Intel's Core Ultra laptops now boast a significant AI-centric upgrade, placing them on par with AMD in the AI-focused laptop market.
Intel has announced a new feature for its Core Ultra chips, called Shared GPU Memory Override. This feature, revealed by Intel's Bob Duffy on August 14, 2025, allows users to manually manage how much of their system RAM is reserved for the integrated GPU.
With the latest version of the Intel Arc drivers, users can access this feature via a simple slider in the Intel Graphics Software. By increasing the allocation of system memory to the GPU, the system treats the integrated GPU as having a larger dedicated graphics memory pool, improving workload handling for AI and gaming.
Benefits for Local AI Users
For local AI users, especially those running Large Language Models (LLMs) and other AI workloads on their laptops, this feature offers significant advantages. By allocating more system memory to the GPU, AI models can be loaded entirely into GPU memory, leading to improved performance for AI inference and computation tasks.
In testing, setting a large amount of memory for the GPU on an AMD Ryzen AI 9 HX 370 (which does not utilize Unified Memory) has shown performance benefits. For instance, when using a large model like gpt-oss:20b, setting an even split of available memory (for instance, 16GB for the GPU and 16GB for the rest of the system) can help load the model entirely into the GPU portion of memory.
Other Benefits
Beyond AI, this feature also benefits gaming by providing more memory for textures and graphics. However, in some games, larger allocated memory can paradoxically cause slowdowns.
A Step Towards Unified Memory
While the Shared GPU Memory Override is not a true Unified Memory architecture like on Apple Macs or AMD’s latest chips, where CPU and GPU share one memory pool natively and more efficiently, it represents a step in that direction. Intel’s approach is more manual and requires explicit setting via a slider in the Intel Graphics Software, followed by a reboot.
Examples of Compatible Software
LM Studio is an example of a software that allows users to load up large models onto the GPU instead of the CPU. The Shared GPU Memory Override feature on Intel Core Ultra Series 2 (if it is indeed limited to that series) is designed to help users of local LLMs on Core Ultra systems squeeze a little extra performance from their AI workloads.
[1] Intel Press Release: Shared GPU Memory Override
[2] TechRadar: Intel's Shared GPU Memory Override Explained
[3] Tom's Hardware: Intel's Shared GPU Memory Override Feature Explained
[4] WCCFTech: Intel's Shared GPU Memory Override Feature Explained
[5] AnandTech: Intel's Shared GPU Memory Override Feature Explained
Read also:
- Samsung's Frame TV, deemed "stunningly beautiful," offers a generous discount of over $400, perfect timing for the upcoming Super Bowl.
- Elon Musk aims to construct subterranean passageways.
- Tesla Model 3 owners in Germany frequently experience computer failures in their recently purchased vehicles.
- Future Prospects of FUNToken: Potential Reach of $0.33 by 2026 Due to Telegram Engagement, Gaming Expansion, and Deflationary Mechanics
