Abaco: graphics XMC minimizes latency for mission-critical applications - Embedded.com

Abaco: graphics XMC minimizes latency for mission-critical applications

Abaco Systems further expanded its growing range of video/graphics solutions with the launch of the rugged, high bandwidth NVP2102A XMC Graphics and GPGPU Card. Designed to complement the recently-announced NVP2102, the NVP2102A offers additional support for legacy interfaces and peripherals.
Based on the NVIDIA Quadro Pascal (GP107) P2000 GPU architecture and providing up to 2.3 TeraFLOPS of peak performance, target applications for the new board include those that require very high-end graphics capabilities or CUDA support when performing general purpose processing, and that require raw video capture and display. The NVP2102A is competitively advantaged by its provision of four 3G-SDI inputs – twice the number typically found. It also provides two NTSC/PAL video inputs as well as supporting two audio inputs. Video output is via two 3G-SDI and two DVI or DisplayPort™ interfaces.
Of particular note is the NVP2102A’s support for direct video capture to GPU memory. This significantly reduces latency, minimizing glass-to-glass time – enabling the delivery of actionable information in closer to real time. The input resolution of incoming video is automatically detected and raw video frames are transferred directly to GPU memory (or host memory). In GPU memory, processing such as image analysis, image enhancement, 360-degree video stitching, sensor fusion, target detection and so on takes advantage of the extraordinary capabilities of NVIDIA GPGPU (CUDA/OpenCL) technology to deliver output to the user far faster than would otherwise be possible. The captured data can also be encoded using the GPU-native H.265 (HEVC) or H.264 encoders directly in GPU memory. Windows or Linux drivers and API are available for x86 systems.
The NVP2102A supports Abaco’s AXIS ImageFlex. AXIS ImageFlex is an image processing and visualization toolkit enabling rapid development of high performance image processing, visualization and autonomy applications aimed at size, weight and power sensitive platforms. It is focused on high performance GPU processing and graphics with interoperability with other programing paradigms such as OpenGL, OpenCL, CUDA and OpenCV.
The NVP2102A is offered in conduction cooled or aircooled variants with rear XMC I/O on Pn6. The rear XMC pin-out is compatible with VPX systems that follow VITA 46.9 x12d+x8d+24s. 3U VPX and PCIe form factors are supported with carrier boards.