Military electronic systems must take their share of punishment, since reliable operation is often a matter of life or death. Considering the number of different vendors contributing to those systems, just getting different computer boards and input/output (I/O) modules to work together can be a challenge.
That’s why the OpenVPX interconnect standard was created for single-board computers (SBCs) and other commercial-off-the-shelf (COTS) electronic subsystem modules. Established in 2009, it aims for complete interoperability among all compliant systems and subsystems for military systems where real-time, high-data-rate operation is critical.
The OpenVPX Industry Working Group (IWG), an organization of defense contractors assembled by Mercury Computer Systems, created the OpenVPX specification in 2009. IWG, which comprises VME International Trade Association (VITA) members, is now the VITA 65 standards working group.
The VPX Marketing Alliance (VMA) supports the VITA 65 standards working group. VMA’s goal is to spread awareness of the value that VPX and OpenVPX interoperability brings to critical embedded system and subsystem applications.
The OpenVPX standards are often called the best of the VMEbus standards, building on those prior specifications with an eye for ever-enhanced performance. While the VMEbus format has proven effective for military and aerospace systems for many years, the steady increase in processing speeds in those systems had created a need for an interface standard with increased bandwidth.
OpenVPX replaces the parallel-bus architecture of VMEbus with multiple serial buses that support faster data rates while also minimizing the amount of signal routing required in a system. An OpenVPX backplane supports speeds of 3.125, 5.0, or 6.250 Gbaud/s.
OpenVPX maintains the mechanical specifications of earlier standards, such as VITA 46, designing modules into mainly 3U and 6U Eurocard form factors. It defines profiles for board modules and backplane slots so similar modules will work within certain defined slot configurations.
The backplane configurations define compatible modules as well as information on data rate, routing topology, and the type of switching fabric. The OpenVPX standard is designed to ensure that any card module conforming to the standard, no matter which manufacturer it’s from, can be used seamlessly in the same backplane as any card module from any other manufacturer.
The VITA working group included not only board-level suppliers such as Curtiss-Wright, Elma Electronic, GE Intelligent Platforms, Kontron, and Mercury Computer Systems, but also major systems contractors such as Boeing and Northrop Grumman. VITA standards were created to support high-speed serial switched fabric interconnections, notably in multiprocessor, multiuser systems that might require real-time support of sensors, communications, and control functions.
OpenVPX was established as an enhanced form of existing VITA standards (it’s essentially the VITA 65 standard, which is based on VITA 46) for designers to achieve full interoperability with their different modules and boards within a system while also maintaining some flexibility to respond to future (unforeseen) requirements.
VITA, which began in 1984, is an incorporated, nonprofit organization. Its standards group, the VITA Standards Organization (VSO), is an accredited American National Standards Institute (ANSI) developer of standards. The organization has been involved with standards for systems ranging from medical imaging to space launch control.
Earlier this year, VITA announced ANSI’s ratification of the second edition of the OpenVPX system specifications, under the heading of the ANSI/VITA 65.0-2012 standards. As new applications emerge, modifications may be needed to the basic OpenVPX framework along with the addition of profiles.
This latest update to OpenVPX, which was planned, adds profiles for payload, peripheral, switch slots, and backplanes in support of the InfiniBand protocol. It also adds VITA 67 coaxial (blind-mate RF) connectors to the OpenVPX specifications.
The VPX 67 connectors accept flexible and semirigid cables with coaxial contacts, using contacts on a 0.240-in. centerline for high isolation between connectors and channels. The VPX 67 RF connectors feature float-mounted jacks for reliable ground contacts, with contact float of 0.079 in. to allow blind-mating of connectors.
The VITA 65 working group is constantly reviewing the OpenVPX/VITA 65 specifications as well as profile candidates for additions to the standard to make OpenVPX an evolving standard adaptable to changing technology.
“As planned in the original OpenVPX effort, we knew that the specification would need room to grow the technology and meet future industry needs without impacting current projects. This revision has taken steps to make additions possible without disturbing existing content,” said Valerie Andrew, chair of the VPX Marketing Alliance.
VITA maintains a directory of OpenVPX products on its site, with about 300 products from different manufacturers currently listed (under the “VPX” heading). The support for the OpenVPX standard has been impressive, with new products announced almost on a weekly basis.
Earlier this year, Curtiss-Wright Controls Defense Solutions (CWCDS) introduced its VPX3-1257 single-board computer (SBC), its first OpenVPX SBC based on the quad-core third-generation Intel Core i7-3612QE microprocessor from Intel Corp. The microprocessor is designed for an improved thermal profile compared to earlier models for enhanced performance and reliability over the harsh environmental conditions faced by military and aerospace applications.
