Motherboard selection for extended support of industrial computer systems
Industrial and military users of servers or workstations based on ATX form factor iX86 motherboards require excellent performance as well as extended life and long-term support. Choosing the right motherboard goes beyond selecting a product for long lifecycle. The selection of the right chipset as well as the processor is key to obtaining the best performance and upgradeability over the lifespan of the system. The following discussion compares various embedded solutions that are available for different motherboards, expected lifespan, and upgradeability.
Commercial computer systems based on available iX86 architecture technology have a lifespan of 6 months before a newer version is available. The industrial or military user that plans to deploy the same architecture for a period of 2 years or longer is expecting long-term availability of all parts to ensure a constant supply of product. All products are eventually going to reach their end of life, but it should be a planned process that allows time for upgrades. Military systems in particular are specified and documented around a particular hardware set. The cost of changing the documentation and requalifying hardware to replace an EOL motherboard can run hundreds and thousands of dollars.
Select processor and chipset for optimum performance
Selection of the processor and the supporting chipset is based on more than just raw performance numbers. The I/O and memory requirements must also be considered. If a design requires only limited I/O or minimal I/O performance, a Pentium solution is fine. However, if multiple channels of x16 PCIe are required, the best solution is a single Xeon or, for the ultimate performance, dual Intel Xeon processors.
Intel originally provided a three-chip solution for Pentium implementations: a processor, north bridge, and south bridge. The north bridge provided the memory interface and PCI interface, and the south bridge implemented the low-level I/O features. Intel announced its new architecture "core" processors in 2008 that implemented the memory controller in the processor, which improved performance. The south bridge was renamed and a new chip providing I/O function was released as the Platform Controller Hub (PCH). The PCH also included the display controllers as well as the I/O functions.
In 2010, with the release of the Q57 series of PCH chips, Intel provided long-term support for the PCH parts. Figure 1 is the timeline for the currently available series of PCH chips. Each PCH part that Intel lists for embedded applications has a 5- to 7-year lifecycle. Intel will announce the end of life and last buying date about 18 months before the actual end of life. Each motherboard vendor and system provider will work with Intel to determine when the end of life for the motherboard or system will be. Intel publishes its embedded roadmap, which provides a timeline for anticipated availability of the products. You may hear a processor is "on the embedded roadmap," which typically indicates a product with multi-year availability. Each processor is typically available in a variety of speeds, but only one or two speeds for that particular processor will be on the embedded roadmap.
As new I/O specifications are announced in the industry, Intel releases new PCH parts to implement the faster I/O capabilities. As an example, the Q77 added USB 3.0 support and the Q87 added support for PCIe 3.0. In addition, each version of PCH is matched with a release of Pentium core processors.
Considering the generation of core processors that utilized the Q57 as first generation product, Table 1 outlines the processor technology used with each PCH family. The challenge is choosing the right combination of processor and I/O performance for the application. The Q57 PCH is still available, but without USB 3.0, PCIe 3.0, and the latest display technology, the part would not be recommended for new designs. However, since the Q57 part will still be available for 2 more years, it will still be used as a replacement product for an existing application.
Selection of motherboard for I/O performance
For industrial and military applications the I/O requirements can go beyond the standard USB, storage devices, and display connections. Many applications utilize coprocessors or specialized I/O add-in cards that require PCI or PCI express connections for the best performance. The choice of a system with the right selection of slots is essential. Often legacy I/O boards are available with only PCI interfaces and must be accommodated in new systems.
PCI/PCIe bandwidth is controlled by the specification revision number supported by the processor and chipset. The highest bandwidth available currently is provided by the PCIe 3.0 specification. Bandwidth is also controlled by the data path width. The maximum PCIe data size is 16 lanes; the smallest is 1 lane. Most common are x1, x4, x8, and x16. The number of PCIe expansion slots is dependent on the number of ports provided by the PCH as well as the processor. The Pentium Core and Xeon processors provide PCIe controllers that are internal to the processor. Tightly coupled with the processor and the memory, these processor-based PCIe controllers provide higher bandwidth than the PCH-based controller.
Figure 2 is a comparison of the PCI features of the chipsets and processors available for long- term supported devices. This figure shows the evolution of I/O support from the second to fourth generation of Intel Pentium products. It also shows the higher availability of 16- or 8-lane PCIe solutions using a Xeon processor. Many applications, such as gas and oil exploration, image processing, and large data mining, can benefit from high-performance coprocessors that utilize the 16-lane PCIe interface for improved memory bandwidth. With the need for more than one coprocessor solution, multiple 16- or 8-lane PCIe add-in slots are becoming more useful in the overall system design.
Systems now collect massive amounts of data, such as from high-resolution 360-deg airborne surveillance cameras, and that data often needs to be processed and presented to the operator in real time. Multiple graphical processor units (GPU) coprocessors such as Tesla boards can provide the required horsepower, but they require high-bandwidth interfaces with the system board to be effective. Processing oil exploration seismic data is another data-intensive application that benefits from multiple GPU boards.
As an example, the system shown in Figure 3 is a dual Xeon industrial computer with four x16 PCIe 3.0 expansion slots that are double-spaced allowing up to four GPU accelerator coprocessor cards to be installed for unrivaled performance. This system is providing mainframe computing performance in a rugged rackmount computer that can be installed in a transit case and hauled anywhere in the world. Add in a rackmount keyboard, display, and a UPS and you’ve got a complete computing station. And, because the components were carefully selected for long-term availability, identical systems compatible with existing software will be available for many years.
It is important to look beyond the processor clock speed and memory size when specifying a system for an industrial application. The required I/O performance and number of expansion ports are just as important as the processing performance. As most industrial applications have long-term operation requirements, the availability of long-term support is also critical.
Anthony T. Bowers is senior applications engineer/project engineer responsible for all custom and semi-custom chassis as well as display products at Chassis Plans, LLC. Anthony holds a masters certificate in program management from Villanova and has over 30 years experience in product development and engineering.
– Edited by Joy Chang, digital project manager, Control Engineering, firstname.lastname@example.org