They are designed to power intelligent systems operating locally across a wide range of use cases, including industrial robots and machines, smart cameras and inspection systems, autonomous vehicles, retail infrastructure, and medical imaging and healthcare devices.
To address the diverse deployment environments demanded by various AI applications, ADLINK's MXM modules, powered by integrated NVIDIA RTX PRO™ Blackwell Generation GPUs, offer exceptional flexibility in power and performance configuration. Supporting power budgets from 45W to 150W, the platform allows system designers to fine-tune performance profiles based on application priorities, whether maximizing compute capacity or operating within tight thermal and power constraints.
The ADLINK EGX-MXM-BW2000, for example, sets a new standard for compact AI performance at the edge by delivering up to 100W of power in an MXM 3.1 Type A form factor (82x70mm), maximizing compute density for embedded systems with limited space. At the high end, the EGX-MXM-BW5000 features NVIDIA Multi-Instance GPU (MIG) technology, which allows developers to partition GPU resources for parallel workloads, improve isolation, and manage resources more efficiently in advanced edge AI deployments. This robust combination enables the efficient execution of large language model (LLM) applications, deep learning pipelines, and other data-intensive workloads directly at the edge.
“Having worked closely with our customers for many years on a variety of embedded applications, from medical imaging to industrial automation, we’ve witnessed a significant shift in how AI is being deployed at the edge,” says Edgar Chen, General Manager of ADLINK’s Edge Visualization Business Unit. “As AI adoption accelerates, our customers are demanding higher GPU performance, greater memory capacity, and uncompromising reliability. With our MXM modules based on the NVIDIA Blackwell architecture, we offer a powerful combination of performance, efficiency, and flexibility, enabling system designers to confidently build next-generation edge AI solutions tailored to their environments.”
Built on NVIDIA Blackwell, the new MXM modules deliver revolutionary AI inference performance with FP4 precision, achieving up to twice the throughput while reducing memory bandwidth demands. NVIDIA’s latest CUDA cores, 4th Gen RT cores, and 5th Gen Tensor cores provide powerful GPU acceleration for AI inference, computer vision, and real-time data analytics, with peak FP32 throughput ranging from 9.2 to 49.8 TFLOPS. All models include next-generation GDDR7 memory, offering high transfer speeds and low latency for consistent and efficient processing of complex, high-demand workloads at the edge.
Depending on the configuration, the GDDR7 memory subsystem supports up to 24GB, memory interfaces up to 256 bits, and bandwidth up to 896GB/s, significantly increasing throughput while minimizing latency for demanding applications such as deep learning, high-resolution imaging, and advanced simulations.
All modules support up to four DisplayPort 1.4a outputs, enabling multi-monitor visualization for industrial, transportation, and autonomous systems. With a 5-year availability and wide operating temperature ranges (Standard: 0°C to 55°C, Extended: -40°C to 85°C), the platform ensures long-term reliability, lifecycle stability, and dependable operation in both commercial and extreme industrial environments.
