Glossary
Cloud and AI terms, in plain words
Short answers first. The sums and sources are on each page.
See GPU hostingAI and GPUs
Models, memory and the chips that run them.
- LLMA model trained on text to predict the next token.
- GPTGenerative Pre-trained Transformer, a family of LLMs.
- GPUA chip that runs many calculations at the same time.
- VRAMGPU memory. The model and its working memory must fit in it.
- HBMStacked memory beside the GPU, with far more bandwidth than GDDR.
- CUDANVIDIA’s platform for running general code on its GPUs.
- QuantizationFewer bits per weight: smaller model, less memory.
- RAGLook up your documents, then let the LLM answer.
- InferenceUsing a trained model to answer new inputs.
Cloud and servers
What a server is, and what a provider promises.
Security
Testing a site the way an attacker would.
Tell us what you run.
Describe the workload. An engineer replies within one business day.