CES

Ryzen AI Max: AMD competes against Apple's M4 Pro with a thick integrated GPU

The SoCs of the Ryzen AI Max processor family raise the performance of integrated graphics units to a level that Windows devices have never seen before.

listen Print view

(Image: AMD)

6 min. read

In the Windows world, the term "integrated graphics unit" has for years been associated with the connotation that 3D games only run with limited graphics settings if at all. This is due to the fact that until now, there have only been comparatively weak variants, so that an additional GPU was required for high 3D performance. However, this is not due to the design per se: Apple's Pro, Max and Ultra mobile processors contain powerful integrated GPUs, and games consoles such as Playstation 5 and Xbox Series X also have powerful graphics units that are integrated into the respective system-on-chip (SoC).

AMD is now launching the first Ryzen SoC at CES 2025, which is heading in the same direction: the series developed under the codename Strix Halo can be found on the market as Ryzen AI Max. The package combines two CPU chiplets with eight Zen 5 cores each with an SoC chiplet containing 40 compute units (CUs) or 2560 shader cores of the RDNA 3.5 graphics architecture. To feed them with data, Strix Halo has a memory interface that is twice as wide as usual, namely 256 bits ("quad channel"). Naturally, the CPU cores also make use of this. For comparison: the normal Ryzen AI 300 is content with 1024 shaders.

What Strix Halo actually brings to the ground is still unclear. No concrete benchmark results were included in previously distributed material, only relative percentages with changing counterparts: AMD had chosen Intel's (clearly weaker) Core Ultra 200V for 3D benchmarks and Apple's M4 Pro for CPU benchmarks. Unfortunately, there was no trace of CPU comparisons with more powerful Intel CPUs or 3D bar charts against Nvidia GPUs such as GeForce RTX 4050/4060. In practice, however, it will probably depend on what the notebook manufacturers make of it anyway. In any case, AMD generally allows a wide TDP range of 45 to 120 watts, which has a massive impact on performance.

The first devices with Ryzen AI Max come from HP and Asus.

(Image: AMD)

One of the first devices with Ryzen AI Max will be the ROG Flow Z13 from Asus, a tablet in the style of the Surface Pro. Asus has already equipped the 2022 predecessor with a 14-core CPU and GeForce RTX 4070, but the latter was run at very low speeds: The GPU delivered noticeably less 3D performance there than the same chip in gaming notebooks with expansive cooling systems. HP is only planning Ryzen AI Max for workstations, but wants to launch two Strix Halo devices with the compact ZBook Ultra G1a notebook and the Z2 Mini G1a mini PC. We do not yet have any details on local prices or specific equipment variants. AMD has only given a rough availability range of between the first and second quarter of 2025.

Like Ryzen AI 300 (Strix Point), Ryzen AI Max (Strix Halo) contains an AI accelerator (Neural Processing Unit, NPU) that fulfills Microsoft's requirement for the Copilot+ marketing label and the associated AI functions of Windows 11. However, the latter devices, which were announced six months ago, have so far been scarcely available and costly (and Strix Halo will undoubtedly be more expensive).

Model overview of the Ryzen AI Max.

(Image: AMD)

The mainstream variant Krackan Point (previously referred to as Kraken Point in the rumor mill) is now being launched at CES, which should enable more affordable notebooks. Like Strix Point, the chips run in the Ryzen AI 300 family –, but no longer only as Ryzen AI 9 with up to 12 cores, but as eight-core Ryzen AI 7 350 and six-core Ryzen AI 5 340.

Model overview of the smaller Ryzen AI 300.

(Image: AMD)

We assume that the availability of notebooks with Ryzen AI 300 will now improve overall: Strix Point and Krackan Point share the same platform (FP8). However, this was previously limited: This is because all previous Strix Point devices have soldered-on LPDDR5 memory; only with Krackan Point is the firmware ready to accept plugged-in DDR5 SO-DIMMs. Some notebook manufacturers had to wait for this and can only now make their devices with replaceable RAM ready for the market – even if they then use a Strix Point twelve-core processor that is actually already half a year old.

Videos by heise

Below the CPUs that are suitable for Copilot+ thanks to the powerful NPU, AMD previously had Ryzen processors with 8000 model numbers in its portfolio. These will mainly be given new names in the 2025 refresh: From now on, they will run as Ryzen 200. The name suffix "AI" is missing as the older NPU is not suitable for Copilot+ and, moreover, it is deactivated in some Ryzen 3 and Ryzen 5 variants.

AMD's Z2 processor family is used in gaming handhelds.

(Image: AMD)

Another offshoot of this older chip generation is the Z2 family, which is intended for the comparatively new product category of gaming handhelds. Specifically, AMD is promising new editions of Asus' ROG Ally, Lenovo's Legion Go and Valve's Steam Deck. What is new is that there are now three chip variants compared to the Z1 generation: Z2 Extreme as a full expansion, Z2 with a reduced number of graphics cores and new Z2 Go with CPU additionally halved to four cores.

Overview of the new Ryzen 200.

(Image: AMD)

The announcement will be made as part of the CES 2025 tech trade fair. Heise Medien is an official media partner of CES 2025.

(mue)

Don't miss any news – follow us on Facebook, LinkedIn or Mastodon.

This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.