Graphics Cards – GPU & Video Processing Boards
Discover cutting-edge solutions from leading global suppliers
Curtiss-Wright Corporation has introduced the VPX3-730, a rugged, high-performance 3U VPX GPU module aligned with the Sensor Open Systems Architecture (SOSA) standard and powered by the NVIDIA RTX PRO 5000 Blackwell GPU.
The VPX3-730 is the latest addition to Curtiss-Wright’s Fabric100™ ecosystem, a comprehensive suite of VPX system building blocks designed to deliver proven interoperability and minimize system integration risk.
NVIDIA’s Blackwell architecture marks a significant advancement in artificial intelligence (AI), machine learning (ML), and vector processing, offering substantial improvements in compute performance and memory bandwidth over previous-generation GPUs. It delivers up to 15x higher performance compared to prior-generation GPUs or other accelerated solutions, depending on workload, and allows system integrators to utilize NVIDIA’s extensive software ecosystem, including NVIDIA AI Enterprise, deep learning frameworks, and neural network libraries. As an Embedded Enterprise Platform Partner with NVIDIA, Curtiss-Wright is positioned to meet increasing demand for high-performance processing solutions with reduced integration complexity.
The VPX3-730 supports a full solution stack, from rugged, high-reliability hardware to application development, optimized to leverage NVIDIA’s software tools. The module provides Modular Open Systems Approach (MOSA) system designers with significantly increased GPU resources, including additional NVIDIA CUDA and Tensor cores, expanded local memory, and support for NVIDIA ConnectX-7 100Gb/s Ethernet NICs with PCIe Gen5 connectivity. It is capable of delivering up to 50 TFLOPS of peak compute performance.
Designed to support GPU-based processing and hardware acceleration for CUDA computing, graphics rendering, and AI workloads, the SOSA-aligned VPX3-730 is engineered for deployment in challenging environments requiring size, weight, and power (SWaP) efficiency. Typical applications include advanced radar systems, sensor fusion, high-performance displays, and edge computing tasks that demand deep learning inference and high-throughput vector processing.
The Fabric100 portfolio supports 100 Gigabit Ethernet (GbE) and PCI Express® (PCIe) Gen4/5 connectivity, enabling developers to architect advanced, rugged sensor processing and AI/ML inference solutions while minimizing program risk. Curtiss-Wright also offers a total lifecycle services management (TLCM) program that supports long-lifecycle applications by mitigating technology obsolescence. For the first time, these lifecycle capabilities are being extended to NVIDIA-based solutions, helping extend product availability to align with mission and platform lifecycle requirements.
Brian Perry, Senior Vice President and General Manager, Curtiss-Wright, commented, “Furthering our commitment to accelerate the development of advanced processing and communications systems for rugged applications at the extreme edge, we are proud to introduce the industry’s highest performance 3U VPX GPU module accelerated by NVIDIA, designed for edge processing.
“The SOSA aligned rugged VPX3-730 module, the newest addition to our comprehensive Fabric100 family of high performance rugged system building blocks, provides a compelling solution for SWaP-constrained vector and AI/deep learning inference applications.”
The VPX3-730 is compatible with NVIDIA Linux drivers, which enable OpenGL and Vulkan accelerated graphics, along with OpenCL and NVIDIA CUDA for math and vector processing, as well as NVIDIA AI Enterprise for its extensive ecosystem of deep learning frameworks and neural network libraries.
