Agentic AI and Why Everyone Wants CPUs Again
Clip title: Agentic AI and Why Everyone Wants CPUs Again Author / channel: TechTechPotato URL: https://www.youtube.com/watch?v=Q_QVlJkyaWc
Summary
Here’s a summary of the video:
The speaker, Dr. Ian Cutress, a semiconductor analyst and speaker, shares his unexpected experience at a conference in Long Beach. Originally scheduled as an attendee, he was asked to fill in for a keynote speaker who couldn’t make it. Cutress presented a talk titled “Agentic AI and Friends: Why Everyone Suddenly Wants CPUs Again.”
The main topic revolves around the intersection of AI, specifically agentic AI, and the semiconductor industry, highlighting the resurgence of CPU demand driven by AI advancements. Cutress discusses various contenders in the CPU market, such as AMD, Intel, and Nvidia, and their approaches to supporting agentic AI workflows. He explores the concept of agentic AI as a loop rather than a simple prompt, emphasizing the need for robust computing power, memory, and predictability to manage these complex workflows effectively.
Cutress addresses the challenges of CPU design and packaging, including the shift towards multi-die architectures, thermal management, and advanced packaging techniques. He emphasizes the importance of understanding the demands of agentic AI and the need for the industry to adapt to these new requirements. The speaker concludes with three main takeaways:
- Agentic AI is a loop, not a prompt.
- The rack is the end product providing the agents’ tokens.
- It all lands on the package, emphasizing the complexity of packaging and the need for continued innovation in semiconductor technology.
Video Description & Links
Description
Agentic AI runs its tools on the CPU, and no vendor tells you how many cores an agent needs. In this keynote at the Design Automation Conference, I compare AMD Venice, Intel Xeon 6, the Arm AGI CPU and Nvidia Vera, then follow the problem down into advanced packaging, CFETs and thermals. You also hear what EDA tools need to handle once agents and 3D stacking reach chip design.
[00:00] Intro and the last-minute call [06:02] Defining agentic AI [10:24] How much CPU per agent [12:17] AMD Venice and Intel Xeon [16:28] Arm AGI CPU and Nvidia Vera [19:06] x86 versus Arm [22:16] CPUs move to advanced packaging [25:57] CPU market and EDA agents [29:59] Wafer scale, glass and memory [34:39] Moore’s law and CFETs [43:44] Audience Q&A [52:33] Wrap up
Mark Papermaster interview: https://www.youtube.com/watch?v=yUBzu7oTTDo&t=2325s Huawei LogicFolding: https://www.youtube.com/watch?v=N9aSPnbz3co IBM 7A Nanostack: https://www.youtube.com/watch?v=_mRJw2F6j5Q Jim Keller interview: https://www.youtube.com/watch?v=AFVDZeg4RVY&t=4900s
If you’re in the market for something from Amazon, please use the following links. TTP may receive a commission if you purchase anything through these links.
Ending music: https://www.youtube.com/watch?v=2N0tmgau5E4
Welcome to the TechTechPotato (c) Dr. Ian Cutress Ramblings about things related to Technology from an analyst for More Than Moore
AgenticAI DAC CPU
Tags
agentic ai, ai cpu, design automation conference, amd, intel, nvidia, arm, amd epyc, epyc venice, zen 6, intel xeon, xeon 6, diamond rapids, nvidia vera, arm agi cpu, x86 vs arm, cpu vs gpu, tsmc, chiplets, hbm, advanced packaging, eda, cadence, synopsys, cerebras, wafer scale, ibm, cfet, moore's law, risc-v, tenstorrent, qualcomm, lisa su, lip-bu tan, semianalysis, brian fuller, mark papermaster, amd advancing ai, data center, ai hardware, semiconductors, inference, keynote, ian cutress