News
Inferenceai& to Provide NEC with Sovereign AI Inference Platform to Accelerate AI-Native TransformationDealsGPUs reshape colocation versus private data center decisionsSovereignDeutsche Telekom boosts its sovereign AI push with 100 million for MistralNeocloudsSpaceX has a CoreWeave problem that no one is talking aboutInferenceAMD Completes Taalas Acquisition to Strengthen AI Inference PortfolioSovereignIndia’s Sarvam AI launches GPU service for open modelsClustersNew UB North Campus structure to house AI supercomputer system for research usesClustersIntroducing Together Storage Accelerator: fast, free data caching for your GPU clusterNeocloudsJensen Huang’s Charity Committed $350 Million to CoreWeaveClustersOak Ridge's AI Factory Moment: Lux, Nuclear Power, and the Next Data Center FrontierInferenceai& to Provide NEC with Sovereign AI Inference Platform to Accelerate AI-Native TransformationDealsGPUs reshape colocation versus private data center decisionsSovereignDeutsche Telekom boosts its sovereign AI push with 100 million for MistralNeocloudsSpaceX has a CoreWeave problem that no one is talking aboutInferenceAMD Completes Taalas Acquisition to Strengthen AI Inference PortfolioSovereignIndia’s Sarvam AI launches GPU service for open modelsClustersNew UB North Campus structure to house AI supercomputer system for research usesClustersIntroducing Together Storage Accelerator: fast, free data caching for your GPU clusterNeocloudsJensen Huang’s Charity Committed $350 Million to CoreWeaveClustersOak Ridge's AI Factory Moment: Lux, Nuclear Power, and the Next Data Center Frontier
EveryNeocloud

Nvidia Hopper

H100, H200 and GH200. 43 providers on EveryNeocloud so far.

Hopper is the Nvidia GPU generation that powered the first wave of the generative AI boom. The H100, announced in March 2022 and shipped from late that year, became the most sought-after chip in the world in 2023. It was followed by the H200, with more memory, and the GH200 Grace Hopper Superchip, which pairs a Hopper GPU with an Arm-based Grace CPU on one board.

The chip

The H100 has 80 billion transistors on TSMC's 4N process and introduced the Transformer Engine, hardware that switches to 8-bit numbers (FP8) for parts of a model where precision matters less. It comes as an SXM module for eight-GPU servers or as a PCIe card. The H200 of 2024 raised memory to 141 GB of HBM3e. For China, Nvidia sold cut-down versions, the H800 and later the H20, under US export rules.

Where it runs

Most neocloud fleets were built on H100s: CoreWeave, Nebius, Lambda, Northern Data and India's Yotta and E2E all grew on them, and Colossus began with about 100,000. The GH200 is the base of several national supercomputers, including JUPITER, Alps and Isambard-AI. As Blackwell arrives, H100 rental prices have fallen, and many H100 clusters now serve inference and smaller training jobs.

Famous nvidia Hopper providers

Still to be written: Jupiter (supercomputer), AI Bridging Cloud Infrastructure.

Further reading

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 provider of the day, new pages on EveryNeocloud, and the conferences and anniversaries of the week ahead.