Graphics processing unit (GPU)
Massively parallel graphics processor that became general-purpose; now the main engine of AI training and inference.
- 1999

First version
NVIDIA (GeForce 256)
Image: Hyins · Public domain · Wikimedia Commons - 2001

GeForce 3
Programmable vertex and pixel shaders replace fixed-function graphics, letting developers write effects as code.
Image: Hyins · Public domain · Wikimedia Commons - 2006

GeForce 8800 (G80) and CUDA
128 unified stream processors and CUDA open GPUs to general computing.
Image: Hyins · Public domain · Wikimedia Commons - 2017

Volta V100
Tensor Cores added for neural-network matrix multiply-accumulate.
Image: Tom's Hardware (Hot Chips 2017 photo, Aug 2017) · © Future plc / Tom's Hardware; no open … · Tom's Hardware - 2018

Turing (RTX 20 series)
Dedicated RT cores add real-time hardware ray tracing to consumer GPUs.
Image: Fritzchens Fritz · CC0 · Wikimedia Commons - 2024

Blackwell
208 billion transistors; two dies act as one GPU; adds 4-bit floating-point inference.
Image: 极客湾Geekerwan · CC BY 3.0 · Wikimedia Commons
Contested: NVIDIA's 'first GPU' label is marketing: 'GPU' was used in the 1980s, and 3Dlabs' Glint Gamma (1997) already did transform and lighting.