News
EveryChip

AI accelerators

H100, B200, MI300X, the TPU, Trainium, Gaudi, Cerebras WSE and the NPUs of phones and laptops. 96 chips on EveryChip so far.

An AI accelerator is a chip designed to train or run neural networks, the layered models behind image recognition, speech and large language models. Most of that work is multiplying large tables of numbers, called matrices. Accelerators pack in thousands of multiply units, feed them from fast memory and use low-precision number formats that save power. Nvidia's data centre GPUs, Google's TPU and many start-up chips belong here.

What makes them fast

A neural network layer multiplies a matrix of inputs by a matrix of learned weights. Accelerators have dedicated matrix units, such as Nvidia's tensor cores or the systolic array of Google's TPU, in which numbers flow through a grid of multipliers. They use formats with fewer bits, such as 16-bit, 8-bit and even 4-bit numbers, so each operation costs less energy and more fit on the chip.

Memory and networks matter as much

A large model has billions of weights that must be read again and again. That is why AI accelerators sit beside stacks of High Bandwidth Memory, and why training clusters link thousands of chips with fast networks. Nvidia's H100 of 2022 has 80 billion transistors; its Blackwell B200 joins two dies with 208 billion. Demand for these chips since 2023 has strained supplies of HBM and advanced packaging.

Famous AI accelerator chips

Further reading

Hanguang 800 🇨🇳 ChinaHanguang 800 AI inference chip by Alibaba for cloud and edge computingAI accelerator · Other · 2019 Meta MTIA v2 🇺🇸 United StatesMeta MTIA v2 Custom AI inference accelerator for Meta ranking modelsAI accelerator · Other · 2024 Biren BR100 🇨🇳 ChinaBiren BR100 AI GPU for data centre compute workloadsGPU · GPU architecture RNGD 🇰🇷 South KoreaRNGD 150W AI inference accelerator chip by FuriosaAIAI accelerator · Other · 2024 Qualcomm Cloud AI 100 🇺🇸 United StatesQualcomm Cloud AI 100 AI inference accelerator for data centres and edge devicesAI accelerator · VLIW · 2021 Telum II 🇺🇸 United StatesTelum II Mainframe processor for IBM z17 with AI accelerationCPU · Other Trillium 🇺🇸 United StatesTrillium Google’s sixth-generation AI accelerator chip for training and inference.AI accelerator · Tensor design · 2024 Ironwood 🇺🇸 United StatesIronwood Google's seventh-generation AI accelerator for inference workloadsAI accelerator · Other · 2025 TPU v5p 🇺🇸 United StatesTPU v5p Google AI accelerator chip for large-scale model trainingAI accelerator · Tensor design TPU v5e 🇺🇸 United StatesTPU v5e Google AI accelerator for training and inference workloadsAI accelerator · Tensor design GB10 Grace Blackwell S 🇺🇸 United StatesGB10 Grace Blackwell Superchip AI superchip for desktop workstationsSoC · Arm A100 🇺🇸 United StatesA100 Data centre GPU for AI and high-performance computing.GPU · GPU architecture · 2020 Nvidia B300 🇺🇸 United StatesNvidia B300 AI accelerator with dual-die design and 288 GB memoryGPU · GPU architecture SambaNova SN40L 🇺🇸 United StatesSambaNova SN40L AI inference chip with three-tier memory architectureAI accelerator · Other · 2023 AWS Inferentia 🇺🇸 United StatesAWS Inferentia AI inference chip designed by Amazon Web ServicesAI accelerator · Other AMD Instinct MI325X 🇺🇸 United StatesAMD Instinct MI325X AI accelerator with 256GB HBM3E memoryAI accelerator · GPU architecture AMD Instinct MI355X 🇺🇸 United StatesAMD Instinct MI355X AI accelerator using CDNA 4 architecture on a 3nm process.AI accelerator · GPU architecture · 2025 TPU v4 🇺🇸 United StatesTPU v4 Google AI accelerator chip with 3D mesh networkingAI accelerator · Other Ascend 910C 🇨🇳 ChinaAscend 910C AI accelerator chip by Huawei for data centresAI accelerator · Other ET-SoC-1 🇺🇸 United StatesET-SoC-1 RISC-V system on a chip for AI inferenceAI accelerator · RISC-V Trainium2 🇺🇸 United StatesTrainium2 AWS second-generation AI training accelerator chipAI accelerator · Other · 2024 🇺🇸 United StatesHopper Nvidia datacentre GPU microarchitecture for AI and high-performance computing.GPU · GPU architecture · 2022 WorldElectrochemical RAM Non-volatile memory for analog deep learning accelerationMemory · Other 🇺🇸 United StatesBlackwell Nvidia GPU microarchitecture for AI and gamingGPU · GPU architecture · 2024

Photo credits on this page

The small pictures above come from Wikimedia Commons under free licences. Some are cropped to fit.

The weekly digest

Every Monday: the chip of the day, new pages on EveryChip, and the conferences and anniversaries of the week ahead.