AMD Unveils Its Fastest Edge AI Chips Yet: The Ryzen AI Max and Ryzen AI Max+ Strix Halo Families
Delivering up to 126 TOPS of compute, AMD claims its top-end laptop chip can run an LLM twice as fast as NVIDIA's RTX 4090.
AMD has taken to the stage at the Consumer Electronics Show (CES) 2025 to declare itself king of the artificial intelligence hill — unveiling the Ryzen AI Max and AI Max+ chips for running generative AI workloads on laptops, delivering up to a claimed 126 TOPS of compute.
"As consumers and professionals increasingly recognize the productivity benefits of AI PCs, AMD is further increasing its performance leadership in the market," claims AMD's Jack Huynh of the unveiling. "With the next generation of AI-enabled processors, we are proliferating AI to devices everywhere, and bringing the power of a workstation to thin and light laptops."
Like longstanding rival Intel, AMD is pushing the idea of the "AI PC" — devices, primarily laptops, featuring chips with integrated neural processing unit (NPU) accelerators for on-device machine learning and artificial intelligence (ML and AI) workloads. In June last year the company unveiled the Ryzen AI 300 series, its first to feature an NPU delivering a claimed 50 tera-operations per second (TOPS) of compute at minimum precision — designed to not only meet but exceed the requirements of partner Microsoft's Copilot+ program for AI-equipped Windows systems.
Now, the company has unveiled higher-end parts: the Ryzen AI Max and AI Max+ chips. Sitting at the top of the tree as what was previously codenamed the "Strix Halo" family, the new chips include the same 50 TOPS accelerator as their predecessors but with up to 16 Zen 5 processing cores running at a base clock of up to 3GHz and a maximum boost clock of up to 5.1GHz plus on-board Radeon graphics with up to 40 cores running at up to 2.9GHz. All told, the range-topping Ryzen AI Max+ 395 delivers a claimed 126 TOPS of compute at minimum precision — enough, the company claims, to run generative AI models, including large language models, on a laptop with good performance.
To prove that claim, AMD's morning keynote at the Consumer Electronics Show in Las Vegas today made a bold claim: that the Ryzen AI Max+ 395 can deliver more than double the performance for Meta's source-available Llama 3.1 70B-Q4 large language model than NVIDIA's GeForce RTX 4090 24GB graphics card — in a laptop form factor and using dramatically less power, ranging between 45W and 120W "configurable thermal design profile" depending on how the laptop manufacturer sets the part up. The parts also support up to 128GB of unified memory, of which up to 96GB can be dedicated to the graphics card.
Other parts announced during AMD's keynote speech include additional Ryzen AI 300 chips, all featuring the same 50 TOPS AMD XDNA 2 neural coprocessor, and the business-oriented Ryzen AI 7 PRO 350 and Ryzen AI 5 PRO 340 with its latest AMD PRO security subsystem.
More information on the Ryzen AI Max and AI Max+ parts is available on the AMD website; the company has not announced pricing, but states that devices built around the chips will be available in the first quarter of this year.