# Mesa adds YOLOX support for NPUs via PyTorch

Digest AI · Hardware & Compute · published 2026-10-14T01:07:58Z

Canonical: https://digestai.news/story/mesa-adds-yolox-support-for-npus-via-pytorch

## Summary

Open-source developer Tomeu Vizoso has merged code into the Mesa driver stack to support the YOLOX object detection model on Neural Processing Units (NPUs). This update utilizes Torx, Mesa's hardware backend for compiling and executing machine learning models with PyTorch ExecuTorch. The change allows users to run YOLOX using NPU drivers within Mesa, expanding the utility of the driver stack beyond its traditional GPU focus.

The implementation has been tested with the Arm EthosU NPU on the NXP i.MX93 SoC. While the Etnaviv driver for Vivante NPUs reportedly supports all necessary operations, it still lacks the infrastructure work required to function with Torx. This development is part of Mesa 26.3 and represents a continued effort by Vizoso to advance NPU capabilities in the open-source ecosystem.

## Key points

- Tomeu Vizoso merged YOLOX support into Mesa's Torx backend for PyTorch ExecuTorch.
- The update enables running YOLOX on NPUs, tested with Arm EthosU and NXP i.MX93 SoC.
- Etnaviv driver for Vivante NPUs supports operations but lacks Torx infrastructure work.

## Why it matters

This update broadens the scope of Mesa from GPU drivers to NPU support, enabling efficient on-device AI inference for object detection models like YOLOX in open-source environments.

## Sources

1. [Mesa expands NPU capabilities with support for YOLOX with PyTorch](https://phoronix.com/news/Mesa-Torx-YOLOX-PyTorch) (phoronix.com, 2026-10-14)

## Cite

Digest AI, "Mesa adds YOLOX support for NPUs via PyTorch", 14 October 2026, https://digestai.news/story/mesa-adds-yolox-support-for-npus-via-pytorch

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/mesa-adds-yolox-support-for-npus-via-pytorch.json