Atlas record: https://www.aitimeline.dev/en/events/evt-0270/
Availability · March 2022
NVIDIA H100
The new generation got a dedicated transformer unit: hardware designed for one specific model architecture.
Why it matters
The circle closed: first the method was fitted to the hardware, now hardware is designed for a method named in a 2017 paper.
The unit picks precision automatically for each transformer layer, speeding training without losing quality. The moment is telling: the name of a neural network architecture appears in a chip's specification. By 2023 demand for these accelerators exceeded supply: The Batch reported in August 2023 that cloud providers had begun having trouble finding GPUs earlier that year and that the shortfall had spread to AI companies large and small, so that access to them became a limiting factor for whole companies. NVIDIA's own announcement of 22 March 2022 gave availability from the third quarter of 2022 and claimed a speed-up of up to 6x for transformer networks.
Event record
- Event date
- March 2022
- Timeline date
- Event date
- Verification
- Sources gathered automatically · September 17, 2026
- Lines
- ID
- evt-0270
The Hopper architecture was announced in March 2022.
Related events
Earlier
- Extends NVIDIA A100 2020
The next generation, with a unit aimed specifically at transformers.
- Related Transformer 2017
The hardware unit is designed for the attention mechanism.
Later
- Enables HBM3 for the H100 2022links to this record
The press release names the buyer: HBM3 goes into the H100, and supply is matched to the shipping schedule of Hopper systems in the third quarter of 2022.
- Related QLoRA 2023links to this record
Instead of bigger hardware, lower precision and fewer parameters that change.
- Extends Blackwell 2024links to this record
The next generation after Hopper, from the same manufacturer.
- Related NVIDIA becomes the most valuable company in the world 2024links to this record
Demand for these accelerators in particular produced the revenue behind the valuation.