AMD's new Ryzen AI MAX+ 395 "Strix Halo" APU has exhibited over three times the performance of the NVIDIA RTX 5080 in the DeepSeek R1 AI benchmark, marking a significant breakthrough in AI processing for consumer laptops. Its robust hardware configuration and innovative memory design not only excel in AI workloads but also unlock unprecedented possibilities for high-end thin and light devices.

This APU incorporates 16 Zen 5 architecture CPU cores and 32 threads, reaching speeds up to 5.1 GHz. Alongside it is a Radeon 8060S integrated GPU equipped with 40 RDNA 3.5 compute units and an XDNA 2 Neural Processing Unit (NPU) delivering 50 TOPS of computational power. Perhaps its most impressive feature is the substantial 128GB of unified memory, far surpassing the 32GB common among competitors. With Variable Graphics Memory technology, users can allocate up to 96GB as VRAM, providing ample support for large AI model operations. In contrast, the RTX 5080 offers just 16GB of VRAM, and the RTX 5090 only 32GB, giving the Strix Halo a distinct advantage in memory-heavy tasks.

During AI performance evaluations, the Ryzen AI MAX+ 395 displayed remarkable strength. Using LM Studio as an example—an application powered by llama.cpp that allows ordinary users to run the latest language models locally—the Strix Halo's 128GB memory pool delivers up to 3.05x higher performance than the RTX 5080 when model sizes exceed the 16GB VRAM limit. Even compared to the RTX 5090, although it has greater computing capabilities, it also consumes significantly more power: the 5080 and 5090 require 360W and 575W respectively, whereas the Strix Halo operates between 45W and 120W, marking a much more efficient alternative.

Specifically in AI workloads, this APU eclipses its competition. Compared to Intel Arc 140V, its token throughput is boosted by 2.2x. In small models like Llama 3.2 3b Instruct, the first token generation is 4x faster; in 7-8B parameter models, speed increases to 9.1x; and for 14B parameter models, it is 12.2x faster than Intel Core Ultra 258V. For 14B parameter models, it is 12.2 times faster than Intel Core Ultra 258V. In vision models, performance is equally strong: IBM Granite Vision 3.2 3b is up to 7x faster, while Google Gemma 3 4b and 12b are 4.6x and 6x faster, respectively. Additionally, the Strix Halo can effortlessly handle large models like Google Gemma 3 27B Vision, a challenge for many other APUs.

Beyond AI capability, the Strix Halo's gaming performance is also commendable. The Radeon 8060S integrated GPU nears the performance of the NVIDIA RTX 4070 Mobile Edition in 3DMark Time Spy tests, achieving 39.4 fps and 36.8 fps in games such as Cyberpunk 2077 and Stars Hollow at the highest 1600p image quality setting, slightly surpassing the RTX 4070's 37.3 fps and 34.7 fps. While it slightly lags in certain games like Wild Dart 2, its low power consumption and integrative design still distinguish it. Unlike traditional discrete GPUs that demand more power and additional cooling, Strix Halo strikes a balance between performance and efficiency in a lightweight notebook.

AMD is positioning this APU to challenge NVIDIA and Intel's high-end offerings. At CES 2025, AMD claimed that its AI performance is 2.2x higher than the RTX 4090 while utilizing only 13% of its power, which is impressively efficient. Compared to the Apple M4 Pro, its score in the Cinebench 2024 multi-core test was close (1669 vs. 1715) and it was 86% ahead in V-Ray rendering. Against Intel's latest Lunar Lake processor, the Core Ultra 9 288V, the Strix Halo averages 1.4x better 3DMark graphics performance and 2.6x faster rendering results.

In terms of hardware design, Strix Halo's 441 square millimeter chip area hosts powerful computing capabilities and an innovative memory architecture. Its unified memory technology eliminates the traditional separation between CPU and GPU memory, offering data transfer speeds up to 273GB/s, which optimizes AI and graphic processing support. This architecture is particularly suited for local AI tasks requiring substantial VRAM, such as operating Llama models with 70B parameters, key to outperforming the RTX 4090.

Currently, Strix Halo is featured in several devices, including high-end thin and light laptops like the ASUS ROG Flow Z13 and HP ZBook Ultra 14 G1a, along with GMKtec's EVO-X2 mini-PC, showcasing its potential in portability and performance, especially for complex tasks, despite not needing a discrete graphics card. More devices powered by this APU are anticipated to enter the market in the second quarter of 2025.

The Ryzen AI MAX+ 395 "Strix Halo" redefines high-end APU standards with superior AI performance, powerful integrated graphics, and an innovative memory design. It enables tech enthusiasts to run complex models locally and opens new horizons for thin and light devices in gaming and professional applications. As more test data and real-world application scenarios emerge, this processor will undoubtedly be a focal point in the tech landscape of 2025.