Where Intel's new CPU designs are a little too split between optimizing for AI and optimizing for everything else, the chips AMD debuted at CES 2025 show more consistency: If they have an NPU, it's the current generation rather than a mix. So if it's got "AI" in the name, that means it has the 50 TOPS (max) architecture, so that's one less tradeoff to worry about when buying a laptop. And AMD's CES looks more interesting, with new designs like the Ryzen AI Max and Max Plus chips, the mobile launch of its gaming-optimized X3D stacked-memory architecture in the Ryzen 9 9955HX3D, and an expansion to its desktop X3D line with the Ryzen 9 9950X3D that has a technology twist intended to make it as good for creating as it is for gaming.
The company also delivered more information about its new RDNA 4-generation of graphics architecture, which underpins the new Radeon RX 9000 series desktop GPUs, starting with the RX 9070 and RX 9070 XT (shipping soon). AMD's FidelityFX Super Resolution 4, the new version of its upscaling and optimization software, builds on the capabilities of RDNA 4.
There's also a profusion of the usual expansions of CPU lines. In this case they go to lower-end parts of the Ryzen AI chips launched this summer, the new generation of Ryzen 9 HX-series performance-focused CPUs, a minor refresh of its Z series (with the Ryzen Z2, Z2 Extreme and a new lower-end Z2 Go) and more.
Radeon RX 9070, 9070XT, 9060
We don't have real specs yet for these new desktop graphics cards. We just know they'll be shipping this quarter from AMD's usual partners, such as Acer, Asus, Sapphire, XFX, ASRock, Gigabyte, PowerColor and more.
We do know AMD has reworked its branding to match Nvidia's. Notice the -70 and -60 in the names? That's intended to indicate which GeForce RTX card they compete with... at least for the RTX 4000 cards. What happens if, say, a new RTX -60 performs like the 4070? The company has also skipped an RX 8000 series and gone straight to RX 9000, because its current desktop chips are there. RX 8000 will live on in laptops, as a way to indicate that the GPU is of the RDNA 3.5 generation. Which brings us to to 9000 series architecture...
RDNA 4. This generation incorporates XDNA 2, the same AI architecture as the current CPUs. It also updates the ray tracing (third gen) -- I really hope it's gotten faster -- and an updated Radiance Display Engine (second gen), which controls color bit depth, HDR and connections. The compute units are supposedly more efficient and AMD says the media encoding is better. Like everyone, AMD's Adrenalin application offers some look-what-AI-can-do apps, including image generation, a chatbot to answer questions about AMD's products and more.
The other "4" belongs to FidelityFX Super Resolution, which adds ML-powered upscaling -- that's what Nvidia DLSS has been doing for a few years -- specifically for RDNA 4 GPUs. It's targeting 4K high quality or high performance performance gameplay, and includes all the features currently FFXSR 3.1.
Ryzen AI Max and Max Plus
The Max series are steroidal expansions to the line, flagship APUs for larger laptops that don't include a discrete GPU -- primarily for working with larger AI models and some gaming. They've got a new memory interface that increases memory bandwidth and a ton more graphics compute units, needed by the larger AI models and heavy graphics and video work. In fact, AMD says it's faster than Apple's 14-core version of the M4 Pro.
AMD has renamed the higher-power integrated GPU, designating it the 8060S; it's fundamentally different than the M series GPUs because of its unified memory architecture (which, like Apple's chips, lets all the processing units share the total system memory).
Ryzen AI 300 series specs
8 (4/4) | 5 | 16 | 50 TOPS | 15W-54W | 860M | 8 | Q2 2025 |
6 (3/3) | 4.8 | 16 | 50 TOPS | 15W-54W | 840M | 4 | Q2 2025 |
8 (8/0) | 5 | 16 | 50 TOPS | 45W-120W | 8060S | 32 | H1 2025 |
12 (12/0) | 5 | 24 | 50 TOPS | 45W-120W | 8060S | 32 | H1 2025 |
16 (16/0) | 5.1 | 32 | 50 TOPS | 45W-120W | 8060S | 40 | H1 2025 |
Ryzen 9 9000HX series
The X3D versions of AMD's CPUs have a large memory cache underneath the processor (3D V-cache), which is closer than usual. This minimizes the frequent trips between the cores and the cache -- frequent trips that generate heat. It helps keep the processor cores cool, which in turn allows it to run faster.
Introducing a mobile version of that architecture in the Ryzen 9 9955HX3D will make it AMDs fastest mobile chip. Because it's expected to rely on a discrete GPU for its graphics processing, the integrated GPU on the HX series is typically low power, something it unfortunately shares with Intel's HX series. But with the discrete GPU and the relatively powerful NPU, laptops with it should be fast for almost any work or gaming.
Ryzen 9 9000HX series specs
16/5.4 | 32 | 50 TOPS | 55W-75W | H1 2025 |
16/5.4 | 32 | 50 TOPS | 55W-75W | H1 2025 |
13/5.2 | 24 | 50 TOPS | 45W-75W | H1 2025 |
The desktop flagship that mirrors this is the Ryzen 9 9950X3D. Because the stacked cache design can result in a small drop in clock frequency on nongaming tasks, AMD tweaked the new CPU by adding a second die, which can run at the maximum frequency, essentially unstacking the 3D V-cache. Thus, AMD proclaims this chip not just the best for gaming, but for content creation as well.
Ryzen 9 X3D specs
16 | 32 | 4.8/5.7 | 170W | H1 2025 |
12 | 24 | 4.4/5.5 | 120W | H1 2025 |
Ryzen Z2 series
These are seemingly minor upgrades, but this processor line is popular for handheld gaming consoles, and there's a new lower-end Z2 Go which hints at perhaps a lower-priced class of these devices, which tend to be pricey (the relatively inexpensive Nintendo Switch uses an Nvidia Tegra processor).
Ryzen Z2 series specs
8/0 | 16 | 5.1 | 15W-35W | 12 | Q1 2025 |
4/0 | 8 | 4.3 | 15W-35W | 12 | Q1 2025 |
3/5 | 16 | 5 | 15W-35W | 16 | Q1 2025 |
For more coverage of CES 2025, check out some of the new CES products that you can actually buy right now, or scroll through the most ridiculous and bizarre CES products of this century.