Google is opening up a tool that helps developers make better use of the GPU for AI tasks.
AI-generated concept illustration / Mais FPS. Not an actual capture.
Google open-sourced ML Drift on October 8. The tool organizes AI execution on the GPU and can now be studied and adapted by other developers. The project is on GitHub under the Apache 2.0 license.
According to Google, Photos and YouTube Shorts already use ML Drift. Improving an image or applying a video effect are uses where reducing processing time means less waiting after a click.
One optimization described in the project’s research paper is memory reuse. During calculations, the application produces temporary results. Once one is no longer needed, its space can be used for another. This avoids reserving separate memory for data that does not need to exist at the same time. ML Drift handles that organization during execution.
PC results are still preliminary
Google presented its PC results as a preview. Performance will depend on the task and how each application integrates the tool.
The published tests concern AI tasks and do not support predictions of higher FPS in games.
Sources
- Google Developers Blog — ML Drift announcement (2026-10-08)
- Google AI Edge — ML Drift repository (Accessed 2026-10-13)
- Scaling On-Device GPU Inference for Large Generative Models (2025-05-01)