“CWCDS is proud to have built one of our industry’s leading Intel Core i7 processor-based product families. This third-generation Intel Core processor allows us to extend that family of products by creating a 3U OpenVPX SBC with a thermal footprint never before obtained with earlier-generation processors,” said Lynn Bamford, senior vice president and general manager for CWCDS.
Bamford added that having a solid partner like Intel will also encourage users concerned with supporting long-lifetime applications. “With Intel’s commitment to long life-cycle supply, this advanced low-power processing engine will drive a new generation of higher-performance and power-efficient compute-intensive applications,” Bamford said.
“By announcing longer-term availability of this high-performing, integrated processor family, Intel is giving our customers options for development of innovative and dependable long-term solutions,” said Matt Lamgham, director of product marketing for Intel’s Intelligent Systems Group.
The VPX3-1257 is a 3U OpenVPX SBC with a wide set of features, including on-board I/O, x8 PCI Express Gen2 fabric, and XMC mezzanine module expansion (Fig. 1). Since it’s based on the Intel Core i7-3612QE processor, the SBC can operate according to the requirements of reduced size, weight, and power (SWaP) applications.
The SBC is built for use in air- and conduction-cooled aerospace and defense applications. It’s designed to fill computing requirements in unmanned aerial vehicles (UAVs), unmanned ground vehicles (UGVs), tactical aircraft, and armored vehicles.
The VPX3-1257 is available with 8 or 16 Gbytes of 1333-MHz DDR2 SDRAM and a variety of high-speed I/O, including dual Gigabit Ethernet, USB 2.0 ports, DVI, SATA, and an XMC site supported by with eight lanes of PCI Express. The company also plans a x16 PCI Express variant of the SBC, the VPX3-1267. Programs based on the Windows WES7, Linux, and VxWorks operating systems will support both.
CWCDS is also enhancing the performance of its CHAMP-AV8 OpenVPX DSP engine with the addition of a pair of third-generation Intel Core i7-3612QE microprocessors (Fig. 2). This processor is expected to improve the power-per-watt performance of the 6U OpenVPX DSP at elevated temperatures by about 50% compared to DSPs based on the second-generation Intel Core i7-2715QE microprocessor.
The latest CHAMP DSPs will also support the emerging Gen2 PCIe-to-sRIO protocol conversion bridging capability from Integrated Device Technology, bringing the advantages of Serial RapidIO switch fabrics to Intel Core i7 microprocessor embedded computing applications.
The CHAMP-AV8 OpenVPX DSP engine is available in a number of different air- and conduction-cooled configurations, suitable for signal-intelligence (SIGINT) and radar systems where power efficiency is critical. The DSP board’s pair of quad-core processors employs the Intel Advanced Vector Extensions (Intel AVX) 256-bit instruction set for performance to 269 GFLOPS.
The OpenVPX DSP includes a DDR3 memory subsystem with peak bandwidth of 21 Gbytes/s. It can be equipped with 8-Gbyte flash memory and as much as 16-Gbyte SDRAM (8 Gbytes per processor). The CHAMP-AV8 boasts as much as 32-Gbyte/s fabric performance, with support for Gen2 SRIO and Gen2 PCI Express interfaces.
In addition to drawing upon its own expertise for its OpenVPX modules, CWCDS is licensing patented air-flow-through (AFT) cooling technology from Northrop Grumman Corp. to effectively dissipate heat generated within the compact modules.
The AFT technology uses a compact heat exchanger to dramatically increase the cooling efficiency of removable electronic modules such as VPX and OpenVPX cards. The AFT modules are housed in rugged shell-like structures that allow cooling without exposing their electronic components and circuit boards directly to air, eliminating the risk of contamination by airborne particles.
Sliding air seals at the inlet and outlet of the AFT cards enable the modules to be removed and replaced in the field. The technology also provides a reliable thermal path for any heat sources in a VPX or OpenVPX module, enabling CWCDS’s AFT-based modules to handle thermal densities as high as 200 W per system slot (Fig. 3). Each ATF card has a heat frame that cooling air passes through.
Because each module’s thermal path is isolated, each card has its own cooling air inlet and exhaust channels. Each of the high-power components on an AFT board is interfaced to the AFT heat frame through a conductive, flexible gap pad. On both the inlet and the exhaust sides of the card a gasket mounted inside the chassis seals the card’s internal air passage to the chassis side walls. The seals prevent air from being blown into the chassis and protect the internal electronics from the harsh external environment.
