Fabless chipmaker MemryX AI introduced its MX3+ accelerator, promising up to ten times greater processing power and memory capacity for edge and industrial AI workloads.

Fabless AI chipmaker MemryX AI announced the launch of its MX3+ accelerator, a new edge‑focused processor that claims up to tenfold improvements in both compute throughput and on‑chip memory capacity for industrial and edge AI workloads.

Key specifications of the MX3+

The MX3+ builds on the company’s MX3 architecture, adding a larger memory fabric that scales beyond 1 peta‑operations‑per‑second (POPS) while keeping power consumption under 10 watts, a critical metric for battery‑powered and rugged deployments.

  • Up to 10× higher inference performance compared with the original MX3
  • On‑chip memory capacity expanded to support larger models at the edge
  • Power envelope optimized for sub‑10 W operation
  • Integrated support for popular AI frameworks via MemryX SDK

Target markets and use cases

MemryX AI positions the MX3+ for applications that require real‑time decision making on the device, such as predictive maintenance in factories, autonomous robotics, and video analytics in smart cities.

By keeping the entire inference pipeline on the accelerator, developers can avoid costly data transfers to the cloud, reducing latency and enhancing data privacy.

Industry reaction

Early adopters in the manufacturing sector have expressed interest in the MX3+ for its ability to run larger neural networks locally, while analysts note that the announced performance gains could narrow the gap between edge and data‑center AI capabilities.

For a broader view of how the MX3+ fits into the evolving edge AI landscape, see the Technology coverage on uxc.news.

MemryX AI’s announcement was distributed via a press release on PR Newswire.

MemryX AI unveils MX3+ scaling AI inference beyond 1 POPS

Track coverage on Momentum · More in Technology.