Teledyne peer-to-peer streaming between PXIe digitisers and GPUs
Teledyne SP Devices at Defence and Security Equipment International (DSEI) Japan, demonstrated the industry’s first implementation of peer-to-peer streaming between commercial-off-the-shelf (COTS) digitisers and graphics processing units (GPUs) in PXIe Systems.
The demonstration employed Teledyne SP Devices ADQ7 PXIe digitisers and RADX Technologies Catalyst PXIe GPUs that are based on NVIDIA Quadro GPUs.
Peer-to-Peer streaming uses Direct Memory Access (DMA) between the digitiser and the GPU directly over the PCIe Bus without having the data pass through the Peripheral Component Interconnect Express Extensions for Instrumentation (PXIe) system’s CPU or memory subsystem. This technique enhances bandwidth per channel, while simultaneously reducing latency and minimising the overhead incurred by the PXIe System’s CPU and Memory subsystems. The net benefit of peer-to-peer streaming is significantly improved real-time signal processing performance and system scalability, which is essential for advanced multi-channel and ultra-wideband applications like Test & Measurement (T&M) and Electronic Warfare (EW) Applications. These applications also benefit from machine learning, where the connection to a GPU is important.
“Peer-to-peer streaming to GPUs is an essential capability that Teledyne SP Devices has long supported in its PCIe digitiser products, said Paul McCormack, Teledyne SP Devices Product & Marketing Manager. “With the advent of the RADX Catalyst PXIe GPUs, we can now add this important capability to our PXIe product portfolio as well, which is essential for many aerospace, defence, automotive, and other long-life-cycle test and measurement and electronic warfare applications.”
At DSEI Japan, Teledyne SP Devices demonstrated an ADQ7 2-CH, 14-bit, PXIe digitiser that is continuously sending 5 Gigasamples per second via peer-to-peer streaming to an adjacent RADX Catalyst PXIe GPU that employs an NVIDIA Quadro T600 and is continuously performing 1 Megasample Fast Fourier Transforms (FFTs) on the incoming stream. The PXIe system employed in the demonstration included an NI PXIe-1095 chassis and an NI PXIe-8881 embedded controller running Ubuntu Linux 20.04. The demonstration shows that the peer-to-peer processing already available for PCIe applications, like computers and servers, is available for more industrial applications requiring PXIe interfaces.
“Teledyne SP Devices continues to demonstrate industry leadership with its support for peer-to-peer streaming between their PXIe digitisers and RADX PXIe GPUs,” said Ross Q. Smith, RADX Co-Founder and CEO. “With peer-to-peer streaming between PXIe digitisers and GPUs, PXIe test system integrators can now, for the first time, conduct ultra-wide-band, low-latency, real-time signal analysis and machine learning in the PXIe systems right where the data is acquired, and at price points and precision levels previously unachievable solely with FPGAs or CPUs.”