News
EveryDatacenter

GPU

A processor with thousands of small cores, first made for graphics, now the main engine for training and running AI models.

Also: graphics processing unit, accelerator, AI accelerator

A graphics processing unit does many simple calculations at once, which suits the matrix arithmetic at the heart of neural networks. Nvidia's data centre GPUs, such as the H100 and the Blackwell family, dominate AI computing, with AMD, Google's TPUs and others as alternatives. Each chip can draw from about 700 W to 1,000 W or more, and they are packed tightly together, which is why AI racks need much more power and often liquid cooling.

See also

Find data centres that mention GPU

The weekly digest

Every Monday: the data centre of the day, new pages on EveryDatacenter, and the conferences and anniversaries of the week ahead.