Windows could soon let you control how your PC splits memory for gaming and AI
A recent experimental Windows 11 build hides evidence that Microsoft is preparing to let users control the split between system and video memory in unified pools. The feature could give users more flexibility when juggling gaming and AI workloads, especially on upcoming unified-memory devices like Nvidia's RTX Spark platform. Read Entire Article
Microsoft is reportedly preparing to introduce a new feature in Windows 11 that would allow users to control how memory is split between gaming and AI tasks on PCs without a dedicated GPU. This could prove crucial as Apple, Qualcomm, Nvidia, and other manufacturers develop AI PCs with unified memory pools that could stretch into the hundreds of gigabytes.
The experimental feature, named "IntelligentCarveout," was discovered in Windows 11 build 29648.1000, which was released on August 17. While it has not been mentioned in Microsoft's patch notes, code strings within the build indicate that it allows for reserving memory for accelerators, graphics, and AI. This would give users the ability to manually set aside unified memory for gaming and AI-intensive applications, effectively excluding it from other system tasks.
If implemented, this feature could significantly alter how performance is configured on unified-memory chips. Currently, unified memory is prevalent in Macs, some laptops, handheld gaming PCs, and other systems with integrated graphics. Although these systems may experience varying performance based on the workload, they generally outperform discrete GPUs with dedicated VRAM.
However, GPU-heavy tasks often draw from system memory, leading to a notable performance penalty. With the rise of high-end GPUs used for AI work, there have been instances where systems with 24GB or more of unified RAM have encountered memory shortages. To address this issue, Microsoft and Nvidia have introduced RTX Spark devices, capable of housing up to 128GB of unified memory.
Nvidia's RTX Spark chip boasts more addressable memory than any of its GPUs, making it an appealing option when combined with the "IntelligentCarveout" feature. This capability could potentially enable the RTX Spark chip to access more memory than even the RTX 5090's 32GB of dedicated VRAM, though it's important to note that the shared, LPDDR5x-based pool is notably slower than the RTX 5090's GDDR7.
Written by urgent.news from TechSpot's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
