Valve hack prioritizes 8GB GPUs on Linux
A Valve engineer posted a VRAM scheduling hack for Linux that prioritizes 8GB GPUs for gaming workloads while deprioritizing background tasks. The change aims to improve gaming performance on constrained GPUs and has implications for system-level resource management in development environments (Tom's Hardware on X).
Video memory is the graphics card’s fast local storage, and on an 8 gigabyte card it can fill before a modern game finishes loading textures. A Valve engineer has posted Linux changes that try to keep that scarce memory reserved for the game in front of you, not the browser or chat app behind it. (phoronix.com) Natalie Vock, who works on Valve’s Linux graphics driver stack and the Radeon Vulkan driver, sent a six-patch series to Linux graphics and control-group maintainers on March 13, 2026. The cover note said the work targets cases where device-memory protection fails and allocations spill into Graphics Translation Table memory, which is slower system memory mapped for the graphics card. (lore.gitlab.freedesktop.org) The kernel side adds device-memory control-group queries and changes in the Translation Table Maps memory manager, the Linux subsystem that decides where graphics buffers live and what gets evicted under pressure. In plain terms, the patches separate memory charging from allocation and adjust eviction behavior so protected foreground work is less likely to lose video memory first. (lore.gitlab.freedesktop.org) Two user-space tools make the policy usable on a desktop. Phoronix reported that `dmemcg-booster` runs as a system service to manage device-memory limits, while `plasma-foreground-booster` tells KDE Plasma to prioritize the full-screen foreground app. (phoronix.com) That setup is aimed at a specific failure mode on Linux gaming systems with 8 gigabytes of video memory or less: when memory fills up, the operating system can shove active game data out of local video memory while background programs keep their share. Digital Foundry said the change reduces stutter by making background tasks yield first. (digitalfoundry.net) The early rollout is narrow. Digital Foundry and XDA both said the easiest way to try it now is on CachyOS with KDE Plasma by installing `dmemcg-booster` and `plasma-foreground-booster`, and newer versions of Gamescope can provide similar foreground signaling outside Plasma. (digitalfoundry.net) (xda-developers.com) Support is not universal across graphics vendors. Reporting on the package rollout said the current work is for Advanced Micro Devices and Intel paths, with no Nvidia support in the initial implementation. (msn.com) The patches also do not create extra memory. Digital Foundry said games that exceed the card’s physical limit can still run into slowdowns, but the new policy changes which process gets pushed out first when the system is under pressure. (digitalfoundry.net) For Linux desktop developers, the thread running through this work is broader than games. It uses the kernel’s control-group machinery to treat graphics memory more like a managed system resource, with the foreground app getting first claim when an 8 gigabyte card runs out of room. (lore.gitlab.freedesktop.org)
