Performance is measured in Bytes per second. The throughput in frames per second is dependent on pixel depth, video capture resolution and any wider system bottlenecks.

 

• VisionSC range ~ 6.4 GB/s per card
• • VisionDP2 ~ 6.4 GB/s
  • VisionAV range ~ 800MB/s per device
  • • VisionAV ~ 800 MB/s
    • VisionAV-HD, VisionAV-SDI ~ 1.6 GB/s
    • VisionHD4 ~ 3.2 GB/s
  • VisionRGB S range ~ 650MB/s per card (devices share the card bandwidth)
  • • VisionRGB-E2S ~ 650 MB/s
    • VisionRGB-E1S ~ 650 MB/s
    • VisionSDI2 ~ 650 MB/s
    • VisionDVI-DL ~ 650 MB/s
  • VisionRGB range ~480MB/s per card

Example data:

• 4K / 30 fps / 16 bit pixel depth ~ 506 MB/s
• 1080p / 60 fps / 32 bit pixel depth ~ 475 MB/s
• 1080p / 60 fps / 16 bit pixel depth ~ 237 MB/s
• 720p / 60 fps / 16bit pixel depth ~ 105 MB/s
• 16 x CIF / 15 fps / 16 bit pixel depth ~ 50 MB/s

MultiStream functionality is made possible through low level hardware control available only in the Vision capture card range. MultiStream technology provides DMA support for 16 independent capture clients, for example, a VisionHD4 card has a 64 clients. Each client can have independent capture size (scale factor), frame rate, pixel depth, DMA destination and cropping rectangle. MultiStream allows 'Preview', 'Capture' and 'Stream' clients with different bandwidth requirements to be supported from a single input source.

Independent cropping allows mosaic style inputs to be cropped into separate buffers or windows. A different pixel depth can be used for each input capture client including: RGB32, RGB24, RGB565, RGB555, RGB8, YUY2, YVYU,UYVY,NV12 and H264.

MultiStream technology does not require the data to transfer via system memory for depth, rate and size conversions.

The Datapath hardware timestamp uses a common clock for both video and audio buffers.

Hardware time stamping provides precise knowledge of when a frame is captured, allowing multiple buffers from different input sources to be synchronized. The start time stamp corresponds to the arrival of the first pixel of a field or frame within the Vision hardware. The hardware timer is synchronized with the systems high precision performance counter across all capture inputs in a system, regardless of whether they are from different physical capture cards. The units of the timestamp are in 100ns 'ticks' with a resolution of 100us. Timestamping functionality can be adapted for high resolution frame stitching across capture cards, stereoscopic interleaving and input signal calibration.

The audio buffer time stamp is represented by the first stereo pair sample captured into the hardware audio store. The common video and audio time stamp allows precision lip-sync streaming when the Datapath clock is used for rendering.

Using various APIs including RGBEasy, Kernel Streaming, DirectX and Open GL, low latency peer-to-peer DMA operations are achieved using a proprietary technology allowing capture lines (not frames) to be transferred as they arrive into the frame store. The process reduces the DMA transfer time from a full frame time to a single line time. Datapath LiveStream: Capture technology is available for selection if and when required, only a single client per input can be considered for LiveStream. Using Datapath LiveStream: Capture technology, clients 2-16 will operate as usual in a 'round robin' transfer schedule immediately after the LiveStream: Capture client transfer. If the LiveStream; Capture client is disabled, all clients will DMA after the completed frame has finished capturing into the frame store. The capture card transfers data to system memory or off-screen memory on the graphics device, depending on the system and how the software is written.

The delay on a third-party graphics system is dependent on several factors, for example, if the data can be sent to the GPU directly or if transferred via system memory, how many buffers are queued, etc. Sample3c of the RGBEasy SDK shows the lowest implementation of latency using DirectX to DMA data directly to a GPUs memory back buffer.

Intel GPU and AMD APU display technologies can achieve a 1-2 frames of 'glass to glass' latency.

The Vision application bundle allows both viewing and control of captured data. Property pages include, 'Window' for full control of the bounding window layout, 'Input' for all signal source related controls, 'OSD' provides screen overlay of text and graphics, and 'Audio' source control of audio input and output formats. The Vision application provides the functionality to save configured windows as .RGB files after user set up. A unique window 'ID' system allows existing windows to be controlled via the extensive Command Line Interface.

A Windows Kernel Streaming driver provides input centric endpoints for supported third party application use. Additional DirectShow filters exist for PiP (picture in picture), diagnostic transform filter for time stamp analysis and a NULL renderer for bandwidth analysis.

The Vision range of capture cards listed above use a single driver, application and interface model. A single Windows installation package is required for all cards and operating systems covering x86 and x64 bit, Windows XP to Windows 8. Microsoft Windows interfaces include Kernel Streaming, COM, DirectShow, MMAPI and WASAPI. A private Windows RGBEasy interface allows finer control of Datapath hardware when required.

