In all industries camera resolutions and camera data transfer speeds are increasing, since ever finer structures and details have to be inspected in ever shorter cycle times. Thereby enormous amounts of data accumulate which have to be captured and processed. But, can the PC-system even guarantee the necessary bandwidth between all the expansion boards?

In Cervantes's novel of the same name, Don Quijote battles windmills which appear to him as giants. Gigantic are also the data streams which are generated by many machine vision applications. Consequently, camera resolutions and camera speed are constantly increasing. There is no end in sight. Line scan cameras with up to 12,288 pixels/line are used today as well as area scan cameras with 16 megapixels or even multi-camera systems. However, applications required to manage highest volumes of data are not restricted to machine vision applications only. Also computer tomographs, hyper spectral imaging and the 3D reconstruction and processing generate huge amounts of image data, which moreover often need to be processed in real-time.

Acquisition and processing of such big data streams does raise quite some issues for current systems: How and where to get the necessary performance? How are all the acquisition and hardware acceleration boards integrated? Is the PC-system able to guarantee the necessary bandwidth between all the expansion boards?

Matrox faces up to the battle against the windmills and presents Supersight, a flexible PC platform with high performance and nearly infinite expansion possibilities. Supersight is a true high performance computer (HPC) platform for highest performance and bandwidth.

Umpteen Slots
Supersight´s core component is the Matrox designed and manufactured PCI-Express (PCIe) 2.0 backplane. It offers four PCIMG 1.3 slots for CPU boards, the Matrox System Host Boards (SHB), and ten additional PCIe x16 2.0 slots for expansion boards like frame grabbers, VGA boards for GPU processing or any other acceleration and expansion boards. The thermal management inside the robust 19'' 4HE rackmount enclosure allows stable operation of up to 14 active cards.

Concentrated Processing Power
In order to provide a maximum of processing performance Matrox uses different, complementary processing technologies in parallel. General purpose CPUs are ideal for complex calculations and analysis of image data. FPGAs show great strength in pipeline and parallelized processes like image preprocessing through shading correction, dynamic thresholding or transformations. VGA boards are getting more and more common in imaging, since their GPUs offer significant processing power at very low cost.
Matrox Supersight allows using frame grabbers for all kinds of image sources - analog, digital, Camera Link, DVI, and HD-SDI. Already during acquisition image data can be pre-processed e.g. with a Matrox Radient High-End grabber with Altera Stratix III FPGA or with Matrox Odyssey Vision processor boards, that use PowerPC CPUs, FPGAs and a pixel accelerator ASICs.

The Matrox System Host Boards with latest generation Intel processors are supported by 64 bit Windows and integrate two Nehalem quad-core CPUs, up to 48 GB of RAM and connectivity for Dual-GigE, USB 2.0 and VGA. Additional processing is done through additional PCIe x16 VGA boards from Nvidia or ATI.

A fully equipped Supersight features up to 8 Intel CPUs with 32 cores, 192 GB RAM and any combination of up to 10 additional VGA boards, FPGA frame grabbers and vision processor boards.

Maximum Bandwidth
By just adding many processing units a massive processing power will never be achieved. Pure processing power is nothing, if data does not get transferred to the specific processing units in time. So all Supersight PCIe slots are connected through an architecture of cascaded PCIe switches, forming a PCIe switched fabric. The switched fabric groups processing elements (CPUs, GPUs and FPGAs) into compute clusters. Data inside such a cluster gets copied with a bi-directional speed of up to 8 GB/sec (PCIe x16 2.0) from board to board.

The architecture of Supersight allows great flexibility on which boards to use inside each cluster. Any configuration of processor boards, GPUs and optional frame grabbers working jointly on all the data without interfering with each other on a common bus.
Since all the links between the separate clusters are based on PCIe x16, communication between the clusters is very fast, too. Unlike classical segmented backplanes, the Matrox Supersight backplane, through the Distributed MIL (DMIL) communication API, enables clusters to look like a unified system.

Perfect Platform
To the advantage of demanding and computing-intensive applications, Matrox Supersight brings together the latest processing technologies - multi-core CPUs, VGA boards and FPGAs - and the flexibility of the switched fabric PCIe backplane for highest bandwidth. The Supersight high performance computing platform is a perfect fit for demanding and data intensive applications.