Ryzen AI 300: AMD shoots against Apple, Intel and Qualcomm
AMD's Zen 5 processors for notebooks are supposed to beat the competition. However, the CPU names are questionable.
![](https://heise.cloudimg.io/width/610/q85.png-lossy-85.webp-lossy-85.foil1/_www-heise-de_/imgs/18/4/6/0/4/0/4/9/Ryzen_AI_300_4-e75f2d1cb6ff7a70.jpeg)
AMD is also building the Ryzen AI 300 monolithically again.
(Image: AMD)
AMD presents its first two notebook processors with Zen 5 architecture: the Ryzen AI 9 HX 370 and Ryzen AI 9 365 (codenamed Strix Point). They are the first high-end models from AMD to be designed as hybrid CPUs with different cores. The first notebooks with these processors should be available from July 2024.
The top model Ryzen AI 9 HX 370 combines four Zen 5 cores with eight Zen 5c cores; the Ryzen AI 9 365 has 4 + 6. AMD has so far remained silent on the differences – the c variants probably have less cache again and do not achieve such high clock frequencies.
AMD has built the Zen 5 core wider compared to Zen 4 to increase throughput. The first two cache stages have been accelerated and the jump prediction should also work more accurately. On average, this should result in 16 percent more performance per clock step (IPC) – just like the Ryzen 9000 for desktop PCs.
Faster CPU cores despite clock stagnation
Thanks to this IPC increase, the new processors are also faster than their predecessors in single-threaded applications, although the maximum boost clock is minimally reduced. The Ryzen AI 9 HX 370, for example, achieves up to 5.1 GHz, which is 100 MHz less than the previous top model, the Ryzen 9 8945HS. The newcomers should generally be faster in multithreaded tasks thanks to the up to four additional CPU cores.
AMD has increased the integrated graphics unit from 12 to 16 compute units (CUs). This increases the number of shader cores from 768 to 1024. AMD's naming RDNA3.5 instead of RDNA3 implies architectural improvements, although the company is not commenting on this. The clock frequency increases minimally from 2.8 to 2.9 GHz.
The full expansion with 16 CUs is initially only available in the Ryzen AI 9 HX 370. 12 CUs are active in the Ryzen AI 9 365. AMD has not yet announced any other models, but they are likely to follow soon.
(Image: AMD)
AI accelerator theoretically ahead of Apple, Intel and Qualcomm
The biggest leap in performance comes from the integrated AI unit, which can perform 50 trillion operations per second (50 TOPS). AMD advertised the previous Ryzen 8040 with up to 16 TOPS. This means that the new processors are fast enough to run Microsoft Copilot locally. Consequently, notebook manufacturers can market their devices as Copilot+.
Normally, companies base their TOPS figures on simple eight-bit integer calculations (INT8), which are sufficient for many AI algorithms. However, the XDNA2 accelerator in the Ryzen AI 300 can also handle the relatively new Block Floating Point (BFP) data format for floating-point calculations with 50 teraflops. It is designed to combine the speed of INT8 with the precision of FP16.
While conventional floating-point types represent each number individually with a significand (mantissa), an exponent and a sign, Block Floating-Point combines several values and uses a common exponent for the entire block of values. AMD also saves chip space with this approach.
In AI applications, the Ryzen AI 300 is said to be faster than all other mobile processors. Specifically, AMD mentions Apple's M4, Qualcomm's Snapdragon X Elite and Intel's Core Ultra 100 + 200. So far, however, AMD has mainly lacked software support. The announcement still did not mention any AI interface of its own –, which means that AMD would still have to fall back on Microsoft's rather poorly optimized DirectML under Windows 11. It is unknown whether this will support Block Floating Point in the near future.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmmung 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.