
Hardware-accelerated GPU scheduling (HAGS) changes how Windows schedules work on a supported graphics processing unit. It can be worth enabling when a game feature requires it, but it is not a guaranteed frame-rate boost. If your PC is stable, keep its current setting unless you have a reason to change it. If you are troubleshooting stutter, recording failures, or crashes, compare both settings under the same conditions.
This guide explains the requirements, how to turn HAGS on or off, and how to judge the result. Compatibility sources were checked on October 13, 2026. The testing method below is a suggested procedure, not a set of measured benchmark results.
Windows traditionally uses a CPU thread to coordinate GPU work. With HAGS, a dedicated GPU scheduling processor handles frequent operations such as context switching and time-slice management. Windows still decides which applications receive priority. Microsoft’s technical explanation describes this division of responsibilities and the original WDDM 2.7 driver requirement.
HAGS does not replace your CPU, add graphics memory, or make an application use the GPU automatically. For that broader distinction, see how GPU acceleration works. Selecting a GPU in an application and changing the operating system’s scheduler solve different problems.
Microsoft introduced the feature with Windows 10 version 2004. Its launch guidance expected little visible performance change for most users, because applications already hid much of the scheduling overhead. That historical guidance does not establish a universal performance result for today’s games, drivers, or creative software.
Make the decision around the software you actually use. A stable gaming setup, a recording workstation, and a machine running overnight renders may need different choices.
| Your situation | Suggested action | What to verify |
|---|---|---|
| A game requires HAGS for DLSS Frame Generation | Enable it if your hardware and driver meet the game’s requirements | Frame Generation availability, responsiveness, and stability |
| Your PC works well with its current setting | Keep that setting, or run a controlled comparison | A repeatable improvement rather than one higher FPS reading |
| OBS or hardware encoding has problems | Test with HAGS disabled | The same recording or stream, including the saved output |
| A creative application becomes unstable after a change | Restore the previous setting and retest | Project completion, output quality, and application logs |
| The switch is missing | Check GPU, driver, and Windows compatibility | Vendor support for your exact configuration |
NVIDIA lists hardware-accelerated GPU scheduling as a Windows requirement for DLSS Frame Generation. This is separate from DLSS Super Resolution. Turning HAGS on does not give an unsupported graphics card access to Frame Generation or add the feature to a game that lacks it.
When comparing HAGS itself, keep Frame Generation off in both test runs. Otherwise, you may compare two different rendering configurations. Then test your preferred Frame Generation setup separately, as the configuration you intend to use.
The OBS troubleshooting guide recommends disabling HAGS if you encounter related performance problems or hardware-encoder failures. This is a troubleshooting step, not proof that every OBS user should disable it.
Test a local recording before an important stream. Include your usual game, overlays, camera, and capture sources. Check both the live preview and the recorded file; a smooth game alone does not establish that the recording is healthy.
Support depends on the GPU, the graphics driver, and the Windows build. A recent Windows Display Driver Model version by itself does not prove that the driver exposes HAGS. The toggle must be available for the configuration you are using.
Use the current supported driver for your exact GPU and operating system. On a laptop or an organization-managed computer, check the manufacturer’s or IT team’s recommended package. Avoid changing drivers and HAGS simultaneously during a comparison: you would lose track of which change affected the result.
Windows 10’s standard support ended on October 14, 2025. Its ability to expose HAGS does not change its support status; consult Microsoft’s lifecycle guidance for your edition and update options.
Save your work before restarting. On Windows 11, begin with Settings → System → Display → Graphics. Depending on the Windows build, the switch may sit under advanced or default graphics settings. Older instructions call this page Change default graphics settings.
On Windows 10, the corresponding route is Settings → System → Display → Graphics settings. Change the HAGS switch and restart. To undo either change, return to the same page, restore the previous position, and restart again.
These instructions describe the Windows setting; you do not need to adjust GPU clock speeds or power limits. If the option is absent, skip registry tweaks intended to force it into view and check compatibility first.
A useful comparison changes one variable and measures the work you care about. Start with a repeatable game benchmark, a fixed gameplay route, a video export, or a saved rendering scene. Keep the same resolution, quality preset, display mode, frame cap, and application version.
Record your CPU and GPU models, driver version, Windows build, and HAGS state. Keep the computer’s power mode and background applications consistent. For laptops, use the same power source in both tests. Allow shader compilation or initial asset loading to finish before collecting results.
For gaming, frame-time spikes may matter more than a small change in average FPS. For video editing or Blender GPU rendering, measure the same export or scene with identical settings. Do not change render engines, sample counts, or denoising options halfway through the test.
GPU usage and CPU load are supporting observations, not the result by themselves. Higher GPU utilization can accompany useful work, but it does not prove lower latency. Likewise, a higher displayed frame rate does not directly measure the delay between your mouse input and the image reaching the screen.
LatencyMon examines a Windows system’s suitability for real-time tasks, particularly audio, using measures such as driver execution times and page faults. It is not a direct mouse-to-display latency benchmark. Choose a measurement tool that matches the claim you want to make.
If the switch is missing, identify the exact GPU and installed driver, then compare them with the vendor’s current support documentation. On systems with multiple graphics processors, identify which GPU runs the application. A remote desktop or virtual machine may expose a different graphics configuration from the physical host.
If problems began after enabling HAGS, restore its previous state and repeat the same workload. If the issue persists, investigate other recent changes, including application updates, drivers, overlays, and graphics settings. A problem occurring after a change is a useful clue, but repeated testing provides stronger evidence of its cause.
For a support report, keep the application version, driver version, reproducible steps, and error messages together. “The same recording fails with HAGS on and succeeds with it off” is more useful than “my GPU is slow.”
HAGS is a Windows graphics scheduling setting. It is not a general switch for Linux GPU servers, model training, or cloud performance. On a remote Windows machine, support depends on the operating system, GPU exposure, and driver configuration. Do not assume that all virtual machines support it, or that virtualization always prevents it.
For remote compute, first establish whether you need a full GPU virtual machine, which operating system your software supports, and how much memory the workload requires. Evaluate Compute with Hivenet against those needs; this Windows feature alone does not establish provider suitability.
Keep HAGS enabled when a required feature depends on it and your applications remain stable. Otherwise, use a controlled comparison to decide. A reliable workflow is a better outcome than chasing a performance increase your own tests cannot reproduce.
Pick one AI, compute, or storage workload and see the difference for yourself. Spin it up in minutes, or let our team map your fastest path to production.