“This innovation opens the door to developing more powerful, rugged electronic systems across the military and commercial electronics fields,” says Pat Antkowiak, vice president and general manager of Northrop Grumman’s Advanced Concepts and Technologies Division. “This improvement in a key method of cooling electronic modules can serve a wide variety of applications.”
CWCDS also offers a 12-slot, 9U development chassis for designers of 3U OpenVPX systems. The rack-mount RME9XC meets VITA 65 power and cooling requirements for air-cooled, 75-W, 3U OpenVPX modules. It supports OpenVPX backplanes with switch fabric data rates to 6.25 Gbytes and can be equipped with power supplies to 2600 W for development use. The RME9XC measures 15.69 by 18.96 by 19.53 in. and weighs 77 lb.
Back To The Future
One of the biggest supporters of the OpenVPX standard is one of the first, Mercury Computer Systems. The company has relentlessly developed new modules based on the interconnect standard, including its recently announced 6U OpenVPX GSC6201 general-purpose graphics processing unit (GPGPU) card (Fig. 4).
Mercury’s StreamDirect technique efficiently delivers streams of sensor data directly to coprocessors, such as the GSC6201, improving performance by a factor of three over standard GPGPU cards. The StreamDirect technology also allows direct communication of data from a sensor to a coprocessor, such as a GPGPU, without intermediate storage in a central processing unit (CPU), eliminating this copy step and speeding the data transfer.
By optimizing the data transfers from the I/O sensors to the GPGPUs, the 6U OpenVPX model GSC6201 can support more than 10 TFLOPS processing capability. The company had deployed the innovative StreamDirect technology previously in a number of programs but is offering it in a standard OpenVPX unit for the first time.
“The GSC6201 is our third-generation OpenVPX MXM-based GPGPU carrier card, building upon our eight -ear track record of deployed GPGPU solutions,” said Scott Thieret, technical director at Mercury Computer Systems.
“With StreamDirect, we can now configure GSC6201-based systems with three times the performance of previous-generation systems by eliminating intermediate data store-and-forward steps and enabling sensors to communicate directly with Nvidia GPGPUs. Additionally, the combination of StreamDirect and the GSC6201 allows multiple GPGPU carrier cards to be hosted by a single CPU, significantly improving the overall system SWaP and GFLOPS/W,” he explained.
The GSC6201 uses the Mobile PCI Express Module (MXM) graphics card form factor developed by Nvidia to combine fast graphics processing speed with the benefits of an OpenVPX module. It can also be upgraded other graphics architectures, including Nvidia’s Kepler MXM.
“The latest fighter jets and drone aircraft incorporate new sensors that demand huge amounts of real-time processing, making them ideally suited for Nvidia’s newest GPGPUs,” said Nvidia’s vice president of global automotive and embedded solutions, Vineet Gupta.
“Mercury Computer Systems’ StreamDirect unleashes the potential of multi-GPGPU systems, bringing leading-edge performance to defense-based systems when using the latest Nvidia GPGPUs,” Gupta said.
Mercury also recently announced the availability of its server-class Ensemble Series 6U OpenVPX HDS6601 compute blade based on the Intel Xeon E5-2600 microprocessor family. Each HDS6601 blade incorporates two eight-core Intel Xeon E5-2648L processors to form a module with 16 computer cores configured as a symmetric multiprocessing (SMP) cluster (Fig. 5). This represents the first 16-way, 32-thread SMP server blade compliant with the OpenVPX standard.
Obviously, this is an enormous amount of computing power in a compact configuration. Any of an HDS6601 cluster’s CPUs can access any part of the 64-Gbyte on-board memory directly. Mercury’s software and other switch-fabric modules support the modules to form a powerful open-architecture computing platform that’s available with the benefits of OpenVPX’s Serial RapidIO and 10-Gbit Ethernet configurations.
“Key to enabling our server-class rugged processor module are Mercury’s mechanical packaging innovations for the Intel Xeon processor E5-2648L and the development of new rugged large memory packaging methods,” said Steve Patterson, vice president of defense product management at Mercury Computer Systems.
“And when the server-class processing capability is configured with Mercury’s GPGPU blades and mobile Intel processor-based products, our customers can configure solutions that quadruple the SWaP performance of currently deployed applications,” Patterson said.
“The Intel Xeon processor E5-2600 family is ideal for high-end processing due to the SMP model supported across multiple devices and is enabled by Intel QuickPath Interconnect chip-to-chip bus technology,” said Steve Price, director of marketing for Intel’s Communications Infrastructure Division.
“Combined with the Intel Advanced Vector Extensions processing capability on each SMP core, this creates a high-performance SMP (symmetric multiprocessing) cluster that performs well in ruggedized servers for high-end embedded real-time applications,” Price also noted. Those applications include advanced signal-processing chores in radar and SIGINT systems.
