AI & High-Performance Computing

 

Rugged Infrastructure for AI, Supercomputing, and Quantum Applications

 

Hardware that transitions breakthroughs from the lab to the battlefield.

 

Advances in artificial intelligence are enabling autonomous navigation and real-time object recognition at the tactical edge. However, transitioning these breakthroughs from the laboratory to the field requires hardware that performs outside of climate-controlled data centers.

Solving the High-Wattage Thermal Challenge

High-performance computing generates significant heat. Modern AI accelerators and GPUs often exceed 125W per module. In compact 3U or 6U VPX or similar form factors, this concentrated heat can quickly lead to overheating and hardware failure.

 

We meet these challenges with thermal management solutions tailored to your specific application. Detailed thermal simulations confirm that every module operates within a safe thermal window before deployment.

 

Working on AI software

Cooling Methods

 

Direct Forced Air Cooling

Uses fans or blowers to move air directly across the electronic modules. This is the simplest and most highly effective cooling approach, but often cannot be employed in rugged systems.

 

Conduction Cooling

A fanless approach that draws heat from components through a conduction plate and wedge locks into the chassis sidewalls. Transfers heat to external fins, which maximize the surface area to ambient air. An external cold plate is often used to enhance the cooling.

 

Airflow over Conduction Cooled Modules

In rugged systems, air is directed over internal fins within the enclosure dissipating heat from conduction-cooled VITA 48.2 modules. This approach utilizes the airflow to enhance the cooling without direct airflow on the plug-in cards.

 

Advanced Cooling Options for VITA Systems

 

Air-Flow-Through (AFT)

Directs cooling air through an isolated heat exchanger within the module per VITA 48.5 or 48.9. (or similar sub-specification)

 

Liquid-Flow-Through (LFT)

Implements VITA 48.4 designs for extreme wattage, using liquid coolants to pull heat directly from the module core.

 

Liquid-Cooled Sidewalls

A hybrid approach that circulates coolant through chassis walls to improve the cooling for standard conduction-cooled modules.

High-Speed Backplanes

Deep learning algorithms often require a communication backbone which preserves signal integrity at high speed. As a leader in MOSA/SOSA-aligned designs, Pixus often has requirements for high-speed backplane performance at 100GbE (over 25Gbauds/s per lane) and PCIe Gen4 speeds. Pixus also has experience in standard and customized solutions at rates in excess of 100GbE. This includes VITA 91 designs at double this rate with the High Density connectors, which is a stepping stone to VITA 100, which will also employ a connector that is double the density of the MultiGig RT3 connector of OpenVPX.   This bandwidth aggregates high-resolution data from radar, sonar, and electro-optical sensors without resulting in bottlenecks.

 

Next-Gen Support

Developing solutions preparing for PCIe Gen 5/6 and 400GbE architectures for future AI accelerators.

 

Optical/RF Interfaces

Significant expertise with profiles following VITA 66.x and VITA 67.x for a wide range of optical and RF interfaces.

 

Predictive Analysis

Signal integrity simulation models performance to prevent data loss in dense, rugged environments.

 


 

Supercomputing and Quantum Computing

Research in high-energy physics and quantum cryptography requires infrastructure that scales to massive processing demands while retaining precise synchronization. Pixus provides modular chassis and backplane systems that support massively parallel processing clusters.

 

We are actively developing infrastructure for the emerging VITA 100 standard to ensure that research platforms progress alongside the technology they support.

The Pixus Advantage

 

Direct Partnership

Solving high-wattage thermal challenges calls for collaboration. We provide direct access to our mechanical and electrical engineers to validate your cooling strategy and ensure signal integrity for high-speed AI payloads.

 

The Authority in Open Architectures

As a leader in SOSA™ and MOSA standards, we ensure your infrastructure is built on a future-proof foundation. This allows for rapid technology insertion, enabling you to upgrade processing modules without redesigning the chassis.

 

Certified Quality & Process Control

Deploy with confidence. Our hardware is designed to meet MIL-STD-810 for shock and vibration and MIL-STD-461 for electromagnetic compatibility, all built within ISO 9001:2015 and AS9100D-compliant workflows.

Get Expert Product Guidance

Selecting the optimal product architecture, format, and configuration for your application can be complex. Get direct access to our mechanical and electrical engineers to navigate these decisions. We’ll work together to find a solution that meets all your technical and program requirements.

 

Contact Our Engineering Team