This year’s developments in laptop processors have been dominated by one trend: AI. The AMD Ryzen AI 300 (aka “Strix Point”) family puts that next-generation capability front and center, with a revved-up neural processing unit (NPU) capable of 50 trillion operations per second (TOPS). This outpaces Qualcomm’s Snapdragon chips’ TOPS rating and is ready to grapple with Intel’s upcoming “Lunar Lake” processors.We’ve enjoyed a steady drip of details regarding the new chips for months, but we finally have one in the lab, powering the new Asus Zenbook S 16 (UM5606) laptop. Driven by an AMD Ryzen AI 9 HX 370 processor, it’s the best you can get from AMD today. From a new 4-nanometer (nm) process to architecture upgrades in every part of the chip, it’s the most advanced silicon AMD has ever put in consumer hands.I’ve put the new system through its paces, running not only our standard panel of tests but also new tests added specifically to measure AI performance and the latest features. Let’s see how it did.
(Credit: Brian Westover)
About the AMD Ryzen AI 300 SeriesThe new AMD 300 series is built from the ground up for AI workloads. From the CPU cores and integrated GPU to the hotshot NPU, AMD has done an architectural revamp on the latest Ryzen chips that’s worth detailing. It all starts with a 4nm process node, letting the manufacturer pack more transistors onto the chip surface and get corresponding improvements to speed and energy efficiency. With dedicated processing hardware for general functions, graphics, and AI, it’s the most advanced system on a chip (SoC) the company has made. (Check out our technical deep dive for a more detailed look at the new chips.)
(Credit: AMD)
The CPU: ‘Zen 5′ CoresAMD’s primary chip architecture gets an upgrade with “Zen 5,” AMD’s latest architectural flavor, boosting the capability of individual processing cores with twice the instruction bandwidth and doubling the data bandwidth. With improved instruction fetch, broad integer execution, and increased data bandwidth, AMD is claiming a 16% boost in instructions per cycle compared with Zen 4.But AMD has also pushed the envelope in terms of core count and thread handling, increasing to 12 high-performance cores (as opposed to the old limit of eight) and handling up to 24 simultaneous processing threads as a result. All of this gets an extra polish with improvements to multitasking, refining how the CPU schedules and prioritizes tasks, and handling larger, more complex tasks with improved data usage.The result is a high-performance CPU that can boost to 5.1GHz and has a thermal design power (TDP) rating of 28 watts—all ready for OEMs’ thin-and-light chassis designs.The GPU: RDNA 3.5The new AMD Ryzen AI 9 HX 370 also gets evolved graphics hardware, using Radeon 890M graphics with RDNA 3.5. The integrated graphics get a major boost with 16 compute units and better performance per watt, squeezing more graphics and gaming performance out of every watt of power. Improvements like a double-rate texture sampler and enhanced interpolation/comparison rates mean smoother visuals and higher frame rates, even allowing gaming at 60fps and 1080p resolution without additional discrete graphics hardware. With improved memory management that reduces retrieval from slower memory units, the whole thing works with less power, delivering 32% faster performance per watt.
(Credit: AMD)
The NPU: XDNA 2Finally, we reach the NPU, the heart of the 300 series’ new AI capabilities. This dedicated processing engine just for AI workloads joins the CPU and GPU on the SoC, handling complex AI tasks without monopolizing the resources used for standard processing and graphics.AMD first introduced the NPU with XDNA, a spatial dataflow architecture that uses tiled processing units and on-chip memory. The new 300 series shifts to XDNA2, which expands the number of engine tiles to 32 (which is 12 more than the previous generation). The result is that XDNA 2 delivers a big boost to power efficiency and total output, with much higher rated NPU TOPS (from 10 TOPS to 50 TOPS).The chip maker has also introduced support for Block Floating Point 16 (BFP16), which combines the performance of 8-bit operations with the accuracy of 16-bit. The math behind it is complicated, but the result is better precision in AI processes without the trade-off in power usage.It all adds up to faster, smarter AI performance for everything from automatic image enhancement and system optimization to on-device large-language models and generative art tools, even video processing and content-creation workflows—all features that used to rely on powerful discrete GPU hardware to run.The Test System: Asus Zenbook S 16 (UM5606)The new 300-series chips are first seen in the Asus Zenbook S 16 (UM5606), which is equipped with an AMD Ryzen AI 9 HX 370 processor and Radeon 890M graphics.
