SIXUNITED has completed the integrated hardware and software adaptation of Intel's variable memory technology, allowing the 30B large model to easily fit into a "pocket"!


In today's era where generative AI has fully entered the edge computing stage, localized operational capability has become the core competitiveness of AIPC. However, due to the technical limitation of fixed VRAM allocation, traditional integrated graphics platforms have always struggled to break through performance bottlenecks. As a pioneer leader in the AIPC field, Intel has taken the lead in launching Variable VRAM Technology, which, through a dynamic scheduling mechanism, breaks the limitation of fixed VRAM allocation in integrated graphics, achieving on-demand adjustment and efficient utilization of VRAM.

As Intel's core partner and full-stack AI service provider, SIXUNITED promptly completed the verification of Intel's Variable VRAM Technology on the Arrow Lake-H series AI PCs, including the MTN25 series 1kg lightweight laptops and MTN79 series flagship performance laptops, and successfully adapted Alibaba Tongyi Qianwen 30B (30 billion parameters) MoE large models—Qwen3-30B-A3B-Instruct-2507 and Qwen3-Coder-30B-A3B-Instruct. This breakthrough allows users to smoothly run ultra-large-scale AI models on lightweight and portable laptops without relying on high-performance discrete graphics or cloud computing power, enabling AI to truly integrate into every scenario of daily office work, study, and creation.

 

Variable VRAM Technology

Breaking bottlenecks to make local AI more powerful

For a long time, the fixed VRAM allocation mode of traditional integrated graphics architecture has limited the ability to run large models locally. Intel's Variable VRAM Technology completely breaks this limitation through a deeply optimized reserved memory dynamic scheduling mechanism. On SIXUNITED's AIPC equipped with Intel® Core™ Ultra 200H processor (Arrow Lake-H) and 32GB memory, this technology can increase the VRAM allocation ratio from 57% to 87%, raising the available VRAM from the industry common 18GB to 27GB, bringing nearly 50% performance improvement.


Especially for the MoE (Mixture of Experts) architecture, which features VRAM fluctuations during expert node switching, Intel's Variable VRAM Technology performs precise software-level tuning and, combined with the hardware characteristics of the Core Ultra 200H platform, deeply optimizes the read/write path of reserved memory. This enhances memory bandwidth utilization, reduces data exchange latency, and ensures that large models run stably and efficiently even under high VRAM usage.

 

Two major product series

Easily "fit" 30B large models into your pocket

SIXUNITED took the lead in completing the adaptation of Intel's Variable VRAM Technology on the 1kg MTN25 series lightweight laptops and MTN79 series flagship performance laptops. Whether for lightweight mobile scenarios or high-load professional creative tasks, SIXUNITED's Arrow Lake-H series can provide stable, smooth, and durable computing power support for running 30B-level large models locally, making AI truly "portable and efficient anytime."

 

01 MTN25 Series: 1kg lightweight portability, the first choice for on-the-go AI

The entire MTN25 series weighs about 1kg, with the representative model MTN25-140M2-YDX only 15.95mm thick, featuring a magnesium alloy body, a 14-inch 16:10 golden ratio screen with a resolution up to 1920×1200, paired with a large capacity 50Wh/60Wh battery for long-lasting endurance and worry-free use. The interfaces support TBT4.0, USB 3.2, HDMI, audio jack, etc., balancing high performance and portability, making it an ideal choice for on-the-go AI computing.

 

02 MTN79 Series: Flagship performance, full-speed AI tool

The flagship MTN79 series covers multiple sizes from 14 to 17.3 inches, supports customization of full metal, semi-metal, and full plastic appearances, and achieves a dual breakthrough in performance and cost with 45W power consumption. The flagship MTN79-140M-HK is equipped with a 14-inch 16:10 OLED screen, a large 60Wh battery, and dual DDR and dual SSD expansions, easily handling long-term high-load model inference and multitasking, making it an efficient tool for developers, designers, and AI engineers.

 

Integrated hardware and software

Adaptation enables 30B large models to run on the go

Hardware breakthroughs can only release true value through deep integration with the software ecosystem. SIXUNITED's self-developed AI assistant has simultaneously completed the adaptation of Intel's Variable VRAM Technology and deeply integrated two 30B large models, Qwen3-30B-A3B-Q4_K_M and Qwen3-30B-A3B-Instruct-2507-Q4_K_M, supporting high-precision quantized versions such as Q4, Q5, and Q6. Enterprise and individual users can achieve millisecond-level response for document summarization, code generation, creative writing, text-to-image, and other local AI interactions in offline or secure intranet environments through SIXUNITED's MTN25 and MTN79 series AIPC, significantly reducing cloud invocation costs and data security risks.

▲ Qwen3-30B-A3B-Q4_K_M running status

▲ Qwen3-30B-A3B-Instruct-2507-Q4_K_M running status

 

The launch of Intel's Variable VRAM Technology opens new performance space for local large model operation on AIPC. As one of the first ODM/OEM service providers to complete verification and application, SIXUNITED not only proves that lightweight and portable devices can carry ultra-large-scale AI models but also sets a benchmark case for integrated hardware and software optimization in the industry. In the future, SIXUNITED will continue to work with Intel and ecosystem partners to promote the popularization and application of larger-scale, higher-precision, and lower-power large models locally, achieving full-chain optimization from hardware, system, drivers to AI applications, making AI truly an accessible productivity tool for everyone.