In Alex Ziskind’s YouTube video titled “This Laptop Runs LLMs Better Than Most Desktops,” published on May 27, 2025, the spotlight is on the Asus Flow Z13 (2025 version). Ziskind captivates the audience by portraying the advanced capabilities of this laptop equipped with an AMD Ryzen AI Max Plus 395 APU, capable of running a 110 billion parameter AI model—a feat even powerful desktops like those equipped with an Nvidia 5090 struggle with. At the heart of this performance is the device’s 128GB of unified memory that facilitates impressive local AI model execution. Ziskind effectively illustrates this breakthrough by showing how larger models—such as the Llama 370 billion—maintain complex reasoning better than smaller models, thereby minimizing errors known as “hallucinations.”

Ziskind adeptly conveys the superiority of larger memory systems in achieving higher scores for speed and efficiency, notably pointing out that using a system with full access to all unified memory heralds better performance—an impressive 51 tokens per second for the Gemma 34 billion model. Although the massive memory capabilities are praised, the segment addressing the limitations and challenges highlights that the GPU’s usage might not always meet expectations due to memory allocation constraints. The video articulates that the system must reboot to adjust memory settings, a process that could feel cumbersome for users often swapping models or configurations.

While larger memory capacity is attractive, the video mentions crucial downsides—specifically system stability issues and limited GPU utilization on the Asus device despite its theoretical capability. Comparingly, Apple’s approach, allowing dynamic memory sharing between CPU and GPU, is noted to handle complex models more fluidly. The critique Ziskind provides acknowledges discrepancies in expected versus actual performance, particularly when comparing defined capacities against real-world task executions.

In closing, Ziskind emphasizes that while the Asus Flow Z13 represents a significant leap in portable AI processing capabilities, potential buyers should remain vigilant about actual performance and setup constraints relative to advertised specs, akin to automatic vs manual transmission preferences in cars. Such nuanced considerations are vital for those increasingly reliant on portable high-performance computing.

Alex Ziskind
Not Applicable
November 24, 2025
Result charts for this vid
PT20M12S