AMD’s latest APU, Strix Halo (Ryzen AI Max+), has finally shed its embargo, allowing for the release of high-resolution die shots that showcase its impressive engineering. The chip boasts 16 cores, 40 Compute Units (CUs), and 128GB of unified memory, thanks to two Zen 5 CCDs and up to eight memory ICs.
The Strix Halo’s design features include a reduced D2D interface between the CCD and I/O die by 2mm, while still supporting 3D V-Cache. The chip also hosts an integrated GPU with 40 CUs developed using RDNA 3.5, taking up roughly 307.58mm² of the 441.72mm² total area.
To address memory bottlenecks, AMD has implemented fast LPDDR5X memory and 32MB MALL/LLC cache. The chip also features an XDNA 2 NPU for 50 TOPS of AI performance and a range of I/O interfaces, including PCIe 4.0, USB 4, and Display Engine support.
The Strix Halo’s adoption is expected to improve as more OEMs adopt the chips, despite initial hesitation. However, pricing remains a concern, with the $2500 option for the 32GB version seeming excessive. AMD hopes to learn from Strix Halo and refine its strategy with future chips like Medusa Halo in a couple of years.
While gamers may be excited about the chip’s potential, it faces stiff competition, particularly when compared to the RTX 4090. Nevertheless, the Strix Halo offers impressive features, including full AVX-512 support and properly-supported native Linux.
Source: https://www.tomshardware.com/pc-components/gpus/amds-game-changing-strix-halo-apu-formerly-ryzen-ai-max-poses-for-new-die-shots