News
IndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value ChainIndustryTSMC’s grip on advanced chipmaking will face competitionIndustryCadence joins European Chip Design Platform to support semiconductor startups and SMEsMaking chipsMarc Hijink on ASML’s Culture, Supply Chain, and GeopoliticsProcessorsIntel's post-Nova Lake CPUs take shape as Razor Lake desktop and laptop variants surfaceAI chipsHow Google’s New AI Chip Software Helps It Battle NvidiaIndiaMore IC Packaging Plants Emerge In IndiaIndustryUS Tightens China AI Chip Access as Nvidia, TSMC and Semiconductor Stocks SlideDesignAxelera AI moves into HPC with TitaniaMaking chipsChinese envoy says chip equipment export curbs threaten EU-China trade thawIndiaSEMICON India 2026 Day 2 Sees 25 MoUs, Strategic Collaborations Across Semiconductor Value Chain
EveryChip
View of the high-performance graphics card NVIDIA H100, which is based on the Hopper architecture. The NVIDIA H100 PCIe has the following features: RAM memory type: HBM2, RAM: 80 GiB, memory speed: 2 TiB/s, 14592 Cuda co

Hopper

Nvidia datacentre GPU microarchitecture for AI and high-performance computing.

Photo: 极客湾Geekerwan, CC BY 3.0, via Wikimedia Commons

Hopper is a graphics processing unit microarchitecture developed by Nvidia for datacentres. Revealed in March 2022, it features a transformer engine and HBM3 memory to accelerate artificial intelligence workloads.

Share on WhatsApp

  • Built on TSMC N4 process with 80 billion transistors
  • Features dedicated transformer engine for AI acceleration
  • Supports HBM3 memory with 3 TB/s bandwidth
  • SXM5 module draws up to 700 watts of power
  • Named after computer scientist Grace Hopper

What it is

Hopper is a graphics processing unit microarchitecture created by Nvidia for datacentre use. It was officially revealed in March 2022 and launched later that year. The design focuses on accelerating artificial intelligence tasks, particularly those involving transformer models. It succeeds the Ampere architecture and introduces new hardware features to handle complex mathematical operations more efficiently than previous generations.

How it came about

The name Hopper honours Grace Hopper, a computer scientist and United States Navy rear admiral. Details about the architecture leaked in November 2019 via social media. Nvidia formally announced the technology during its GTC event in March 2022. By late 2022, export regulations led to modified versions like the H800 for the Chinese market. Demand surged during the AI boom, with major tech leaders seeking access to the hardware.

Inside the chip

The architecture uses up to 144 streaming multiprocessors. It includes a Tensor Memory Accelerator for efficient data movement between memory types. A dedicated transformer engine dynamically adjusts numerical precision to speed up calculations. The design supports thread block clusters for better parallel processing. It also improves single-precision floating-point throughput compared to its predecessor, Ampere.

How it was made

TSMC manufactures the chips using its N4 process node. The H100 implementation contains 80 billion transistors. The design supports High Bandwidth Memory 3, offering significantly higher speeds than prior standards. The SXM5 socket configuration provides superior performance over standard PCIe connections due to increased memory bandwidth capabilities.

Hopper
A drawing made by EveryChip from what this page says. It is a diagram, not a picture of the real chip.

Where it went

These processors power datacentres and supercomputers. They are essential for training large language models and running AI inference tasks. The GH200 module combines a Hopper GPU with a Grace CPU for unified memory access. Companies use them for high-performance computing applications that require massive parallel processing power and rapid data throughput.

Versions and family

The H100 is the primary accelerator model. Nvidia released the H800 and H20 with reduced bandwidth to comply with export restrictions. The GH200 integrates the GPU with a central processing unit. The H200 is a later variant that faced trade disputes in 2026. Ada Lovelace serves as the consumer-facing counterpart to this datacentre-focused architecture.

Why it matters

Hopper became central to the global AI infrastructure build-out. Its scarcity drove high secondary market prices and unusual shipping methods. The architecture influenced international trade policies regarding advanced semiconductor exports. It established new performance benchmarks for transformer-based workloads, cementing Nvidia's role in the artificial intelligence hardware sector.

Problems and reception

Industry leaders expressed urgent demand for the hardware during its initial release. Export controls created uncertainty for international buyers, particularly in China. Analysts noted significant price premiums on the secondary market. The architecture received attention for its specialized transformer engine, which addressed specific needs in modern deep learning applications.

Questions people ask about Hopper

What is the Hopper microarchitecture?

Hopper is a graphics processing unit design by Nvidia for datacentres. It focuses on accelerating artificial intelligence workloads using specialized hardware engines and high-bandwidth memory systems.

Who manufactures Hopper chips?

TSMC manufactures the chips using its N4 process node. Nvidia designs the architecture and handles the overall product development and sales for these datacentre accelerators.

What makes Hopper different from Ampere?

Hopper introduces a transformer engine for dynamic precision scaling. It also supports faster HBM3 memory and offers improved single-precision floating-point throughput compared to the previous Ampere generation.

What is the GH200?

The GH200 combines a Hopper-based GPU with a Grace-based CPU on a single module. This design allows for memory coherence between the processor and graphics unit via NVLink connections.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "Hopper (microarchitecture)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.

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.