BLOCK_SIZE means in Triton: the tile size each program instance processes

Why does your computer sometimes feel slow when you're playing games?

Imagine you're playing a video game on your computer. Suddenly, the game starts lagging, making it hard to move your character smoothly.

Your computer's processor is like a chef cooking a meal. If the chef has too many orders (tasks) at once, they can't cook them all efficiently. The processor divides the game's data into smaller pieces called "blocks" to handle them more quickly. This process is called "block processing."

Example

If your game data is 1024 units and your processor's block size is 256 units, it divides the data into 4 blocks (1024 ÷ 256 = 4).

Remember this

The block size determines how many pieces of game data your computer processes at once, affecting the game's smoothness.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

Triton auto-tunes BLOCK_SIZE: different sizes optimize for different hardware

Ever wondered why videos look different on various devices?

Triton differs from CUDA

Why does a super-fast computer sometimes run slower than a regular one?

tl.dot does in Triton: block-level matrix multiply using tensor cores

tl.dot performs block-level matrix multiplication using tensor cores in Triton

Matrix multiplication algorithm

Ever wondered how computers speed up multiplying huge numbers?

tl.load and tl.store do in Triton: read/write tensors from/to GPU global memory

`tl.load` reads tensors from GPU memory; `tl.store` writes tensors to GPU memory

loop unrolling does: trades code size for reduced loop overhead

Can we speed up our computer programs?

Swipe through 100 ML concepts daily

Open Pocket Polymath