Emergent Vision Technologies continues to break new ground in high-speed imaging and machine vision. In a recent webinar, the team explored how recent innovations like GigEVision 3.0 support, GPU-optimized pipelines, and new development tools are helping engineers and system developers tackle growing performance demands.

GigE Vision 3.0 Compliance

GigEVision 3.0 introduces support for RoCEv2 and RDMA, allowing zero-copy transfers to system memory and closing the performance gap with CoaXPress and CameraLinkHS. Emergent, having already delivered zero-copy capabilities through its advanced driver technologies, has implemented full support for the new standard via a simple firmware and software update. This includes all current models, including the compact and power-efficient EROS 10GigE cameras.

Compatibility and stability are cornerstones of any vision system. Emergent has validated a wide range of off-the-shelf network interface cards (NICs) from Broadcom and NVIDIA / Mellanox on both Windows and Linux. For customers who need long-term support or advanced features like front-panel triggering, the company also offers FPGA-based NIC / grabber hybrids.

At the International Vision Standards Meeting (IVSM) in Quebec, Emergent successfully demonstrated compatibility with leading industry players like Stemmer Imaging and MVTec, using its 10GigE EROS and 25GigE BOLT cameras. These tests confirmed interoperability and high performance across varied system environments.

Shortcomings of RoCEv2 for multi-camera, GPU Direct systems

However, the adoption of RoCEv2 is not without tradeoffs. Notably, it lacks support for multicasting, and Windows support is limited for Intel and Marvell NICs. There’s also no GPU Direct support for Nvidia GPUs on Windows. Emergent’s proprietary drivers overcome all of these limitations, including enabling GPU Direct on Windows—a critical feature for reducing system latency and avoiding performance bottlenecks.

Beyond image transfer, the challenge shifts to processing all that data. Simply getting images into system memory isn’t enough, especially at 25GigE and 100GigE speeds. Without optimized pipelines, systems can hit CPU and memory ceilings almost immediately. Emergent’s solution: real-time, zero-copy pipelines built for GPUs and other high-throughput processing units.

For those aiming to scale their imaging setups, flexibility is essential. Emergent supports not only GPU-based systems but also FPGA cards and NVIDIA BlueField super NICs that combine processing and network interfaces. High-speed SSD recording is another area where Emergent offers validated, performance-tuned solutions.

Proof of Performance in Multi-Camera Systems

Though connecting 200 cameras to a single PC sounds theoretical, Emergent has already helped customers deploy systems at similar data levels. Thanks to EROS 10GigE cameras’ auto-negotiation, competitive pricing, and minimal system overhead, Emergent’s total cost of ownership is often dramatically lower than traditional multi-server solutions.

Two new software technologies, FlexProc and FlexTrans, further simplify system design. FlexProc allows users to build custom CPU, GPU, or FPGA plug-ins without writing low-level image capture code. FlexTrans ensures that image data and results can move seamlessly through a system using best-in-class zero-copy and GPU Direct methods. Combined, they enable rapid prototyping, scalability, and high performance.

Finally, the company has introduced eSDK Pro, a simplified but powerful SDK designed to help developers quickly build multi-camera applications. With eSDK Pro, entire pipelines can be implemented in a fraction of the code typically required, making it easier to achieve performance goals and shorten time to market. Real-world examples showed complete acquisition pipelines for eight 25GigE cameras running on two servers with four GPUs, all implemented in a few lines of code.

Get in Touch

With continued innovation across hardware, drivers, and software, Emergent Vision Technologies is enabling the next generation of performance-driven machine vision applications. If your goal is to create high-performing multi-camera systems, please contact us to begin your project.