Advanced Micro Devices (AMD) has launched a full range of OpenCL-compatible heterogeneous parallel processing products, including traditional x86 CPU, GPU and integrated APU chips, as well as embedded GPU chips/MXM modules and add-on cards, etc.
Not only has AMD won acclaim with its AMD Turbo CORE, AMD Eyefinity, and AMD Dual Graphics technologies, AMD is also committed to establishing a comprehensive heterogeneous multi-core industrial alliance and ecosystem with the midstream and downstream hardware and software industry sectors to bring embedded systems into the new realm of heterogeneous parallel processing.
Careful observation of the embedded market and continued R&D investments
Arun Iyengar, corporate vice president and general manager of the AMD Embedded Solutions Group, indicated that the scope of applications for the embedded market is very broad. The sales cycle for the entire product line is considerably longer than in other markets. During the early stages of the cycle, suppliers would have 3 months to send samples and selections to customers. After the design is adopted (design win) the SI certification stage is approximately 12 to 18 months. Finally, moving a product from engineering prototype to mass production for the customers would stretch for another 36 to 60 months. This entire process could take up to a decade for defense military institutions.
The embedded market is more focused on the optimal heat dissipation rate by adopting low power consumption, no cooling fan, higher functionality integration, and more compact designs to create the smallest sized products with optimal energy efficiency in order to lower the total cost of ownership (TCO). The customers hope suppliers can fully understand its product strengths and weaknesses, provide solutions, and tailor the various solutions to respond to the specific needs of the market.
Iyengar indicated that AMD is carefully observing the development of the entire embedded market and continues to invest in its R&D. Since June of 2012, AMD's Embedded Solutions Division has been organized into an independent group. In October of 2012, CEO Rory Read indicated that the embedded market has become the group's rapid-growing emerging business; and 2013 is regarded as the first year for AMD embedded solutions.
AMD, a pioneer in the embedded market for many years
Iyengar stressed that AMD has been a pioneer in the embedded market for numerous years. From the Am386 and Am486 processors in 1991, the Am5x86 in 1995; the Elan SC520 in 1999; the acquisition of Alchemy Semiconductor and National Semiconductor (NS) by the AMD Geode processor business group between 2002 and 2003; the addition of the 64-bit AMD Opteron, AMD Athlon, AMD Turion, and Mobile AMD Sempron processors into the embedded market; and the ATI Radeon E2400 GPU launched in 2008, and only reached EOL this July. In 2009, AMD launched the ATI Radeon E4690 GPU with 2D and 3D multimedia graphics performance and multi-screen output capacity that is suitable for a wide variety of embedded applications in gaming consoles, consumer electronic products, digital billboards, and industrial control boards, as well as for applications in the medical, defense, and aerospace fields. AMD also introduced the BGA flip chip packaging that is suitable for small size ASB1 processor systems.
AMD released the Embedded G-Series APU platform in 2011, and launched the Embedded R-Series APU platform that focuses on high-end graphic performance in 2012. In 2013, AMD will launch embedded applications that are integrated with high-performance graphics and computing performances as well as the SoC silicon chip program that can connect a wide range of I/O peripherals together as one.
AMD will continue to provide product supply, open-source operating system support, drivers, software libraries, development platforms, and technical support for its embedded market products for 5 to 7 years. AMD will also provide diverse product selection from the initial silicon circuit design stage to the engineering sample stage, and will eventually link and intimately collaborate with the existing industrial ecosystem.
Market is at a heterogeneous multicore emerging stage
Iyengar used a three-stage CPU evolution schematic diagram to illustrate his point. The single-core CPU has long since reached its development bottleneck. The homogenous system would also reach the power, parallel oriented software, and performance bottlenecks as the number of cores has increased to a number that cannot continue to be sustained. The heterogeneous system is at a burgeoning stage. The data parallel mechanism enabled by energy efficient GPUs can ensure the continued improvement of performance, although in extending performance, the traditional programming models must be subverted and changed.
Iyengar also mentioned the developmental history from the early single-core microprocessors to the homogenous multi-core systems, and the advent of the heterogeneous multi-core system. The AMD APU combines the serial data processing of the traditional CPU with the parallel data processing of the traditional AMD Radeon GPU into one. Products like the AMD Embedded G-Series APU (G-T16R APU) have the average power consumption rate of only 2.3W and an I/O controller chip that consumes less than 1W; it is suitable for extremely small and convenient-to-carry embedded mobile devices without a cooling fan. The AMD Embedded R-series APU comprises the quad-core x86 core and the AMD Radeon 7000 GPU core with 384 parallel arithmetic units, which provide high performance that can rival independent GPU graphics performance with DirectX 11 hardware acceleration specifications and can use the OpenCL and DirectCompute software to provide program solutions to enhance the performance of embedded platform computing to another level.
Embedded product line and key technologies of AMD
The AMD embedded product line is divided into different series. The R-Series APU provides appropriate performance, energy efficiency, and above high definition (HD) visual experiences. The G-Series APU emphasizes low power consumption as well as unprecedented GPU integration, and is suitable for mini and fanless system designs. In addition, the AMD Radeon GPU provides amplified 3D and multimedia performance specifically for embedded systems, and its product delivery cycle is ensured for up to five years.
