A manufacturer of high-speed GigE Vision cameras has released a video presentation that details major advancements in high-speed imaging technologies without using RDMA or Roce. The presentation also looks at the technology behind GigE Vision and how this can help deliver the highest camera performance with low system simplicity.

High-speed inspection today requires machine vision cameras capable of handling the required speed and data rates with high reliability and performance. Even with the machine vision industry’s most popular interface GigE Vision differences exist between communications protocols that, when understood, can unlock more performance in high-speed imaging. This presentation explains the differences between a GigE Vision Stream Protocol (GVSP) that leverages the user datagram protocol (UDP) approach, as opposed to remote direct memory access (RDMA), RDMA over Converged Ethernet (Roce), and transmission control protocol (TCP), and how it can help systems integrators, OEMs, and end users achieve top system performance.

“An optimized GSVP approach can deliver true zero copy image transfer with the lowest latency and jitter, while remaining fully GigE Vision compliant,” said John Ilett, president, CTO, and founder of Emergent Vision Technologies. “While the proprietary RDMA/Roce and TCP options offer some of the benefits of GigE Vision, they rely on flow control and resends, which can impact performance and induce latency and jitter.”

Machine Vision systems with 24 and 48 cameras

The presentation also dives into processing technologies – including Nvidia GPUs, FPGAs, and Emergent’s software and proprietary NICs – and how they unlock new capabilities for high-speed imaging, such as zero copy image transfer and GPUDirect. Additionally, it covers the adoption of 10GigE and beyond while showcasing new high-speed imaging demonstrations featuring 24- and 48-camera setups using Ecapture Pro software and GPU Direct, along with several helpful animations that provide additional context to the technology.

“In high-speed imaging applications where multiple 10GigE, 25GigE, or 100GigE streams are used, real-time processing will require offload technologies to more suitable processing technologies than just system memory and CPU, which is a point that does not come up often when alternate interface or protocol methods are discussed,” said Ilett. “By leveraging an optimized, UDP-based GVSP approach, Emergent has dedicated the development of its GigE Vision camera ecosystem to reliably support today’s most challenging high-speed imaging applications.” But it has to be mentioned, that there's no support for multicasting with RoCE.