Valve Saves 8GB AMD GPUs: Major Linux Update Changes Game for Gaming

Video RAM (VRAM) management has become the crux of the matter for PC gamers. As video games increasingly demand more resources, graphics cards with 8 GB of memory often find themselves cramped, causing slowdowns and framerate drops. Faced with this observation, Natalie Vock, member of the graphics driver development team at Valve, revealed a welcome optimization for the Linux ecosystem.

The problem of the “leak” to system memory

Until now, the Linux kernel struggled to intelligently prioritize VRAM allocation. When a game saturates the graphics card’s memory, the system indiscriminately moves data to the slower RAM of the PC. This process does not distinguish between the game in the foreground and a simple browser window open in the background. The game experiences slowdowns even though the available VRAM is not fully saturated.

A priority booster for players

To correct this priority defect, Natalie Vock has developed new patches for the Linux kernel accompanied by two specific utilities. The heart of this solution is based on dmemcg-booster. This tool allows you to favor the active program by giving it absolute priority over VRAM. If the memory reaches saturation, it is now the background tasks which are expelled to the system RAM, leaving the field free for the game to use every megabyte of VRAM.
The tests carried out on Cyberpunk 2077 with an 8 GB GPU are clear. Before the patch was applied, the game only consumed 6 GB of VRAM and dumped 1.37 GB of data into RAM. Once the “booster” was activated, VRAM occupancy rose to 7.4 GB while transfers to RAM fell by more than half, providing significantly increased fluidity.

A gradual integration

A second tool, called plasma-foreground-booster, was also designed for the KDE desktop environment in order to automatically identify the window in the foreground and automate this prioritization.
Although these fixes are already being integrated into the CachyOS distribution, their final integration into the mainline Linux kernel is still pending. However, a major restriction remains: this advancement only concerns users of AMD graphics cards. Nvidia drivers, due to their proprietary and closed memory management, cannot yet benefit from this software optimization.