Discord, Steam, Telegram hoard up to 1GB VRAM via hardware acceleration

A Reddit user on Linux with an Nvidia GPU found that several everyday applications are silently reserving significant VRAM even when minimized. Discord is the worst offender, hoarding 450MB of GPU memory due to its hardware-accelerated rendering. Steam takes 200MB, Telegram 150MB, and a handful of other apps push the total well over 1GB. The issue stems from web-based interfaces that use hardware acceleration—they construct frames in VRAM and never fully release the memory when the window is hidden. This poses a serious problem for AI developers and power users who need every megabyte of VRAM for local LLM inference, 3D rendering, or machine learning tasks.

The fix is simple but comes with trade-offs: users can either close those apps entirely or disable hardware acceleration in their settings. Disabling it, however, causes noticeable UI stuttering and sluggishness. A more practical workaround is to maintain a separate browser with hardware acceleration turned off, used exclusively when running VRAM-intensive workloads. On Linux systems, the `nvidia-smi` command provides a real-time list of all processes consuming VRAM, making it easy to identify and kill these memory hogs. For professionals squeezing every last byte of VRAM out of their GPUs, this hidden overhead can mean the difference between a model fitting in memory or crashing with an out-of-resource error.