Trends In OpenVPX
If there is any one dominant trend in current OpenVPX card designs, it is the move to higher-power Intel microprocessors to gain computing power with improved thermal management. Yet another founding member of the OpenVPX initiative following this trend, GE Intelligent Platforms, recently announced its latest OpenVPX SBC. The SBC325 builds on the capabilities of a third-generation Intel Core quad-core microprocessor in addition to x16 PCI Express GPGPU connectivity (Fig. 6). The 3U SBC ideally suits intelligence, surveillance, and reconnaissance (ISR) systems for radar and for command/control platforms.
“The improved thermal performance of the new Intel processor helps to deliver improved response times and throughput and reduce the thermal footprint of a solution—or reduce the number of boards required in a system, minimizing its size and weight,” said Rod Rice, general manager for military and aerospace products at GE Intelligent Platforms.
The SBC325 is available in air-cooled and conduction-cooled versions, with a 2.1-GHz Intel Core i7-3612QE processor with IntelAVX, as much as 8 Gbytes of DDR3 memory, and as much as 32 Gbytes of solid-state memory.
Also, module designers are finding ways to enhance OpenVPX product performance without stepping outside the boundaries of the VITA 65 specifications, such as the efforts by Curtiss-Wright Controls Defense Solutions to license Northrop Grumman’s cooling technology.
Similarly, German OpenVPX module supplier Kontron is offering an automatic test technology as an option for all of the company’s current and future 3U and 6U VPX, OpenVPX, and VME processor boards. The Power-on Built-in Test (PBIT) measurement routine is stored in nonvolatile memory on the card.
Although it is hosted on a module’s CPU board, PBIT checks components that are part of the main board, such as sensors and memory, but also peripheral components connected across the backplane. It can even check for the proper operation of components connected to SATA and USB ports. This capability enables the PBIT test option to greatly increase the security of an OpenVPX device by preventing system boot-up when it identifies a problem, such as an unauthorized memory stick connected to the USB port.
Kontron’s PBIT functionality features a sophisticated learning mode in which it captures the most intricate details of a system configuration. By storing the configuration of a system when it is working properly, the PBIT routine can check system health upon the next boot-up cycle, even identifying transient errors, such as bad cables and connectors, by comparing the previously captured system configuration with the currently identified configuration. The company offers a free white paper about the PBIT test routine on its Web site (http://us.kontron.com/PBIT_whitepaper).
While many OpenVPX products are CPUs, coprocessors, or SBCs, Pentek uses the compact format for an eight-channel OpenVPX beamforming network: its 53661 software-defined radio (SDR) board. The 3U OpenVPX board includes four 200-MHz 16-bit analog-to-digital converters (ADCs); a timing, clock, and synchronization section; and a Virtex-6 FPGA from Xilinx.
Each of the FPGA’s four digital-downcoverter (DDC) intellectual-property (IP) cores can accept samples from any of the four ADCs. The DDCs can provide downconverted baseband signals over bandwidths from 2.5 kHz to 80 MHz. They also have programmable gain and phase-shift controls accessible via the VPX backplane. When used as a beamforming network, one ADC is assigned to each DDC.
Each DDC includes a power meter at its output to calculate downconverted signal power. Each power meter has a threshold detector capable of generating a system interrupt for signal power above an upper threshold limit or below a lower threshold limit to help simplify signal monitoring.
The FPGA also features a summation block to add the four DDC outputs for the channel combining required for beamforming. The summation block is based on Xilinx’s Aurora gigabit serial protocol. The Aurora interface accepts a propagated sum on one 4X input port and delivers the new propagated sum on a 4X output port including contributions from the four onboard channels.
At a bit rate of 3.125 Gbits/s, each 4X link can transfer data at 1.25 Gbytes/s. The OpenVPX module’s PCI Express x4 2.5-Gbit/s serial interface provides a 1-Gbyte/s connection to the CPU for programming the DDCs and beamforming parameters. As with many OpenVPX modules, the model 53661 beam-forming card is available in air-cooled and conduction-cooled versions.
If the OpenVPX format was initially criticized, it was simply because it was yet another board standard for military users where existing standards, such as VPX and VME, had served adequately for many years. But electrical or mechanical performance levels have never been an issue for these standards.
OpenVPX brings real value to military and other users because it guarantees compatibility between modules and backplanes as well as modules and modules. In a vital industrial application, where downtime is lost revenue, that compatibility has a price tag. But in the field, in hostile military environments, that compatibility can be a matter of life and death.