Strix Halo already being tested
HP set to launch a laptop powered by AMD’s yet unannounced architecture.
The launch of the Strix Point, also known as Ryzen AI 300, is just a day away, but attention is now starting to shift towards the Halo series. Although both processors share names and underlying microarchitectures, they are fundamentally different in design. The Halo series targets the high-end APU segment, currently dominated by Apple’s M-series products.
Like Apple, AMD aims to offer a range of high-performance APUs. The Strix Point series has already demonstrated significant improvements in integrated graphics, reportedly 19-32% faster than its predecessor . This boost convinced several laptop manufacturers to adopt APU-only designs.
However, the Strix Halo represents a more advanced design, featuring up to 16 Zen 5 cores and up to 40 RDNA 3.5 Compute Units . This setup includes 4 additional CPU cores and 24 more CUs for graphics compared to the Strix Point series. While it’s unclear if the 40 CU variant will be available, it is expected to deliver a substantial performance increase.
In addition to the increased cores, the Halo series will feature a larger L3 cache— 32MB per CCD . Unlike the monolithic design of Strix Point, the Halo will utilize up to 8 CCDs, resulting in a maximum L3 cache size of 64MB. The Halo also includes a Memory Access Last Level (MALL) cache, similar to Infinity Cache.
The recent leak on Geekbench reveals specifications for the AMD Strix Halo APU, confirming these cache details through a slide shared by an AMD partner comparing Halo with the Strix Point series.
AMD Strix Halo Specs, Source: Geekbench/HKEPC
As we can see, the leak features an 8-core version, but this isn’t news either, as Halo was mentioned to feature several configurations featuring 6, 8, 12 and 16 cores. And given the leaked specs, this could be any of the four known variants:
| AMD Strix Variants (Speculation) | |||||
|---|---|---|---|---|---|
| VideoCardz.com | Cores Variants | GPU | GPU-Class | Memory (up to) | TDP |
Strix Halo | 16x Zen5 12x Zen5 8x Zen5 ⬅️ | 40x RDNA3.5 CU | RTX 4070 | LPDDR5x-8000 256-bit | ~120W (?) |
Strix Halo | 12x Zen5 8x Zen5 ⬅️ | 32x RDNA3.5 CU | RTX 4060 | LPDDR5x-8000 256-bit | ~70W (?) |
Strix Halo | 8x Zen5 ⬅️ 6x Zen5 | 24x RDNA3.5 CU 32x RDNA3.5 CU (low freq) | RTX 4050 | LPDDR5x-8000 256-bit | ~55W (?) |
Strix Halo (LP) | 8x Zen5 ⬅️ 6x Zen5 | 20x RDNA3.5 CUs | RTX 3050 | LPDDR5x-8000 128-bit | ~45W (?) |
Strix Point | 12 (4x Zen5c, 8x Zen5) | 16x RDNA3.5 CU | GTX 1650 | LPDDR5x-7500 128-bit | ~28W |
In terms of performance, benchmarks from Geekbench 5.4—compatible with versions 5.3 and 5.5—provide some first-hand data. Notebookcheck’s database allows for a preliminary comparison. However, it’s important to note several warnings: power settings are unknown, the test machine likely isn’t using final silicon (which would otherwise display the full SKU name), drivers may require adjustments, and what seems even more important the tests were conducted on a Linux operating system. This means the results are not directly comparable.
The highest single-core score recorded was 2177 points , and the highest multi-core score was 13993 points. This places it at the top of the single-core performance ranking, while the multi-core score is expected to improve with the addition of 8 more cores.
Geekbench 5.5 results (mobile CPUs only), Source: Notebookcheck
The release date for Strix Halo is uncertain, but most rumors suggest it will be released around CES 2025 early next year.
Source: Geekbench , Notebookcheck