(Credit: Brian Westover)
Part of Asus’ seemingly endless range of Zenbook models, the Zenbook S 16 (UM5606) is a 16-inch, ultra-slim laptop. Too big to be an ultraportable but much thinner than most desktop replacements, the Zenbook S 16 has plenty of premium touches, like a ceramic-bonded finish that Asus calls Ceraluminum. A sophisticated chemical and electric process fuses the aluminum chassis with a ceramic layer that’s smooth to the touch, distinctive in style, and highly durable.The premium quality extends to the 120Hz, 3K-resolution OLED touchscreen display and the giant feature-filled touchpad.
(Credit: Brian Westover)
From the rich port selection to the 1TB SSD inside, you can tell that this is a well-built machine. I’ll save the extended discussion of features and specs for our full review, but suffice it to say that AMD’s latest, most powerful hardware is well-matched by the system housing it.The Asus Zenbook S 16 (UM5606) sells for $1,699.99 through the Asus website and major retailers. Available for purchase now, the first units will begin shipping on July 29.Testing the Asus Zenbook S 16: New Chips Demand Old and New TestsTo get a clear picture of how well the new Ryzen AI 300 series fits in the landscape of modern AI-infused processors, I’ve compared it with not only past AMD Ryzen chips but also Intel Core Ultra processors and the recent Qualcomm Snapdragon Elite X chips used to usher in Windows Copilot+ PCs.
That means testing it with our usual batch of system tests, like PCMark 10 and HandBrake, and using newer benchmarking tools, like Cinebench 2024 and Geekbench 6. We’ve also used Geekbench 6 ML, a machine learning benchmark that (at present) seems to be the most straightforward test for AI capabilities.We haven’t run all of these new tests on every machine we compared the Zenbook with, but we have a broad representation, and it’s more than enough to show us how well the Ryzen AI 9 HX 370 fares on these next-gen performance measures.There’s also the issue of chip type. The Ryzen AI 9 HX used in the Zenbook S 16 is closer to an Intel Core Ultra 9 than a Core Ultra 7, but we haven’t reviewed many of those, and those we have tested have a different performance profile due to being paired with dedicated graphics chips. Most of our comparisons also aren’t high-performance “HX”-class chips. But those details will have to shake out in the coming months as we review more systems and adjust our testing regimen.Productivity and Content Creation TestsI started with some familiar names for CPU-specific testing, using the same benchmarks we use in most of our laptop reviews. UL’s PCMark 10 simulates a variety of real-world productivity and office workflows to measure overall system performance, giving a simple, easy-to-compare score. HandBrake 1.4 performs a simple video transcode, converting a 12-minute 4K file down to a 1080p, 30fps version, testing the CPU’s performance in handling intensive media tasks.Maxon’s Cinebench R23 uses the Cinema 4D engine to render a complex scene, providing a simple but intense workload to stress the processor over a 10-minute period. I also tested it with the newer Cinebench 2024, which also uses the Cinema 4D engine to render complex scenes but provides comparable versions of the test that run on Qualcomm’s Snapdragon processors, giving us a comparison point for the first batch of Copilot+ PCs.Similarly, I ran Geekbench 6.3, which is not only cross-compatible with Arm systems but also uses a more up-to-date workload that simulates a range of popular applications to measure single-core and multi-core performance.Finally, I also included 7-Zip’s built-in compression benchmark, which is again compatible across AMD, Intel, and Arm, giving us a point of comparison across the three different processor platforms. The test measures compression and decompression speeds, and uses multithreaded processes, giving a better idea of a CPU’s multi-thread capability.