A Video for Linux installation including support for the G-Streamer application is available for downloaded at the Datapath website. Card model dependent features will return an appropriate error when not supported.

A captured frame can be scaled up or down within a 4K x 4K range using a high quality polyphase filter. The scaling process reduces overall DMA bandwidth by ensuring any downscaling is performed on the capture card before data is transferred to the destination. Any of the 16 MultiStream input clients can be up or down scaled independently.

Signal detection is provided by internal driver algorithms assisted by physical hardware measurements. This provides accurate signal detection for all signal sources including analogue 3/4/5 wire, DVI, HDMI, component, composite SDI, dual link, display port and s-video. Using Datapath event interface call back mechanisms, 1:1 application streaming/display can therefore be maintained (see sample2 DirectShow SDK).

When an agreed buffer size has been negotiated by an encoding application for example, any changes to the input signal thereafter will be scaled to the original agreed data format for both video and audio. As a further example, a stream has been negotiated at 1080p@60 FPS video with a 48KHz audio stream, an encode process is started, the signal is then removed by the user, 'no signal' painted video frames are provided at 60 FPS with audio silence at 48KHz provided by the Vision driver at the agreed capture rate and audio format. If a new signal is inserted into the hardware by the user, for example 640x480 at 29.97 FPS with a 44.1KHz audio stream, the Vision hardware will up scale the video to 1080p at 60 FPS and up scale the audio to 48khz. This functionality will work for both up and down rate conversion. Note, the encoding application will not be aware of any physical changes on the input in this example.

The number of physical Vision capture cards within a system is limited by the motherboard BIOS bus allocator. Any combination of signal input cards listed above can be mixed together. Access to physical inputs can be ordered differently to the default device enumeration using the jumper link settings on the capture card.

Each DVI or HDMI capable input exposes a 128 byte Vision or 256 byte VisionAV EDID. Each EDID has a 'Preferred' and 'Additional' timings block available for configuration. The EDID can be displayed and configured in either 'Video Definition' or 'EDID' format.

Captured data is stored in 24bit 4:4:4 format and can be converted to a number of formats during the MultiStream DMA data transfer. Example output buffer pixel depths include: RGB888, RGB566, RGB555, RGB8, YUY2, YVYU, UYVY, NV12. The VisionHD2-SQX additionally supports the H264 compressed format.

Three memory buffers receive frames sequentially 1,2,3,1,2,3... Frames are read out from the last completed and therefore latest buffer. Triple buffering ensures pixel data is not written and read from the same buffer at the same time avoiding 'image tearing'. A LiveStream client will function differently, this is described in the LiveStream: Capture section.

• 256MB DDR2 Frame store on all VisionAV capture cards
• 128MB DDR Frame store on VisionDVI-DL capture cards
• 64MB DDR Frame store on VisionSDI2 capture cards
• 32MB DDR Frame store on VisionE1s/E2s capture cards

The Vision interface allows text, bitmaps and user defined operations to be overlayed onto the capture data buffer. Capture data overlay is achieved using either the Vision application OSD property page, Vision command line interface or RGBEasy API.

Support for VisionAV range captures of HDCP sources onto ImageDP4 graphic card only.

Low latency transfer interfacing for third party graphic device vendors. See Sample5 of the RGBEasy SDK for working examples of Datapath Vision captures into these technologies using the OpenGL API.

AMD DirectGMA and NVIDIA GPUDirect™ technology in conjunction with the LiveStream: Capture typically produces 2-3 frames of latency. 1080p/60Hz/true colour input to output rate. This rate includes capture, transfer and display. A 'peer to peer' DMA is setup and executed in this model.

SDK features described above are documented by sample applications found in the Datapath SDK bundle including RGBEasy, DirectX, DirectShow, MMAPI, WASAPI, OpenGL and .Net source code.

The Vision SDKs can be found either within the Vision stand alone installation or as a separate downloaded from the Datapath website.

Datapath hardware is manufactured to ISO 9001 standards, each board is subjected to an automated solder paste inspection, automated optical inspection and x-ray inspection. Electronic post assemble tests are conducted in addition to a visual IPC-A-610 Class 2 inspection.

Electrical tests are tailored according to Vision card design and will be based around one or more technologies:

• Boundary Scan
• In-Circuit Test
• Flying Probe
• Functional Test

Prior to acceptance and shipping, all Vision hardware is functionally tested and re-inspected on site at Datapath.

Datapath hardware is certified listed accessory cards under IEC & UL 60950-1 – File number E345219. Each capture card includes a unique serial number, which is traceable back to bare printed circuit board.

Software is passed through both automated test procedures for regression testing and written test procedures, accredited by test certified personnel.

Datapath is large enough to undertake substantial projects, but small enough to offer an unrivalled level of service and support, which is achieved by an accessible, customer focused staff. Datapath offers high level consultancy of performance solutions required for multiple or single projects for existing or new customers.

Datapath support responses are answered by specialized Vision hardware or software engineers.