Atlas25 – 25GigE RDMA Camera with Direct LC Fiber Connection
The Atlas 25 is a compact 25GigE camera purpose-built for OEM integration and system integrators requiring deterministic high-bandwidth imaging in the most demanding applications.
It delivers high performance reliable RDMA data transfer, streaming image data directly to main memory and bypassing the CPU and OS. This increases throughput, lowers latency, and eliminates CPU usage for streaming.
The Atlas 25 combines Sony’s 4th generation Pregius S CMOS sensors with GigE Vision 3.0 and RoCE v2 (RDMA over Converged Ethernet) to deliver up to 25 Gbps data throughput with low latency and minimal CPU overhead. Designed for applications requiring both performance and reliability, the Atlas25 provides a scalable upgrade path from existing 1, 2.5, 5, and 10 GigE systems while maintaining full GigE Vision compliance.
The Atlas 25 camera provides a direct LC duplex multimode fiber connection without the need for an external SFP module, simplifying system design, reducing power consumption, and improving overall reliability. Featuring a compact 55 × 55 mm housing, a new 17-pin GPIO connector with additional input and output lines, and a wide operating temperature range from -20 to 50°C ambient, the Atlas25 ensures reliable operation in harsh industrial environments.
The first Atlas25 models include the Sony IMX530, IMX531, IMX532, and IMX535 CMOS sensors, delivering resolutions up to 24.5 megapixels and frame rates up to 184 fps. With its fanless thermomechanical design, the Atlas25 dissipates heat efficiently without external heatsinks or fans, ensuring quiet and maintenance-free performance in space-constrained systems.
With GigE Vision 3.0 and RDMA (RoCE v2) support, the Atlas25 achieves deterministic, low-latency data transfer by bypassing the CPU and operating system layers. This architecture allows image data to stream directly into host memory, freeing up system resources for image processing, AI inference, or other critical tasks. The result is high throughput, reduced CPU load, and predictable real-time performance over standard Ethernet infrastructure.
