AI vendors are turning to custom hardware as Microsoft winds back the clock on Windows
Vendors are moving way beyond Nvidia’s GPUs in the datacenter
This week's episode of The Kettle delves into the evolving landscape of AI hardware and recent updates to Microsoft's Windows 11 operating system. While GPUs have long reigned supreme in data centers, vendors are increasingly turning to alternative silicon solutions like those offered by companies such as Cerebras and Marvell. Tobias Mann, a Register Systems Editor, explains that Cerebras, founded by Andrew Feldman, set itself apart by designing chips larger than traditional GPU-sized chips.
These large-scale chips boast 44 gigabytes of super-fast SRAM and can serve large language models faster than any GPU on the market. However, Cerebras faced challenges with compute capabilities and turned to partnerships, such as one with AWS, to address the compute-heavy side of inference. The company's latest generation of chips, the CS-4, offers a modular rack-scale system with improved power delivery, doubling the clock speed and enhancing memory and I/O speeds.
As AI hardware continues to advance, Microsoft has also made strides by reintroducing customizable taskbar functionality and providing users with greater control over context menus and neural processing units (NPUs) in Windows 11's Task Manager.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
