Intel has introduced a new driver feature called Shared GPU Memory Override, which allows integrated Arc GPUs on select Core Ultra systems to reserve a significantly larger portion of system RAM for video memory. The control appears in Intel Graphics Software as a slider that increases the percentage of system memory available to the iGPU. Early release notes and driver listings indicate default values near 57%, while vendor examples suggest that the control can allow significantly higher allocations on high-memory laptops, at 87% on Intel's demo. Intel positions the capability as a way to close the gap between integrated and discrete solutions for memory-bound graphics workloads and to give developers and power users greater flexibility for non-gaming tasks such as running local AI models.

Intel's implementation mirrors AMD's Variable Graphics Memory feature for Ryzen AI platforms, which similarly reallocates system RAM to the GPU and has been demonstrated alongside driver-level enhancements, such as AFMF, to increase frame rates in specific titles. Both approaches exploit unified memory architectures to enlarge the effective VRAM pool; however, this is not a universal performance win. Some games will opportunistically load larger assets when more VRAM is available, which can blunt or even reverse the gains. Users must install Intel's latest driver and reboot to enable the override. Minimum RAM and platform requirements also apply. OEMs may expose similar options in BIOS, and manufacturers will publish guidance. However, for workloads that are always VRAM-limited, such as local LLMs, it is essential to have an option to extend the iGPU's memory capacity, allowing larger models to run without an internet connection.

More...