Towards the end of last year in December 2023, AMD introduced its Ryzen 8040 series (Hawk Point) mobile APUs and also briefly talked about next-gen Strix Point. Back then, AMD wanted to follow the same naming scheme as Strix Point APUs would have been labelled as "Ryzen 8050 series," though more recent leaks suggested that had changed as the company was seemingly looking at a "Ryzen AI" based moniker.
Today, at Computex 2024, AMD has officially confirmed its new naming scheme with the Ryzen AI 300 series CPUs. It appears the move comes as a way for AMD to separate its AI PC processors from its traditional gaming and productivity lineup, similar to how we have the Ryzen Threadripper series that only caters to the HEDT (high-end desktop) segment. The announcement comes in addition to Zen5-based Ryzen 9000 and Zen 3-based 5000XT series processors for desktops.
If you are wondering why AMD chose '300' series it's because this is the third generation of AMD processors with NPU after Ryzen 7040 (Phoenix) and Ryzen 8040 (Hawk Point).
 |
 |
As you can see in the image above, both the Ryzen AI 9 HX 370 and the Ryzen AI 9 365 come with an NPU rated to deliver up to 50 TOPS (Terra Operations Per Second) of AI throughput. This is the major headlining feature of these new APUs as it meets Microsoft's minimum requirement of 40 NPU TOPS for Copilot+ AI PCs. AMD is touting these as the "World's best processor for Copilot+ PCs". Till now, this was only supported on Qualcomm Snapdragon X chips.
While both the APUs have a similar performing NPU, the HX 370 has more cores, CPU-wise and GPU-wise, than the 365. Out of the 12 CPU cores on the Ryzen AI 9 HX 370, four of them are Zen 5 while eight are Zen 5C and in the case of the Ryzen AI 9 365, the number of Zen 5C cores is six, making it a 10 core processor.
What sets the Ryzen AI HX 370 ahead of the Ryzen AI 365 is the additional graphics performance. Both of these chips will feature the new RDNA 3.5 graphics architecture and the HX 370 will pack 16 CUs (compute units) compared to 12 on the 365. Paired with high-speed memory, the former can be quite a bit faster.
 |
 |
AMD has also shared details on how it was able to achieve the massive jump in AI processing power on its new NPUs. These are based on XDNA 2 architecture and AMD has actually managed to surpass its previous promise where it stated that second-gen XDNA would be three times faster.
AMD says this was achieved thanks to a combination of doubling of power efficiency and quintupling (5X) of AI compute capacity. It is also marrying the best of both worlds from INT8 and FP16 for high performance and high accuracy respectively, referring to the design approach as "Block" FP16.
 |
 |
AMD has also compared the performance of its new 3rd Gen Ryzen AI APU against those from Qualcomm, Apple, and AI, in both AI as well as non-AI tasks. First, we have the comparisons against Intel where AMD boasts about Ryzen AI 9's superiority over Core Ultra:
 |
 |
 |
Up next, we have a couple of charts against Qualcomm Snapdragon X and Apple M3:
 |
 |
AMD promises over 100 Copilot+ PCs will be available from OEM partners like Acer, Asus, HP, Lenovo and MSI, starting in July. You can check out our entire Computex 2024 coverage in these articles here.
3 Comments - Add comment