AMD punts Ryzen AI 400 to business desktops and laptops

AMD’s Ryzen AI Pro 400 chip will be offered as a desktop processor, giving users another option for a low-power, efficient desktop or mini PC. But AMD is marketing both its desktop and mobile Ryzen AI 400s under its Pro brand for businesses, apparently conceding that consumers will pass them by.

Well, maybe that’s not entirely true. There’s nothing stopping a consumer buying one of the estimated 200 different PCs with a Ryzen AI Pro 400 (Gorgon Point) chip inside, including either a desktop, laptop, or even a workstation. They’ll ship from the usual suspects, including Acer, Asus, Dell, HP, and Lenovo.

Still, AMD’s Ryzen AI 300 performance was outstanding, and the company marketed them at both consumers and corporations alike. Now, the Ryzen AI Pro 400 chips are being designed in with AMD’s security initiatives for enterprises, alongside AMD’s existing Radeon AI Pro lineup of GPUs. Well, AIPUs, anyway: Absolutely nothing of AMD’s marketing mentions games, and instead positions the chips as powerful solutions for local LLMs. To reiterate the obvious: AMD is leaning hard into AI, and the corporate budgets willing to pay for it.

The desktop chips are new, though: Socketed desktop processors that AMD tipped off when it announced the Ryzen AI 400 family. There are three, but in two flavors: The “G” series denotes a 65W chip, and the “GE” suffix indicates a 35W chip. All of the chips include an NPU with 50 TOPS, qualifying them for Copilot+ status on desktop PCs. They’ll slot in to the AMD AM5 platform, supporting DDR5 memory modules of up to 8,533 megatransfers/s.

• Ryzen AI 7 Pro 450G/450GE: 8 cores/16 threads, 2.0GHz base/5.1GHz turbo, 24MB total cache; Radeon 860M graphics w/8 CUs
• Ryzen AI 5 Pro 440G/440GE: 6 cores/12 threads, 2.0GHz base/4.8GHz turbo, 22MB total cache; Radeon 840M graphics w/4 CUs
• Ryzen AI 5 Pro 435G/435GE: 6 cores/12 threads, 2.0GHz base/4.5GHz turbo, 14MB total cache; Radeon 840M graphics w/4 CUs

AMD’s mobile Ryzen AI Pro 400 chips were spelled out earlier by AMD, and contain more information than AMD provided in a recent briefing.

AMD’s Ryzen AI Pro 400 chips don’t use the performance or efficiency cores that Intel’s Core Ultra or Qualcomm’s Snapdragon X Elite chips do; instead, they use a “compact” version of the same core (Zen5c), typically manufactured in a finer 3nm process with a smaller die area.

Remember, gamers still have AMD’s superb Ryzen 9000X3D desktop chips, the Ryzen 9000X, and even the AMD Ryzen AI Max+ to choose from for desktop PCs and gaming. Still, we can at least hope that we’ll see a consumer release of the Ryzen AI Pro 400 chips, eventually.

Performance: It’s all about the CPU

Of the 2025 mobile platforms, AMD’s strength was in CPU performance; Intel excelled in 3D, while Qualcomm’s Arm architecture was the most energy efficient and competed strongly in battery life.

The Ryzen AI Pro 400 should continue the trend, AMD says. AMD is expecting 20 percent faster single-core performance and 30 percent faster multithreaded performance in Cinebench 2026, alongside 50 percent faster rendering in the CPU-only Blender “Classroom” demo file.

AMD also thinks that a Ryzen AI Pro 400 laptop should get about 19.8 hours of battery life using MobileMark 30, which uses a variety of real-world applications to estimate battery life. (I’ve struggled to find a good battery benchmark, and recently tested battery life using Netflix streaming as another metric.)

GPUs? Or AIPUs?

At Computex 2025, AMD launched the AMD Radeon AI Pro R9700, an AI-optimized GPU with 32GB of GDDR6 memory across a 256-bit memory bus. (More VRAM typically allows larger AI models to be run.) It featured 64 RDNA4 Compute Units (CUs), or about 4,096 streaming processors (SPs). Peak 8-bit precision (INT8 Matrix) performance is 383 TOPS, while single-precision FP32 performance is 47.8 TFLOPS.

Now, AMD has announced the Radeon AI Pro 9600 as well for AI workstations, which also uses 32GB of GDDR6 memory across a 256-bit bus. But it has a smaller number of compute units, 48CUs, with 3,072 stream processors, running at a game frequency of 1,080MHz and a boost frequency of 2,020MHz. Peak 8-bit precision (INT8 Matrix) performance is 199 TOPS, or about half that of the R9700. The same goes for peak single-precision (FP32 vector) performance, at 24.8 TFLOPS.

Finally, AMD also launched the Radeon AI Pro R9700S and Radeon AI Pro R9600D GPUs, both designed for enterprise racks.