BROWSE BY TECHNOLOGY



RTC SUPPLEMENTS


TECHNOLOGY CORE

AMD Brings Power into SoC Designs

Qseven Standard Guides AMD G-Series into Even Smaller Form Factors

As AMD’s SoC family becomes smaller and consumes less energy while becoming computationally much more powerful, it is only reasonable to pair it with the small, versatile and rugged Qseven form factor.

DAN DEMERS, CONGATEC

  • Page 1 of 1
    Bookmark and Share

Article Media

In 2013, AMD shocked the embedded systems market by announcing the first x86 quad-core system on chip (SoC). Previously, x86 SoCs were limited to two CPU cores along with modest graphics and minimal I/O interfaces. Four-core solutions required two chips and a large thermal solution. Now the two-chip G-Series platform with discreet-GPU class integrated graphics has merged together into a single chip, thanks to an advanced design coupled with 28nm process technology. This announcement set the stage for an all-out war between AMD and Intel, and a few months later Intel followed with its own SoC announcement.

Embedded system manufacturers have always had a “make versus buy” choice when it comes to x86 processors. The latest wave of low power silicon with very respectable performance is once again flooding the market with consumer motherboards and embedded small form factor single board computers (SFF SBCs). For applications like point of sale (POS) terminals and information kiosks, the I/O requirements resemble desktop and laptop PCs closely enough to allow those motherboards to be used directly.

However, the rest of the embedded x86 space is very diverse when it comes to application I/O. SFF SBCs are very disorganized and fragmented in the area of expansion bus interfaces. Therefore, most system OEMs are way down the path of choosing COM (computer-on-module) form factors like COM Express and using standard interfaces like PCI Express and USB to customize the peripherals in the form of a just-right I/O card called a “carrier board” or “baseboard.” Rather than I/O being stacked right above the heat from the processor, the novel COM approach spreads out the design in X and Y dimensions, thereby allowing a range of affordable thermal solutions from fansinks and heatsinks down to flat heat spreader plates.

While COM Express has been quite successful with the 3-chip and 2-chip x86 platforms, there is significant overhead in terms of size and connector cost that doesn’t scale down very well to all-in-one SoC solutions. What is the point of collapsing two chips into one if the board space doesn’t also shrink? The cost savings of a single chip gets partially squandered when the tiny chip still goes on large module PCBs with expensive board-to-board interconnects to the carrier board. Clearly, a lower cost connector scheme and a smaller board are needed. Rather than inventing yet another form factor, industry leaders in embedded SFF standards are choosing Qseven as the form factor on which to implement the G-series SoC. Before diving into the details of a Qseven solution, let’s first examine what comprises the powerful SoC itself.

The Ultimate Fusion

The embedded evolution marches on with a x86 CPU, graphics processor comparable to stand-alone external GPU card, and I/O controller (southbridge) all on a single die. While x86 SoCs are not a new invention, the attainment of this level of performance and integration is what’s unprecedented. With up to 2 GHz CPU performance per core, and having four such cores along with Radeon graphics, the AMD Embedded G-series SoC blends an impressive fusion of technologies into a tiny chip (Figure 1).

Figure 1
AMD G-series SoC.

The G-Series SoC platform is a high-performance, low-power System-on-Chip (SoC) design, featured with enterprise-class error-correction code (ECC) memory support, dual and quad-core variants, integrated discrete-class GPU and I/O controller on the same die. It achieves superior performance per watt in the low-power x86 microprocessor class of products when running multiple industry standard benchmarks. This brings the kind of exceptional multimedia experience from desktop and laptop computers down into embedded devices, and the CPU+GPU provides a heterogeneous computing platform for parallel processing. The small-footprint SoC sets the new baseline for power efficiency across different workload types found in the variety of embedded applications. ECC memory support helps the SoC support application requirements that were previously inaccessible to x86 products in these power envelopes at this price point.

An AMD G-Series SoC offers 113% improved CPU performance compared to the previous two-chip AMD G-Series APU. Its advanced GPU supports DirectX 11.1, OpenGL 4.2 and OpenCL 1.2, enabling parallel processing and high-performance graphics processing that provides up to 20% improvement over the 2-chip APU. Excellent compute and graphics performance with enhanced hardware acceleration delivers up to 70% overall improvement versus the G-Series APU.

The SoC design offers a 33% footprint reduction compared to the AMD G-Series APU two-chip platform, simplifying design with fewer board layers and simplified power supply. The low-power SoC also reduces overall system costs. It enables fanless designs, enhancing system reliability by eliminating moving parts.

AMD’s standard embedded 5-year availability and support (additional 2 years possible) qualifies the SoC for long lifecycle medical, military, vehicle and avionics applications. From the cost perspective, the G-Series SoC platform brings performance and efficiency with desirable features, delivering lower total cost of ownership and higher ROI.

The Ever-Shrinking Form Factor

With an array of performance options, the AMD G-Series SoC platform allows OEMs to utilize only one board design to cover solutions from entry-level to mid-range. The SoC design enables unprecedented levels of combined graphics and CPU performance in SFFs.

