Compact embedded module for high graphics and computing requirements

The Qseven form factor has seen strong growth in almost all sectors of embedded applications with the exception of display-intensive applications from Digital Signage and POI (Point of Information) to Gaming and POS (Point of Sales). This is going to change with the latest Qseven module products from MSC which are equipped with processors from the AMD Embedded G-Series of APUs (Accelerated Processing Units). The small 70 x 70mm modules can be placed on fairly compact baseboards, and the resulting system can deliver stunning content on up to two independent Full-HD displays (resolution up to 1920 x 1200 at 60 frames per second). Because of the built-in hardware decoders for several video formats as well as DirectX-11 capability including OpenGL 4.0 support, the result will remind of workstation-style graphics rather than a low-power embedded module. Due to the lack of active cooling requirements, the final system can be extremely compact and small, allowing it to be hidden in the monitor’s stand or even added to the display electronics inside the panel’s chassis.

The AMD Embedded G-Series APUs are a new family of embedded x86 processors which combine a low-power CPU and a discrete-level GPU into a single embedded chip. Based on a new power-optimized core, the AMD Embedded G-Series platform delivers levels of performance in a compact BGA package that is ideal for low power designs in small form factors. MSC is focusing on the lowest-power APUs of this family for the new Qseven module: the AMD T40E dual-core processor and the T40R single-core chip both clocked at 1 GHz and featuring a very low power dissipation of 6.4W or 5.5W, respectively. Both APUs come with the built-in AMD Radeon HD6250 Graphics Processing Unit (GPU) which delivers outstanding graphics performance never seen before in such a low-power device.

Each APU supports dual independent high-resolution displays and delivers exceptional multi-media capability, e.g. hardware decode support for H.264, VC-1, MPEG2, WMV, DivX and Adobe Flash. Both 2D and 3D graphics acceleration and full DirectX-11 support are done by hardware without obstructing the CPU. Paired with the AMD Embedded A50M Controller Hub, the platform supports advanced interfaces such as 6Gbit/s SATA, Gen. 2 PCI-Express and HD Audio which find their perfect match in the interface definition of the Qseven connector.

Compact Qseven Module

The Qseven standard was conceived for compact embedded modules equipped with low-power and space-saving processor platforms. The 70 x 70mm module offers a comprehensive set of modern, fast serial interfaces including PCI-Express, SATA, USB, Gbit LAN, I2C, SPI, LPC, SDIO, LVDS and HDMI/DVI graphics. Introduced in 2008, Qseven has been the fastest-growing embedded module standard ever with a large manufacturer base around the world.

The Qseven module with the AMD Embedded G-Series processors provides memory and support chips forming a general-purpose computing device:

• AMD Embedded G-Series CPU T40E (dual-core) or T40R (single-core) at 1GHz
• AMD A50M Fusion Controller Hub
• 2 or 4Gbyte DRAM LV-DDR3
• Optional SATA Flash Disk (bootable)
• 10/100/1000 Ethernet
• Dual-channel LVDS 18/24 bit and DVI/HDMI/DP graphics
• 4x PCI-Express x1 channels
• 2x SATA high-speed 6Gb/s interfaces
• 8x USB 2.0 interfaces
• LPC, SMBus serial interfaces

The AMD A50M Controller Hub supports a rich interface set allowing a seamless coverage of the Qseven interface signals at the 230-pin edge connector. On the module, up to 2Gbyte LV-DDR3 DRAM are soldered, and a SATA Flash Disk (e.g. 4 Gbyte) can optionally be populated making the resulting COM universally usable in a diversity of applications. It comes with a long lifetime and availability period which for embedded products is typically five to seven years from announcement.

The AMD Fusion APUs’ impressive graphics capabilities are at the core of the Qseven module, and hence several display interfaces are available to connect up to two monitors each of which can show independent content at up to Full-HD resolution. LVDS is still a popular interface when TFT panels are to be driven directly, so that the module features dual-channel 18/24 bit LVDS even usable for highest display sizes and resolutions. DisplayPort is considered to be the next widespread standard for the purpose of driving LCD directly, and the Q7-A50M offers both DP and HDMI (which encompasses DVI graphics signals implicitly) on the same digital output lines. The baseboard designers then have the choice between these powerful display interfaces for the realization of advanced graphics systems.

Turbocharging Number Crunching

In computing-intensive applications such as image processing, image recognition or other parallel processing tasks, the AMD Embedded G-Series APUs and their APP (Accelerated Parallel Processing) open up performance dimensions which formerly were unattainable for a small computer module such as Qseven. All the above application fields can be condensed to the requirement of extremely fast number crunching, the massive multiply-and-add of floating-point numbers. And since all inputs to the equations are readily waiting to be processed at the beginning, there is no need to perform these calculations sequentially, i.e. one after another. If the underlying hardware allows computing partial equations in parallel, i.e. at the same time, then the result will be available within a substantially shorter time frame.

Today’s graphics processors are highly programmable, massively parallel compute engines. Using these “turbochargers” so solve complex floating-point equations instead of triangle and shader algorithms, will speed up the number crunching process in an extreme way. But how to control the data flow through these high-powered but unusual and complex machines? With the development of cross-platform, parallel programming languages such as OpenCL and the continued evolution of heterogeneous computing, “General-Purpose computation on Graphics Processing Units” (GPGPU) has finally started to lend the capabilities of these machines to the embedded market.

What makes GPGPU computing so interesting is the availability of extreme floating-point performance in cost-effective GPUs. And in addition to excellent GFLOPS per dollar, GPGPU computing also brings high performance per Watt. Early pioneers in GPGPU computing had no computing language that could access that compute power.  Instead, they selected graphics operations in OpenGL that used the same math they needed and then copied the results from the frame buffer.

That situation led to the development of the OpenCL computing language, which ports across GPU architectures and between CPU and GPU components to provide a truly heterogeneous programming environment.  First introduced in 2008, OpenCL today has wide industry support.  With it the application developers are enabled to partition effectively serial tasks for execution on the CPU and parallel tasks on the GPU. OpenCL is an open standard that was created by an industry consortium including many chip vendors, software companies and research organizations.

Application Variety

Massive floating-point number crunching is not limited to image processing or pattern matching. Among the industrial applications for the Qseven module with the AMD Embedded G-Series APUs there will be many applications which will make use of the high computing performance available from using processor and graphics of the AMD Accelerated Processing Unit in parallel. For automation and medical, vision algorithms may fit well the GPGPU capabilities, and Point-of-Sale applications one day will require face-recognition software to identify the user of the system. And for all security-critical applications a fast, instantaneous antivirus tool may be required which can be scaled down to a complex, fast pattern matching algorithm again fitting well the parallel structure of the GPU.

Last but not least, the strong graphics capabilities of the Qseven module with the AMD Embedded G-Series APUs enable its use for applications such as HMI (Human Machine Interaction) for industrial controls, medical and Infotainment systems and Point-of-Information and Point-of-Sales including Cash Registers, Ticket Printers, Vending Machines and similar applications.