Enhanced 3U VPX GPGPU board incorporates NVIDIA Turing architecture
Aitech Systems has announced that it now offers its C530 GPGPU board with powerful NVIDIA GPUs, based on NVIDIA’s Turing architecture. The rugged, high performance board helps designers overcome major hurdles in the rugged AI landscape by providing accelerated data processing of multiple streams simultaneously, while withstanding extreme environmental conditions.
The enhanced 3U VPX board blends NVIDIA technology with Aitech’s powerful ruggedisation and SWaP capabilities. Because it is based on COTS, open-standard architectures, the C530 can be utilised in a number of platforms and applications.
The use of multi-layered artificial neural networks and the NVIDIA Turing architecture-based GPUs’ ability to concurrently execute floating point and integer operations give the board a distinct performance advantage over non-GPGPU accelerated architectures.
When equipped with the NVIDIA RTX 3000 GPU, the C530 houses 1920 CUDA cores for parallel processing, 240 Tensor Cores for AI inference, 30 RT Ray-Tracing Cores for real-time rendering and 6 GB DDR6 for processing of up to 5.3 TFLOPS (FP32). The T1000 GPU’s 896 CUDA cores and 4 GB DDR5 enable 2.6 TFLOPS (FP32) of data processing.
Dan Mor, GPGPU product line manager for Aitech, noted: “There’s a growing need for more advanced platforms that enable higher computation power and relatively low power consumption, while handling many independent video outputs and providing a large throughput back out to the network. Our enhanced C530 uses the power of NVIDIA to provide these performance needs for compute-intensive applications like AI delivery, video analytics and image processing.”
The C530 delivers accuracy in complex tasks such as object detection, classification, segmentation and motion detection. It is used widely throughout rugged HPEC applications, including unmanned and autonomous vehicles—both ground and aerial—and in surveillance, targeting and advanced weapons systems found in naval, avionics and industrial environments.
Via an MXM site, the C530 currently supports the NVIDIA RTX 3000 GPU, consuming only 80W, and the NVIDIA T1000 GPU, consuming only 50W, with new configurations released as higher-performance MXMs become available.
The C530 operates as a peripheral board with a compatible x86 VPX host SBC, connected to the host SBC over the VPX backplane, via a high speed PCIe Gen3 link of up to 16 lanes. Four independent HDMI video ports each support resolutions of up to 1600x1200 at 60Hz.