In this episode of The Tech Poutine, hosts Ian and George discuss recent developments in space missions, semiconductor innovations including ARM’s new data center CPU, and emerging AI hardware startups, while critiquing industry marketing and technical challenges. They also cover personnel changes, manufacturing dynamics, and broader tech topics such as Nvidia’s investments and memory optimization research, concluding with a call for community support.
The Tech Poutine episode kicks off with hosts Ian and George discussing the unusually warm weather in London and their recent busy schedules filled with travel and events. They touch on the recent Artemis 2 lunar flyby mission, highlighting the stunning Earthset photos and the challenges of sharing raw data from space missions. The conversation then shifts to the semiconductor industry, with George showcasing a new Panther Lake laptop and Ian sharing a rare AMD Seattle ARM-based server chip, reflecting on the evolving landscape of silicon and hardware design.
A significant portion of the discussion centers around ARM’s announcement of their first in-house designed merchant silicon data center CPU, the ARM AGI CPU. The hosts critique the launch event’s lack of detailed technical information and marketing clarity, noting discrepancies in power ratings and performance claims. They explore ARM’s strategy of offering a CPU designed primarily for select customers like Meta, rather than competing broadly with their clients. The conversation delves into the CPU’s specifications, comparisons with AMD and Intel processors, and the challenges ARM faces in product marketing and communication.
The episode also covers emerging AI hardware startups such as Talis, which specializes in ASICs tailored to specific AI models, offering high token processing speeds with a unique compute-in-memory architecture. The hosts discuss the trade-offs of such specialized chips, including potential obsolescence and deployment strategies. Additionally, they mention Dmatrix’s acquisition of Giga.io’s fabric business, highlighting ongoing innovations in AI hardware and interconnect technologies like CXL and PCIe advancements, including the recent PCIe 8.0 live demo showcased at an optical fiber conference.
Industry personnel changes and market dynamics are also addressed, including the retirement of Leslie Barnes, AMD’s head of CPU cores, and the appointment of Suzanne Plameumber as his successor. The hosts touch on the challenges of maintaining profit margins in hardware manufacturing, the impact of memory pricing on devices like the Raspberry Pi, and Intel’s recent repurchase of a 49% stake in their Ireland Fab 34, signaling confidence in their manufacturing capabilities. They also discuss the complexities of nested chip designs, such as Nvidia’s DGX Spark and MediaTek collaborations, and the software ecosystem challenges these pose.
Towards the end, the hosts reflect on broader industry topics, including Nvidia’s investments in various companies, the significance of Nvidia’s Vera CPU architecture, and the importance of gender balance in tech interviews. They highlight innovative research in memory refresh optimization by Lori Wyatt and share personal anecdotes about gaming preferences and industry events. The episode concludes with a call to support the channel through likes, subscriptions, Patreon, and merchandise, emphasizing the community’s role in sustaining their content creation.