THANK YOU FOR SUBSCRIBING
Aerospace and Defense Review Weekly Brief
Be first to read the latest tech news, Industry Leader's Insights, and CIO interviews of medium and large enterprises exclusively from Aerospace and Defense Review
Mike Knowles, President and CEOFor the defense market, the U.S. and its global allies understand it will be difficult to win a war of attrition against peer adversaries, such as Russia and China, which base their strategy on centralized command and control. This is why the U.S. and its allies have chosen to implement a distributed or decentralized command and control strategy. In this model, commanders at the battlefield edge use advanced sensors and systems to assess, decide, and act on military theater scenarios faster than their adversaries.
Key to this strategy is the ability to implement AI and associated data processing to fuse sensor information, assess scenarios and implement decisive action. This approach is characterized by the DoD Joint All Domain Command and Control (JADC2), and this has been driving the increased demand for AI-enabled edge processing, sensor fusion, autonomy, and simulation.
In the commercial market where AI is considered part of the Fourth Industrial Revolution, we see similar implementation at the edge, where sensor and decision systems interact to support rapid conversion from assessment to action. Most notably, we see this in use by trucking, agriculture, and in other industries where the sensor fusion of radar, LIDAR, laser, and infrared are collated and processed by AI to support autonomous operation.
You mention “edge processing.” What exactly does this mean and how is it applied?
Edge processing represents computing at the point where events and data acquisition and related user decision making is taking place. This compares to the popular “cloud” concept where computing is done at centralized data centers, and the results are delivered to users at the edge via the cloud.
We are proud that our products and capabilities are being adopted across the Department of Defense as well as in several commercial applications.
You can think of our products as datacenters on the move that are designed to operate in the harshest of defense or commercial environments. Since we can make a massive scale of computing power available at the edge, we enable users to benefit from the power of AI in real-time where seconds, or even milliseconds, count.
What are the key differentiators that OSS brings to edge processing, particularly in addressing the increasing demands by the defense and commercial sectors?
OSS is differentiated in three key areas: packaging/ruggedization, technology application, and industry relationships. For ruggedization, OSS is the only company which is successfully ruggedizing and certifying datacenter-class AI processing hardware, such as NVIDIA® HGX-A100 and AMD or Intel server-class CPUs for use in vehicles. It takes leading industry expertise in electrical, mechanical, and power design to reliably deploy datacenter products in the harshest edge processing environments.
From a technology perspective, we can leverage our expertise in development and integration of leading-edge technology to match the pace of compute and latency requirements in some of the most demanding domains such as AI, sensor fusion, and autonomy.
For example, OSS is a recognized expert in PCI Express technology. For nearly 20 years, we have led the market in reducing latency, increasing bandwidth, and creating large-scale rugged edge processing systems. Our leading-edge PCI Express products provide the highest speed interconnects required for achieving these goals. Even today, we continue to reduce latency for mobile AI systems by many orders of magnitude compared to competitor’s products.
We can achieve these differentiated offerings by leveraging our deep relationships with key technology providers and turn that privileged access into differentiated performance. We have been a close Intel partner for 23 years and NVIDIA NPN partners for 13 years, which has enabled us align our products closely with their roadmaps. Thanks to these longstanding relationships, we hold a special NVIDIA to design components with their highest-end datacenter GPUs and interconnects,
Could you provide insights into the product types OSS offers, and particularly those that meet the technical demands of the military and commercial markets?
Our products focus on the delivery of high-performance compute and storage capabilities. Two important products in these categories include Rigel, which is the world’s most compact, rugged GPU supercomputer, and our Short Depth Server, which is a ruggedized, multi-purpose GPU compute and storage server. Both systems provide scalable levels of compute and storage capability for rugged defense and commercial applications.
Using A100 GPUs and third generation NVIDIA® NVLINK™ technology, Rigel delivers 5 PetaOPS of INT8 AI inference performance on up to 28 simultaneous AI workloads for natural language processing, object recognition, threat detection and sensor fusion processing. The GPUs are integrated with OSS PCIe Gen 4.0 switch fabric technology that provides low latency 256Gbps direct memory transfers between network or FPGA sensor capture, GPU processing, and solid-state storage. Its performance and compactness enable large scale, multi-workload, AI inference and autonomous applications at the edge.
