Nvidia Unveils Hopper H100 Datacenter GPU With 80 Billion Transistors
Nvidia introduces the Hopper H100 GPU at GTC 2022, packing 80 billion transistors on a custom TSMC 4N process to succeed the Ampere A100 in supercomputing and AI workloads.
Nvidia reveals its new Hopper architecture and the H100 GPU at GTC 2022, marking a major leap for datacenter computing. Built on a custom TSMC 4N manufacturing process, the H100 packs an impressive 80 billion transistors compared to the 54 billion found in the previous Ampere A100 generation.
The new chip focuses heavily on AI and supercomputing performance to compete with rivals like AMD's Instinct lineup. It features fourth-generation NVLink technology that delivers up to 900 GB/s of bandwidth and includes PCIe 5.0 support for systems that require up to 128 GB/s of throughput.
Memory capabilities also see a significant upgrade with the H100 supporting 80GB of HBM3 memory right out of the gate. This memory configuration provides 3 TB/s of bandwidth, representing a 50 percent increase in both memory capacity and speed over the existing A100 datacenter cards.