Krackan Point 2: AMD is apparently planning another Die for Ryzen AI 300
AMD could soon expand its Ryzen AI 300 CPU series with a Ryzen 5, for which a new low-end die is to be used for the first time.
Notebookplatine mit AMDs Krackan Point
(Image: c't / mue)
Even if the Microsoft Copilot+ label currently only provides end customers with a few really useful AI functions in Windows 11, the associated requirements have far-reaching implications for CPU manufacturers. The label is only awarded if the AI unit (Neural Processing Unit, NPU) integrated in the processor achieves a minimum performance of 40 tops (trillions of operations per second, measured in INT8 data format).
As the label is associated with substantial marketing subsidies, no notebook manufacturer wants to do without it. So the CPU manufacturers have to deliver. This causes rotation, especially in the entry-level market: They cannot simply sell old CPUs under a new name as in the past – new chips with cutting-edge NPUs must be produced.
2 Ă— Strix ...
At AMD, this means that the Ryzen AI 300 processor family (only this is suitable for Copilot+) currently consists of three different chips. It all started a year ago with Strix Point, which has up to four Zen 5 and eight Zen 5c cores. The latter are space-optimized variants of the former, which do not clock quite as high. This makes no difference in notebooks, where CPUs always work at their thermal limit, while the space savings mean smaller and therefore cheaper chip sizes. The integrated graphics unit comprises up to 16 compute units (1024 shaders).
AMD added Strix Halo to its portfolio at the beginning of the year. This is a powerful combination chip with up to 16 Zen 5 cores and 40 compute units (2560 shaders), which can be seen as the x86 counterpart to Apple's M4 Pro. We recently tested the first notebooks equipped with it.
... and 2 Ă— Krackan
Krackan Point is intended for cheaper notebooks. This is a smaller version of Strix Point with four Zen 5 and four Zen 5c cores; AMD has halved the integrated graphics unit to eight compute units. Not all processors use their chips in full expansion: The following table shows AMD's current range of Ryzen AI 300 CPUs.
According to reports, however, this is not the end of the story: AMD is reportedly working on another cut-down, which will simply be called Krackan Point 2: Only two Zen 5 cores would be available there (but still four Zen 5c) and only four compute units.
Geekbench leak
The chip is to be used in a processor called Ryzen AI 5 330, among other things. This has not yet been officially announced, but an entry in the Geekbench database reveals that it is likely to have one Zen 5 and three Zen 5c cores. As AMD always includes the GPU name in the processor name and a Radeon 820M has not yet appeared anywhere, it is reasonable to assume that the graphics unit could be limited to just two GPU cores. There is no mention of the NPU in Geekbench, but without one that is suitable for Copilot+, Krackan Point 2 makes little sense.
In the future, Krackan Point 2 could also power other processor models, such as an even smaller Ryzen AI 3, and even the well-known Ryzen AI 5 340 could be converted (with a slightly different core configuration). However, this might not happen until early 2026: According to industry insiders, Gorgon Point is due then, which will have to last the entire year as the Ryzen AI 400. Technically, however, apart from slight clock speed increases, not much is expected to change; Gorgon Point will be a classic refresh. Zen 6 cores and a graphics unit based on RDNA4 are not on the agenda until a year later with Medusa Point – and in combination with 2-nanometer production.
Videos by heise
Thinking outside the box
AMD's approach is reminiscent of Qualcomm's Snapdragon X processors: The twelve-core processor, codenamed Hamoa, shook up the notebook market in mid-2024 thanks to Copilot+; later, slimmed-down processor models came drop by drop. For entry-level models, there is the Purwa entry-level chip, which only has eight CPU cores and a halved GPU (X1-45 instead of X1-85). But that's not all: Qualcomm is said to be working on an even cheaper chip called Canim (possibly with only four CPU cores). It may come alongside the next Snapdragon X generation, which is due to be unveiled in fall 2025 and is certainly intended for expensive premium devices for now.
Even Apple has recognized that there is something to be gained from affordable notebooks thanks to the AI hype. According to rumors, Cupertino is working on a MacBook (with no other name suffix) that Apple wants to position below the MacBook Air family. However, it will not contain an M4, but a weaker processor. There are rumors of the iPhone chip A18 Pro, which is already powerful enough for everything Apple Intelligence is competing against Copilot+ with. It is possible that the entry-level MacBook will materialize before the end of the year.
On the other hand, there are bigger question marks surrounding the stumbling chip giant Intel. The former industry leader has been struggling for years with home-made problems such as botched production processes, which are now taking massive revenge. Only the special case Lunar Lake (Core Ultra 200V) currently manages Copilot+, but not its sister series Arrow Lake (Core Ultra 200H/U), which is stronger in terms of CPU performance. Some particularly inexpensive Intel notebooks still use processors from the old 12th Core i generation, completely without an NPU. We are curious to see what Intel's next CPU generation Panther Lake (Core Ultra 300) will bring in this respect. This is also on Intel's roadmap for this year.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
(mue)
