Year: 2025
-
BWR 5: AI Creates Seismic Waves in Data Center Interconnect
Bob and Joe discuss Marvell’s Celestial.AI acquisition, Amazon Trainium using NVLink Fusion, Meta deploying Google’s TPU, and more.
-
Google TPU Draws Attention Amid Meta’s Interest
The recent focus on Google’s TPU, following Gemini 3.0 and Meta’s potential adoption, signals an evolving AI chip landscape. Meta’s move to own and house a TPU pod would be a significant shift from Google’s standard cloud-service model, posing a fresh challenge to GPU vendors.
-
Power11 Is a Small Step for IBM, Lodestar for Others
IBM’s Power11 processor introduces technologies like external memory buffers, AI acceleration, and SMT that will become common in other systems. Power11 refines the Power10 design, achieving a 14–50% speedup in IBM’s applications and offering up to 256 cores in a chassis.
-
The End Is Not Nigh
Leading up to Nvidia’s recent earnings call, concerns mounted that the company would confirm fears that the AI bubble would soon burst. Too many companies are prodigiously spending to build the next data center to create the next foundation model. Profitability is well over the time horizon, and their annular funding arrangements are unsustainable. Nvidia…
-
Arm Neoverse Fuses NVLink
Arm is adding Nvidia NVLink Fusion to its Neoverse platform. Neoverse is Arm’s family of infrastructure CPUs. NVLink Fusion is the licensable version of Nvidia’s chip-to-chip interconnect. Companies licensing Neoverse CPUs and peripheral technologies (IP), including soft IP or a CSS hard macro, will now receive Fusion support. This will help them to build chiplets…
-
AMD’s Embedded Strategy Lacks Bite
AMD revealed its embedded-processing strategy at its recent analyst day and set growth targets. Less noticed than data-center and client businesses, the embedded unit is called out in the company’s quarterly financial statements. It faces the same growth pressures but unique dynamics. The diverse embedded market is in disarray at the high end, and AMD’s…