# GPU training and mixed precision

Source: https://onco.cc/terms/mixed-precision-gpu/  
OnCo record `mixed-precision-gpu` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

Foundation models are trained on graphics processors, and mixed precision stores most numbers in 16-bit floats to halve memory and double speed; a laptop-class chip can fine-tune small models but pretraining at scale waits for a data-centre GPU.

## Summary

Mixed-precision arithmetic uses floating-point numbers of different widths in a single computation (Wikipedia); deep learning stacks many layers of artificial neurons and trains them on large data, which drives its hardware demand (Wikipedia). NVIDIA A100 and H100 cards through CUDA are the standard training hardware, Apple's Metal backend (MPS) runs smaller jobs on laptops, and a fine-tuning harness written on one and awaiting the other is a normal state for an academic project.

## Fields

- Kind: Term
- Last checked: 2026-09-24
- Also known as: mixed precision; mixed-precision training; bfloat16; fp16 training; GPU training; A100; H100; CUDA; MPS backend; Apple silicon MPS
- Tags: cansim-terms

## Notes

- Listed in the CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme; CanSim page path /terms/mixed-precision-gpu.

## Sources

- Wikipedia: https://en.wikipedia.org/wiki/Mixed-precision_arithmetic
- Wikipedia: deep learning: https://en.wikipedia.org/wiki/Deep_learning
- Wikipedia: https://en.wikipedia.org/wiki/Mixed-precision_arithmetic

## Connected records

- terms: [Cancer AI vocabulary (CanSim terms map)](https://onco.cc/terms/cancer-ai-vocabulary/), [Fine-tuning versus frozen features, and LoRA](https://onco.cc/terms/fine-tuning-vs-frozen/), [Foundation model](https://onco.cc/terms/foundation-model/)
- technologies: [AI compute and model platforms for oncology](https://onco.cc/technologies/ai-compute-platforms/)

---
JSON: https://onco.cc/api/v1/entities/mixed-precision-gpu.json