MemryX Announces MX4 Roadmap for Asynchronous Dataflow in Data Center AI
MemryX Inc. has announced its roadmap for the MX4, a next-generation accelerator designed to enhance AI inference in data centers through a 3D hybrid-bonded memory architecture. The MX4 aims to improve memory capacity, bandwidth, and energy efficiency, building on the MX3's performance advantages, while focusing on asynchronous processing to support larger workloads like Large Action Models and high-resolution multimodal vision. A dedicated test chip is expected by 2026, developed in partnership with a 3D memory provider.
Theia Market Signal Identification - AI Assisted
MemryX Inc. has unveiled its roadmap for the MX4, a next-generation accelerator aimed at enhancing AI inference in data centers. The MX4 will utilize a 3D hybrid-bonded memory architecture to address memory capacity, bandwidth, and energy efficiency constraints.
Currently, MemryX's MX3 silicon delivers over 20 times better performance per watt than mainstream GPUs. The company has partnered with a next-generation 3D memory provider to develop a dedicated test chip by 2026, focusing on a ~5µm-class hybrid-bonded interface and direct-to-tile memory integration.
The MX4 will leverage the existing MX3 software stack to facilitate adoption and support larger memory configurations. The architecture emphasizes asynchronous processing to eliminate bottlenecks and improve scalability, making it suitable for frontier workloads such as Large Action Models and high-resolution multimodal vision.
Comments