Fujitsu and AMD are partnering to develop energy-efficient, open-source HPC and AI platforms by 2027. Fujitsu will contribute ARM-based processor technology, including a new 288-core chip, while AMD provides GPU expertise. The collaboration aims to advance technology, commercialisation, and ecosystem expansion for AI and HPC applications.
Google Cloud has unveiled Trillium, its new AI chip with 4x faster training and up to 3x higher inference throughput than its predecessor. Trillium features doubled memory and bandwidth, supports large language models, and can link up to 256 chips in a pod. It boasts 67% better energy efficiency. Google also announced upcoming preview of A3 Ultra VMs with Nvidia H200 GPUs.
An $11 billion debt market has emerged for AI companies buying Nvidia chips, raising concerns about Nvidia's AI market dominance. This funding targets 'neocloud' groups, likely new AI-focused cloud computing companies, reflecting Wall Street's enthusiasm for AI technologies.
A study predicts AI-related e-waste could grow from 2,600 tonnes in 2023 to 2.5 million tonnes by 2030, potentially reaching 1.2 to 5 million tonnes cumulatively between 2020 and 2030. North America is expected to produce the majority, followed by East Asia and Europe. US restrictions on GPU sales to China may exacerbate the issue. Researchers propose a circular economy strategy to potentially reduce AI-related e-waste by 86% globally, including extending hardware lifespan, reusing components, and improving computing efficiency.