Mid-range performance in a smaller package deserves a smaller form factor. Qseven was invented for 1-1.6 GHz single core processors. Now that dual core Gigahertz processors with desktop GPU have arrived within the same 10-12W power envelope, Qseven becomes a natural choice for implementing a G-series SoC design.

Originally invented when 2-chip x86 solutions appeared in late 2007 to provide an entry level below the 3-chip notebook processors, the tiny 2.75 x 2.75” (7x7 cm) board size has already replaced ETX as the high volume cross-architecture open standard for low- and mid-range embedded computing.

The “Q” in the name Qseven comes from the word “quadratic” (square), and “seven” is the module’s 7 cm form factor. This base area allows the deployment of a powerful and efficient cross-architecture platform with extensive interface options, which at the same time retains the compact dimensions needed to facilitate its integration into handheld equipment.

Unlike most of today’s board standards that use costly board-to-board mated pair connectors, Qseven is the first standard to use the reasonably priced laptop internal MXM card socket, and is easy for carrier board routing with 230 pins—30% fewer than similar connectors—arranged on a 0.5 mm grid. This socket was originally used with fast laptop GPU cards and can therefore handle the high data transmission rates required by some interfaces such as PCI Express. Despite its small dimensions, it has an extremely robust construction and therefore is suitable for mobile applications.

The Qseven Computer-on-Module, unlike memory modules, is not held by the card socket itself, but is instead secured using four screws and a spacer (2.7 or 5 mm, depending on socket height). This type of mounting allows high shock and vibration specifications to be achieved.

Many board manufacturers around the world already produce Qseven modules, and the market adoption rate continues to climb rapidly due to the low-cost gold-plated card edge “fingers” that make it easy to install and upgrade over time. The mechanical outline of the processor board and mounting holes is shown in Figure 2.

Figure 2
Qseven outline drawing.

Module Added-Value

While the SoC is rich in features, embedded developers often have requirements for controllers not found in the base platform. The Qseven 2.0 specification allows additional interfaces beyond the processor’s I/O to come off the module, such as serial ports (UARTs) and a dedicated I2C controller. Sometimes carrier circuits are easier to design hanging off these interfaces.

Three processors in the G-Series SoC family are available in the Qseven form factor. The AMD GX-210HA 1.0 GHz dual core (1 Mbyte L2 cache) has a 9-watt thermal design power (TDP). The AMD GX-210JA 1.0 GHz dual core (same cache) has just a 6-watt TDP and an expected average power consumption of just 3 watts in many embedded applications. For extreme environments, the module is also available in an extended temperature range of -40° to +85°C featuring the AMD GX-209HA 1.0 GHz dual core (L2 cache 1 Mbyte). The SoC is designed to require 33% less power than previously available AMD G-Series processors.

Four single (x1) PCI Express Gen 2 Lanes, one USB 3.0 and five USB 2.0 ports, up to two SATA 3 Gbit/s ports and a Gigabit Ethernet interface allow for flexible system expansions with high data bandwidths. A microcontroller, on-module SATA SSD up to 64 Gbyte, ACPI 3.0 CPU power management and high-definition audio complete the rich feature set of the conga-QG Qseven module.

Embedded developers benefit from outstanding multimedia performance, excellent performance-per-watt ratios and flexible task distribution between CPU and GPU. The fanless module design is particularly suitable for cost-sensitive applications in the control and automation industry, digital gaming, communications infrastructures, and graphics-rich devices such as thin clients, digital information boards and medical-imaging equipment.

Thanks to onboard error-correction code (ECC) memory support, the module is also suitable for communications, safety-critical situations and applications in harsh environments. It can be equipped with up to 8 Gbyte ECC DDR3L onboard memory. ECC provides additional functions to check the data flow and adjust it as necessary in order to correct errors. The correction mode of this memory type can detect and correct both single and double bit errors. It therefore differs significantly from the so-called “parity bit,” where errors can be detected but not corrected.

The integrated AMD Radeon graphics supports DirectX 11.1, OpenGL 4.2 for fast 2D and 3D imaging, plus OpenCL 1.2 to execute program code with the integrated GPUs. Third-party tools are available that can generate code to run on the GPUs too, effectively increasing the number of processor cores available to the application.

The dedicated hardware, Universal Video Decoder 4.2, for the seamless processing of BluRays with HDCP (1080p) decodes H.264, MPEG4, VC-1, MPEG-2 video streams. The available display interfaces include single/dual channel 18/24-bit LVDS and DisplayPort 1.2, as well as DVI/HDMI 1.4a for the direct control of two independent displays. DisplayPort 1.2 also supports multi-stream transport (MTS), enabling the control of up to two displays per graphics port in daisy chain mode (Figure 3).

Figure 3
High-density conga-QC, a Qseven module with G-Series SoC.

Embedded developers finally get to have their cake and eat it too. The unprecedented graphics and CPU performance within a tiny footprint module is a novel solution to the almost insatiable need for graphics to drive next-gen user interfaces for equipment and instruments. The SoC platform is well-suited to increase performance while reducing system power and size in a broad range of markets including portable medical devices, industrial automation, digital signage, thin client terminals and electronic gaming machines.

congatec
San Diego, CA
(858) 457-2600
www.congatec.com