In the last few weeks, we’ve heard much about Microsoft’s Windows Copilot+ PC features dropping in mid-June, first with new Qualcomm Snapdragon-based hardware. At Computex 2024, AMD unveiled its latest AI-ready processors, the Ryzen AI 300 series, specifically made to power these next-generation AI PCs. With this announcement, AMD claims it has the first x86 processor to meet the requirements for Windows Copilot+ PCs, opening up a whole new branch on the system-on-chip (SoC) family tree.Meet AMD Ryzen AI 300, Now Running on Zen 5Slated to appear in several new AMD-powered laptops, the new Ryzen AI 300 features Zen 5 CPU architecture, with up to 12 high-performance cores (the upper limit used to be eight cores) and 24 processing threads. These chips’ graphics are powered by RDNA 3.5, which powers up to 16 compute units, delivering more graphics capability without a discrete GPU.
(Credit: AMD)
But the natural cherry on top is AMD’s next-gen neural processor unit (NPU) using XDNA2 chip architecture. After hearing both Qualcomm and Intel make big claims about the performance of their NPU hardware, AMD tops both, claiming up to 50 trillion operations per second (TOPS). That claim makes it the most powerful laptop SoC yet for AI laptops, at least in the spec-number wars.
AMD will launch the 300 series with two new chips, part of the Ryzen 9 product line, with the Ryzen AI 9 HX 370 and the Ryzen AI 9 365. Confused about AMD’s naming conventions with these new processors? You have every right to be, as the chip maker is once again revamping how it names its mobile processors. Check out our separate explainer on that.
(Credit: AMD)
Aimed at creators and gamers, these new third-gen Ryzen AI chips use AMD’s Zen 5 CPU architecture for multicore chips (with 12 and 10 cores) with Radeon 890M and 880M integrated graphics (with 16 and 12 compute units), respectively.
(Credit: AMD)
NPU Performance Reaches New HeightsThese Ryzen AI 300 chips use XDNA2 NPUs, which have received a flurry of improvements over the first generation. AMD has increased the number of compute tiles, providing five times the compute capacity—first-gen XDNA could only produce 10 TOPS. XDNA2 also doubles the chips’ multitasking capabilities as it runs AI functions in the background, whether that’s nearly invisible webcam image enhancement or the more exciting new features like Copilot Recall.
(Credit: AMD)
AMD is bragging hard on that 50 TOPS number, and rightly so. Microsoft’s Copilot+ requirements call for 40 TOPS, and Intel and Qualcomm have produced TOPS numbers in the mid-40s. As the first to 50 TOPS, AMD’s Ryzen AI 300 AI computing power also outpaces Apple’s claims of 38 TOPS using the new M4 chip in the iPad Pro. We usually try to add a dose of reality when companies make big claims in their marketing and press briefings, but if the 50 TOPS claim is accurate, it’s a real win for AMD.The Red Team has also boosted its NPU efficiency (compared with the original XDNA NPUs), leveraging architecture changes to get up to twice as much work done in generative AI tools for the same power. As a result, you’ll enjoy longer battery life, even as the size of the AI workload grows.AMD’s Big AI Win With Blocks and BitsPerhaps the most significant advancement made by the new Ryzen AI 300 series and XDNA2 technology is an obscure development called “Block FP16.”AI models generally process data in 8-bit and 16-bit formats, which refer to the amount of information or data that a large-language or image-generation model works with. Models using 16 bits are more accurate but use more memory bandwidth, which slows performance. Smaller 8-bit models allow for faster performance without that memory bottleneck slowing things down. But because 8-bit models work with less information to begin with, they’re not as accurate.
(Credit: AMD)
This creates a trade-off between faster models and more accurate models. Even though most models use the 16-bit datatype, performance requires shrinking the model data down (a process called quantization) to 8-bit to get faster performance at the cost of accuracy.AMD has introduced Block FP16 functionality to allow its NPU to run 16-bit models without quantizing them down and at speeds comparable to 8-bit datatypes. AMD won’t share the specifics of how this is done. Still, it means that developers can rev the engine a bit more on AMD machines with full 16-bit capability. Users will find that AI functions like image generation run smoother and produce better outcomes.Windows 11, No Emulation RequiredAMD threw some shade at recent Microsoft darling Qualcomm, the most significant player in the long-promised Windows-on-Arm revolution. But running Windows, designed around x86 hardware, on the Arm-based Snapdragon processors coming to many laptops and tablets, requires emulation for running most apps.
Recommended by Our Editors
(Credit: AMD)
That’s not so on AMD, an x86 platform that shows speedy performance across the board with no added CPU overhead from emulation requirements. AMD also presents better performance across the board in general productivity, multitasking, and graphics, as measured by benchmarks GeekBench 6.3, Procyon Office, Cinebench R24, and 3DMark.
(Credit: AMD)
AMD also claims faster performance than Apple M3. However, this comparison is slightly less exciting, given that AMD compares its top-of-the-line Ryzen AI 9 HX 370 with Apple’s basic M3 chip instead of its more capable M3 Pro or M3 Max options.
(Credit: AMD)
Performance against Intel also factored into AMD’s big reveal, with the firm comparing against Intel Core Ultra 185H processors and claiming better performance in productivity, creative tasks, and gaming.
(Credit: AMD)
People rarely buy laptops based on the gaming performance of the CPU’s integrated graphics, but this is a helpful illustration of the differences in graphics capability. We’ll just have to wait for our own testing and reviews to determine how these systems compare under real-world circumstances.AMD Ryzen AI 300 Is Coming SoonAMD’s new AI laptop chips will arrive as early as July, appearing in systems from Acer, Asus, Dell, HP, Lenovo, and MSI. More than 100 models will drop from various partners. We’ll see these Ryzen 300-series chips in models from Asus and MSI (and possibly others) during Computex.
(Credit: AMD)
Pricing and availability on the new Asus and MSI laptops have not yet been announced, but AMD says we should start seeing these systems sometime next month.
Get Our Best Stories!
Sign up for What’s New Now to get our top stories delivered to your inbox every morning.
This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You may unsubscribe from the newsletters at any time.