Recommended by Our Editors
In PCMark 10, the Zenbook S 16 edged ahead of the other top scorers, achieving nearly double the 4,000-point threshold that indicates acceptable everyday productivity. In HandBrake, where shorter times indicate faster performance, the Zenbook S 16 was one of the fastest of the bunch, sometimes by a dramatic margin. In both versions of Cinebench, the AMD-powered Zenbook S 16 produced excellent results, putting last year’s AMD chips to shame and edging ahead of Intel Core Ultra 7 models. However, you’ll see slightly better results in the Snapdragon-powered systems I compared it with. The highlight of the newer tests I ran was Geekbench 6, where the Zenbook’s Ryzen 9 processor delivered one of the best scores.These results point to stellar performance in everything from basic productivity to more advanced media work, like video editing. And, with full x86 support, the AMD system won’t be held back by the app support and emulation issues we saw on the recent Snapdragon products.Graphics TestsI again used a mix of old and new tests for graphics testing. These mostly include a handful of newer tests from 3DMark, on top of the usual 3DMark Time Spy and Night Raid tests we normally use. I added 3DMark Wild Life (both regular and Extreme versions of this test). These two tests, run in their Unlimited variants, are compatible with laptops and smartphones, using the Vulkan graphics API at 1440p resolution to measure GPU speeds. The Extreme version ups the resolution to 2160p, or 4K, further stressing the graphics chips.For more graphics testing, I added 3DMark’s Steel Nomad and Steel Nomad Lite, which focus on APIs more commonly used for game development, like DirectX 12 and Metal, to produce insights more closely aligned with how games can expect to perform on the system, with an increased focus on more geometry and particle effects. None of these tests is ray-traced. Higher scores are better.Finally, I ran our standard GFXBench tests, a cross-platform GPU benchmark that stresses low-level routines like texturing and high-level, game-like image rendering. The 1440p Aztec Ruins and 1080p Car Chase tests, rendered offscreen to accommodate different display resolutions, exercise graphics, and stress compute shaders using the OpenGL programming interface and hardware tessellation, respectively. The more frames per second (fps), the better.
For a system with only integrated graphics, its results were superb. Using the AMD Radeon 890M graphics solution built into the SoC, the Zenbook S 16 managed to top most competitors in most tests. Sometimes, the scores proved high enough to support 1080p gaming at decent frame rates, even without a dedicated GPU. For average use, however, the clear takeaway is that everything from streaming to content creation will be well supported.AI TestsAI features, particularly NPU hardware, are rarely addressed in most benchmarks, so we’re still hashing out how our regular testing will account for these new capabilities. But for this round of tests, I relied on Geekbench ML, a machine-learning test that focuses on either the CPU or the entire SoC.
In the DirectML version of this test, the Ryzen’s XDNA2 NPU showed serious chops, giving the Zenbook S16 the best score of every laptop tested, with nearly three times the performance of the Snapdragon systems (which, admittedly, are not optimized for this particular benchmark set). In CPU-specific testing, however, the older AMD Ryzen 7 from 2023 pulled ahead—indicating better performance, yes, but only in scenarios where solely the CPU would be relied upon to handle this particular subset of AI tasks. Again: AI benchmarking is in its very early days, especially across platforms. AI performance, by and large, varies wildly depending on the API in play, and whether all or some of the parts of the processor are employed. And a key thing to remember: In some cases, running an AI task efficiently (if more slowly) on a sufficiently resourced NPU is the whole point. In other words, sometimes “the race is not to the swift,” in the larger picture.Verdict: An AI Powerhouse, Ready to RoarWhether it was basic productivity tasks, demanding graphics benchmarks, or advanced AI workloads, the AMD Ryzen AI 9 HX 370 proved a worthy contender in the ongoing AI laptop title fight. It consistently delivered top scores in this first taster, but the efficiency and capability improvements delivered by AMD’s new NPU hardware are perhaps the most exciting thing we’ve seen in the relatively new AI device category. As the laptop industry shifts to focus on AI features running on-device, that combination of powerful CPUs, capable graphics, and beefed-up NPUs will determine whether these AI capabilities sink or swim as a long-term industry change.I have some doubts near-term about Qualcomm Snapdragon X-based laptops (our first taste of the “AI PC”), due to the viability of Windows on Arm and pressures from the established x86 crowd. And we still have yet to see and test Intel’s Lunar Lake chips, expected later this year. AMD may not stay on top, but for the moment, the AMD Ryzen AI 9 HX 370 makes a standout performance that sells the concept of AI laptops better than anything we’ve seen so far.
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.
