CUDA becomes generally available

NVIDIA opened a software platform for general-purpose computation on graphics processors to developers.

Why it matters

Access to parallel computation on ordinary graphics cards made training large models practical outside specialised computing centres.

The year is marked approximate: the beta and the first stable release both happened during 2007 (NVIDIA released the public beta of the SDK on 26 February 2007 and dates the introduction of CUDA itself to November 2006), and the archived page snapshot cited here is from 2008.

Event record

Event date

2007 · Approximate date

Timeline date

Event date

Verification

Sources gathered automatically · September 16, 2026

Lines

ID

evt-0049

NVIDIA

Sources

primary NVIDIA CUDA developer pages
NVIDIA (archived) · Published 2008
secondary Nvidia Releases Public Beta of Cuda Software Developer Kit (26 February 2007)
Computer Graphics World (NVIDIA release reprinted) · Published February 26, 2007
secondary NVIDIA Takes Direct Aim at High Performance Computing (HPCwire, 22 June 2007)
HPCwire · Published June 22, 2007

Related events

Earlier

Related Amazon EC2 2006links to this record
The other half of what made training large models affordable: renting computation instead of buying it.
Related GeForce 256 1999links to this record
The class of device CUDA opened to general-purpose computation eight years later.

Later

Builds on Deep networks on graphics processors 2009links to this record
The computation runs on graphics processors through CUDA.
- Large-scale deep unsupervised learning using graphics processors · ICML 2009
Builds on AlexNet at ImageNet 2012links to this record
The AlexNet paper describes training on GPUs using this platform.
- ImageNet Classification · NeurIPS
Related Google's Tensor Processing Unit 2016links to this record
Instead of a repurposed graphics chip, a chip designed for neural networks.
Related Jalapeño chip 2026links to this record
Both records are about the computing platform models are trained on.