Run real-time AI at the edge without external memory, lower latency, less power, seamless integration for industrial, automotive, and defence applications.
As Edge AI expands into industrial automation, transportation, healthcare and defence, deploying high-performance AI at the edge remains a challenge. Many solutions struggle with high power consumption, latency, and complex integration, slowing adoption in mission-critical applications. Developers need compact, efficient modules that accelerate AI workloads without extensive system redesign.
To address this need, Virtium has introduced a new M.2 AI accelerator module. Powered by a low-power AI inference processor, the module delivers 24 TOPS of compute while running entirely from on-chip SRAM, eliminating external DRAM. This reduces latency, power consumption and system complexity, enabling real-time vision AI for demanding environments.
The module conforms to the standard M.2 2280 form factor for seamless integration across ARM and x86 platforms and supports industrial-temperature operation from -40°C to +85°C. Its software ecosystem allows one-click model compilation without retraining or code modifications, simplifying development from prototype to production. Floating-point activations ensure high accuracy for vision-model tasks, while compliance with trade agreements makes it suitable for government applications.
Key features of the module include:
- 24 TOPS AI compute with ultra-low latency
- On-chip SRAM operation, no external memory required
- Industrial-temperature support for harsh environments
- One-click model deployment with full software access
- Standard M.2 form factor for easy integration
Keith Kressin, CEO of MemryX, said, “By combining low-power inference technology with Virtium’s industrial expertise, we’re delivering real, immediate value to customers,” positioning the company at the forefront of Vision AI opportunities, providing a high-performance, U.S. made solution for emerging applications.
