source