TECH NEWS | AMD Pushes Agentic PCs With Ryzen AI Max
AMD defines an Agentic PC as a system combining a GPU, CPU and NPU to run AI agents.
The shift toward PCs that can run AI agents locally is changing the role of the personal computer, as AI workloads move beyond simple prompt-and-response interactions toward tasks that require planning, reasoning and repeated execution.
AMD used an online briefing Wednesday, Sept. 30, to outline its approach to agentic computing, with Michael Nordquist, corporate vice president for client product marketing, discussing the company’s Ryzen AI Max processors and broader PC portfolio.
Nordquist said the distinction is increasingly about where and how AI work is performed. “As we move into agentic, that means you have an agent working on your behalf on more complex tasks where you really try to assign it a task or an outcome and actually reason through it,” he said.
In its accompanying presentation, AMD defines an Agentic PC as a system combining a GPU, CPU and NPU to run AI agents. “The term “Agentic PC” is defined as a PC with a GPU, CPU and NPU combination that allows the user to run agents performantly,” the presentation said. AMD added that Ryzen AI Max+ systems with 64GB or more of memory are designed to provide an experience for demanding agentic workflows.
The presentation said AMD had shipped more than 500,000 Agentic PCs across more than 50 designs, compared with tens of millions of AI PCs across more than 250 designs. The company’s portfolio spans consumer and commercial AI PCs, developer platforms and workstation-class systems.
For Nordquist, the move to local agentic computing changes the balance of hardware requirements. Earlier AI PC workloads could rely heavily on the GPU, but agents can repeatedly execute tasks, retrieve information, reason over data and call applications. That puts more emphasis on CPU resources, memory capacity and the ability to run larger models locally.
“I think people have seen all of a sudden it’s the resurgence of the CPU, right? It’s not just about pinging a GPU somewhere and having it respond, it’s running tasks over and over,” Nordquist said.
Ryzen AI Max combines a Zen 5 CPU with up to 16 cores and 32 threads, up to 192GB of LPDDR5x unified memory, AMD RDNA 3.5 graphics and an AMD XDNA 2 NPU, according to the presentation. AMD said the architecture can allocate as much as 160GB of the unified memory to graphics workloads, allowing larger models to be held locally.
The presentation also highlighted local execution of large language models. AMD reported that its Ryzen AI Halo platform, powered by a Ryzen AI Max+ PRO 495 processor with 192GB of memory, was tested running a 320-billion-parameter GLM 5.3 Flash model locally at up to 20 tokens per second. A separate test involving Qwen 3.8 Flash Next models totaling 176 billion parameters reached up to 42 tokens per second.
AMD said the results were based on testing conducted in September 2026 and that performance may vary. The tests used specific software, memory and quantization configurations, making the figures performance measurements under AMD’s stated test conditions rather than general benchmarks for every system.
Nordquist said local processing could also address the cost of running agentic workloads in the cloud, where usage is increasingly tied to token consumption.
“We’ve seen all the big cloud vendors right now as we moved into agentic, and the agents are the ones that are driving the tokens, they have a meter turning that are saying how many tokens are you using and you’re paying per token,” he said. “If you’re able to take that and run it all locally, sure you have the device cost or you have the power cost, but it’s essentially almost like free tokens that you’re able to do.”
The strategy extends beyond processors. AMD’s presentation lists Ryzen AI Max processors, the Ryzen AI Halo Developer Platform, Ryzen AI processors, Ryzen AI PRO processors, Radeon AI PRO graphics and Threadripper Halo Station as parts of its PC portfolio, with ROCm software spanning the platforms.
Nordquist said software and ecosystem support would be critical as enterprises move from experimentation to deployment.
“Having hardware is great, but if you don’t have the ecosystem and the software infrastructure for it, you really can’t capitalize on some of the capability,” he said.
The presentation identified Perplexity, Microsoft and Cisco among AMD’s ecosystem partners. It said Perplexity’s Portable Computer runs on Ryzen AI Max-based systems and can handle local AI tasks and recurring workflows, while Microsoft selected Ryzen AI Halo as the first platform for its Project Zenith developer experience. AMD also said Ryzen AI Halo and Cisco are intended to help enterprises deploy and manage local agentic AI.
For enterprise users, AMD is positioning its Ryzen AI PRO line around security, manageability, quality and long-term support. Nordquist said enterprises are looking for ways to introduce local AI without abandoning existing management and security practices.
“We know enterprise wants to have that security, they want to have that governance, they want to have the manageability to see seamlessly deploy from names that they trust,” he said.
The broader portfolio approach reflects AMD’s view that agentic computing will not be confined to one class of machine. Its presentation places consumer and commercial AI PCs alongside developer platforms and commercial workstations, with Ryzen AI Max positioned between thin-and-light PCs and higher-performance workstation systems.
The practical question for users and IT departments is therefore shifting from whether a PC has an NPU to what workloads can actually be run locally, how much memory they require, which applications support local execution and how those systems can be managed. AMD’s latest portfolio is aimed at that transition, while its presentation notes that performance varies by workload, software configuration and system design.
Full disclosure: All news articles published on the TechSabado website are written by human journalists, unless otherwise specified. Final text editing is also performed by human editors, with artificial intelligence (AI) used only to assist with additional grammar and style guide corrections..
————————————————————————-
TEN YEARS OF TECHSABADO!
PLEASE LIKE our FACEBOOK PAGE and SUBSCRIBE to OUR YOUTUBE & SPOTIFY channels.
WATCH TECHSABADO ON OUR YOUTUBE CHANNEL:
WE ARE ALSO ON SPOTIFY
WATCH OUR OTHER YOUTUBE CHANNELS:
PLEASE LIKE our FACEBOOK PAGE and SUBSCRIBE to OUR YOUTUBE CHANNEL.