Iyengar introduced the first key technology of the AMD APU, the AMD Turbo Core. Taking the AMD R-464L as an example, the x86 core clocks at 2,300 MHz and its GPU clocks at 496MHz. Under the general load balanced state of the x86 core and its GPU, each clock remains unchanged. When the system enters the multi-threaded parallel processing state, the clock of the x86 core would dynamically increase by a maximum of 39% (up to 3,200 MHz), but the clock of the GPU remained unchanged. When implementing programs that emphasize 3D graphic processing, the clock of the x86 core remained at 2,300MHz, but the clock of its GPU would dynamically increase by a maximum of 38% (685MHz). Depending on the needs of the load, the APU can provide bidirectional and dynamic clock adjustments for the GPU and the x86 core.
The other key technology is the AMD Eyefinity technology. The AMD R-series APU can provide external connections and has parallel display ports for four monitors. When connected to an external PCIe add-on graphics card with six DisplayPort (DP) v1.2 interfaces, the system can expand its access capacityto up to ten monitors. The general embedded systems designs based on the Intel CPU can only enable either the internal GPU or the external Radeon GPU chip even with the additional add-on standalone GPU chip. However, the AMD APU is equipped with AMD Dual Graphics technology that enables the internal Radeon7000 GPU core to operate in parallel with the external Radeon E6460/E6760 GPU chip to further compact and enhance the 3D and multimedia graphics computing performance.
Iyengar indicated that the AMD G- and R-series APUs can be applied to digital billboards, casino gaming machines, IP-TVs, x86 set-top boxes, medical applications, POS terminals and phone booth kiosks/automatic ticketing systems, thin clients, video conferencing, human machine interfaces (HMI), and industrial control applications, as well as communications machines/infrastructures.
Unifying the industry to form the heterogeneous system architecture (HSA) foundation for the promotion of the parallel processing technology
Iyengar stated that parallel computing can be applied to video and image processing such as machine recognition, medical imaging, intelligent billboards, and video surveillance; digital signal processing such as telecommunications and national defense; network traffic packet processing; and green high performance computer (HPC) applications. AMD offers a full series of OpenCL-compatible heterogeneous parallel processing application products ranging from the low power consumption G-Series APU that is designed to consume only 4.5W to 18W of power with 80GFLOPs of maximum parallel computing performance to the high-performance R-Series designed to consume 17W to 35W of power and provide 500GFLOPs of parallel computing performance.
The AMD Radeon GPU series, including the AMD Radeon E6460 GPU, adopts a BGA package and the MXM module or the PCIe add-on card format, and is designed with a power consumption rate of 20W in conjunction with the 25GB/s bandwidth GDDR5 memory in order to deliver its 192GFLOPs floating-point performance. The E6760 GPU adopts a BGA package, the MXM module, or the PCIe add-on card formats; and is designed with the power consumption rate of 35W in conjunction with the 51GB/s bandwidth GDDR5 memory in order to deliver its 576GFLOPs floating-point performance. The E6970GPU adopts the MXM module, and is designed with a power consumption rate of 95W in conjunction with the 115GB/s bandwidth GDDR5 memory in order to deliver its 1.3 TFLOPs floating-point performance.
In June 2012, AMD brought together ARM, Imagination, MediaTek, Texas Instruments (TI), Samsung, and Qualcomm to form the nonprofit HSA Foundation in order to establish a complete industrial chain from the Silicon Intellectual Property (Silicon IP) to the software developers and to promote the standardization of the heterogeneous system architecture.
AMD has unified many of its industry partners such as Sage Electronic Engineering, LLC.; ALT Software; CORELIS; Core Avionics & Industrial, LLC.; Express Logic; ELTAN; Green Hills Software; Texas Multicore Technologies, Inc. (TMT); Fluendo; Viosoft; etc.; to cooperate in finding a solution that can ensure prompt R&D and system integration for the OEM manufacturers. For example, Gizmo, the AMD Embedded G-Series APU based small development platform, has the circuit board area of only 4x4 inches and provides I/O peripheral interfaces such as USB ports, 5.1 channel connectors, VGA ports, PS/2 keyboard and mouse ports, and SATA ports. The Gizmo development platform was originally developed under collaboration by AMD, Sage, Viosoft, and Texas Multicore; provides Windows, Linux, and RTOS operating system and associated drivers; and its back-end software and technical support is provided by the nonprofit technology community GizmoSphere.
Iyengar also indicated that AMD and ARM are long-term and strategic cooperation partners with close and well established working relationships. AMD will add the 64-bit ARM authorized ARM processor chips into its production line, and continue to supply the x86 architecture chips, while increasing investments in software development. He concluded that AMD will continue to expand investments in the embedded market, as well as provide superior products that focus on key market applications while satisfying the needs of the consumers and the market.
Arun Iyengar, corporate vice president and general manager of the AMD Embedded Solutions Group