MX4
More AI. Less infrastructure.
High-performance AI inference—from edge systems to data centers. MX4 brings MemryX’s efficiency-first architecture to advanced vision, multimodal, generative AI, and LLM workloads, delivering greater compute and support for larger models without the traditional power and infrastructure burden.
Bigger models shouldn't mean bigger compromises.
AI is moving toward larger models, richer inputs, and more complex pipelines. Delivering that capability today can mean higher power, more memory, more cooling, and more infrastructure.
MX4 is being developed to change that equation.
With dramatically expanded compute and model capacity, MX4 extends MemryX into increasingly demanding AI workloads while preserving the efficiency-first approach at the foundation of our architecture.
DO MORE WITH AI
More capability where it matters.
MX4 builds on the MemryX dataflow architecture to keep data moving efficiently through the accelerator and reduce the costly movement of model data that limits conventional compute architectures.
125x
More Compute
1600x
More model weight capacity
30x
More memory bandwidth
10x
More power efficient
Relative to MX3
Move beyond vision
From perception to generative AI.
MX4 expands the MemryX platform into a broader generation of AI workloads, including large language models and multimodal applications.
Whether intelligence begins with a camera, a sensor, text, audio, or multiple inputs at once, MX4 is being designed to provide the compute and model capacity required for increasingly sophisticated inference.
Video Management Systems
Industrial & Robotics
Transportation
Smart Devices
Build what's next.
MX4 has availability targeted for 2029. Talk to our team about the MemryX roadmap, future deployment requirements, and opportunities to build with the next generation of MemryX AI acceleration.
