AI training and inference
Accelerators for training and running neural networks. 129 chips on EveryChip so far.
AI chips train neural networks and run them, a task called inference. Training a large language model takes thousands of accelerators running for weeks; inference runs every time someone uses the model. Nvidia's data centre GPUs have led since the deep-learning revival of 2012, while Google's TPU, AMD's Instinct, Amazon's Trainium and many start-up chips compete. Phones and laptops now carry small AI engines too.
Training and inference
Training adjusts billions of weights by passing data through the network and back, again and again, so it needs huge amounts of arithmetic and memory bandwidth and fast links between chips. Inference runs the finished model, often at lower precision, and is spread across data centres and devices. Many chip designs specialise in one or the other.
Why Nvidia led
In 2012 a neural network called AlexNet, trained on two Nvidia gaming GPUs, won the ImageNet image recognition contest by a wide margin. Nvidia's CUDA software, available since 2007, made its GPUs the default for researchers. Its A100 (2020), H100 (2022) and Blackwell (2024) chips became the main tools of the AI industry, and demand outran supply in 2023 and 2024.
Famous AI training and inference chips
Further reading
🇺🇸 United StatesApple A11 Bionic
Six-core phone SoC with a Neural Engine, used in iPhone 8 and iPhone XSoC · Arm · 2017
🇺🇸 United StatesLarrabee
Cancelled Intel x86-based GPGPU project that became the Xeon Phi lineGPU · x86 · 2009
🇺🇸 United StatesAmpere
Nvidia's 2020 GPU architecture behind GeForce RTX 30 and the A100 acceleratorGPU · GPU architecture · 2020
🇺🇸 United StatesIBM Heron
156-qubit tunable-coupler quantum processor by IBMQuantum · Other · 2023
🇺🇸 United StatesVolta
Nvidia's 2017 GPU architecture, first with Tensor Cores for deep learning.GPU · GPU architecture · 2017
WorldMajorana 1
Microsoft quantum processor using topological superconductorsQuantum · Other · 2025
🇺🇸 United StatesSycamore
Google's 53-qubit superconducting quantum processorQuantum · Other · 2019
🇺🇸 United StatesApple M3
Apple's 3 nm Mac and iPad SoC family with CPU, GPU and Neural EngineSoC · Arm · 2023
Photo credits on this page
The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.
- Apple M1: Henriok, CC0, source
- AMD Instinct: Мой Компьютер, CC BY 3.0, source
- Meteor Lake: Fritzchens Fritz, CC0, source
- Power10: Henriok, CC0, source
- Intel i860: Konstantin Lanzet, CC BY-SA 3.0, source
- GeForce RTX 50 series: 极客湾Geekerwan, CC BY 3.0, source
- Lunar Lake: LaptopMedia, CC BY 3.0, source
- Granite Rapids: Intel Corporation, Public domain, source
- Tesla Dojo: Steve Jurvetson, CC BY 2.0, source
- Xeon Phi: Capitalistcakes, CC0, source
- Nvidia Tesla: division, CSIRO, CC BY 3.0, source
- Nvidia RTX Spark: Daniel Lu ( User:dllu ), CC BY-SA 4.0, source
- Pascal: Fritzchens Fritz, CC0, source
- GeForce: ZMASLO, CC BY 3.0, source
- GeForce RTX 30 series: PantheraLeo1359531, CC BY 4.0, source
- Radeon RX 9000 series: MoreThanTech Personality rights Personality rights Personality rights, CC BY 3.0, source
- SpiNNaker: Pabogdan, CC BY-SA 4.0, source
- GDDR7 SDRAM: 极客湾Geekerwan, CC BY 3.0, source
- Sycamore: Google ( https://www.youtube.com/channel/UCK8sQmJBp8GCxrOtXWBpyEA ), CC BY 3.0, source
- Tensor Processing Unit: Norman P. Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan,, CC BY 4.0, source
- High Bandwidth Memory: Fritzchens Fritz, CC0, source
- Nvidia Drive: NVIDIA Taiwan, CC BY 2.0, source
- Hopper: 极客湾Geekerwan, CC BY 3.0, source
- Apple A11 Bionic: Henriok, CC0, source
- Blackwell: 极客湾Geekerwan, CC BY 3.0, source
- Larrabee: Intel, CC BY-SA 3.0, source
- GeForce RTX 20 series: 极客湾Geekerwan, CC BY 3.0, source