Our 3U short-depth server, or SDS, can record and store up to 720 terabytes of data on ruggedized, removable, and transportable SSD data canisters. It can simultaneously process the data using four of the most powerful NVIDIA H100 Tensor Core GPUs that provide up to 15.8 PetaOPS of INT8 AI inference performance. Given this high level of performance, the entire AI workflow—from data capture and storage to processing and networking—can be done in a single, all-in-one system. To support the scale out for any AI Transportable compute applications, additional GPU or FPGA can be added to the 3U SDS using OSS expansion enclosures.
Can you provide some examples where OSS rugged edge processing has delivered the power of AI, autonomy or sensor fusion/processing?
We are proud that our products and capabilities are being adopted across the Department of Defense as well as in several commercial applications.
We have expanded the storage technology capability on the Navy P-8A reconnaissance aircraft, allowing the Navy to store high volumes of real-time data collected from the aircraft’s advanced airborne sensors comprised of multifunction radar and associated tracking systems.
We recently engaged with the U.S. Army on efforts to replace vehicle crew computers with GPU-accelerated systems that can deliver low latency video for 360-degree situational awareness.
For the U.S. Air Force, we have delivered a data storage and compute system that supports electronic warfare assessment. Our system allows the Air Force to record massive volumes of simulation data and deliver it at high speeds with low latency to scientists and operators on the network.
For commercial markets, we have announced several projects in autonomous long-haul trucking over the last year. This has included providing datacenter-class compute for autonomous navigation and high-speed data logging for safe and efficient fleet operations.
We also recently engaged with a number of commercial maritime OEMs to provide autonomous ship control systems for harbor patrol, harbor monitoring and container shipping applications using our 3U SDS that feature NVIDIA GPUs.
Edge AI Acceleration for Defense Decision Cycles
AI accelerator systems in aerospace and defense are no longer evaluated solely by processing power. Their real value lies in how effectively they bring advanced computing closer to sensors, operators and autonomous systems without overwhelming the platforms that carry them. Modern aircraft, ground vehicles, ships, submarines and unmanned systems generate enormous amounts of radar, video, infrared, lidar and electronic data. The challenge is often not whether an AI model can run in a data center, but whether it can process information at the edge quickly enough to support decisions while operating within strict power, space and thermal constraints.
Defense leaders must not approach AI acceleration as a mere hardware upgrade but as an architectural issue. Today's mission demands the concurrent execution of many AI applications,, ranging from sensor analysis and data fusion through decision support to autonomous functions. The prevailing CPU-based architecture struggles to meet the demanding requirements of processing large amounts of data in real time. GPU acceleration offers much-needed parallel processing capabilities for such applications, but the system's performance also depends on the efficiency of data movement, storage and management across the platform.
The most effective solutions bring data center-level performance into operational environments while meeting the practical requirements of defense programs. Size, weight, power and cost remain critical considerations because every additional component affects payload capacity, endurance and maintenance demands. Low latency is equally important. Information that arrives after a mission may support analysis, but it cannot assist with decisions that need to be made onboard and in real time. Organizations should also consider scalability, choosing architectures that can support future sensors, applications and AI models as operational requirements continue to evolve.
Ruggedness also has to be understood beyond basic enclosure strength. Defense buyers need systems that can tolerate motion, vibration, temperature variation and electromagnetic constraints while still supporting current GPU technology. The best fit is usually a vendor able to combine accelerator hardware, high-bandwidth interconnects, storage and management software into a deployable architecture instead of forcing integrators to stitch together data-center products and field computing gear. That breadth reduces integration risk and keeps future refresh cycles practical as GPU families, standards and mission software change. Procurement teams also need proof that cooling, firmware control and data movement will remain manageable after deployment. AI hardware is only valuable when crews can monitor heat, adjust power use, move stored data and update systems without creating a maintenance burden or locking the program into one GPU path.
One Stop Systems stands out because it concentrates on moving data-center performance to the edge, rather than merely hardening lower-performance embedded computing. Its work emphasizes rugged servers, GPU accelerators or extenders, PCI Express switch fabrics, storage systems and software that helps align AI inference demands with efficient GPU and fabric choices. Its relevant portfolio includes Rugged Edge AI, PCIe Expansion, rugged supercomputers, compute accelerators, expansion systems, flash storage arrays and Ion Accelerator software for AI workflows. For aerospace and defense teams that need compact, low-latency AI acceleration near sensors and platforms, One Stop Systems is a clear recommended choice.
...Read more
| Share this Article: |
