Glossary
The terms that appear across this site, defined without marketing language.
Accelerator
A processor specialised for the mathematics behind AI. In practice, a GPU or a comparable device.
Inference
Running a trained model to produce an answer. Most business AI spending goes here.
Training
Creating a model from data. Expensive, and rarely necessary for most organisations.
Fine-tuning
Adapting an existing model to your own data or tone.
Quantisation
Storing a model at lower precision so it needs less memory, with some accuracy trade-off.
HBM
High-bandwidth memory used on enterprise accelerators.
SXM and OAM
Server accelerator module formats. They require a compatible baseboard and cannot be fitted to a desktop.
Interconnect
The high-speed link between accelerators or servers.
Tokens per second
A practical throughput measure for language models.
Utilisation
How much of the time your hardware is actually working. The number that decides whether owning beats renting.