Infrastructure
Floating-Point Operations Per Second
FLOPSA measure of raw computing speed used to size AI hardware and training runs.
FLOPS, floating-point operations per second, quantifies how much calculation hardware can do, and total operations is also used to describe how much computation a training run consumed. Training a frontier model can take on the order of ten to the twenty-fifth operations or more, which is why it needs massive GPU clusters. The figure is a common currency for comparing chips and for policy proposals that would regulate models above a compute threshold.
Related terms
Compute
The raw processing power used to train or run a model, a key driver of cost and capability.
Graphics Processing Unit
The specialized chip that does the heavy parallel math behind training and running AI models.
Scaling Laws
The empirical finding that model performance improves predictably as you increase model size, data, and compute together.
Tensor Processing Unit
A chip designed by Google specifically for machine learning, as an alternative to general-purpose GPUs.