An industry perspective from GE Intelligent Platforms
SBCs gain 256-bit vector processing
SBCs are key elements of military embedded computing subsystems and systems. They represent the lifeblood of every COTS vendor’s offerings, from the smallest computer on a module to the extensive, sophisticated offerings in 6U VME, CompactPCI, and VPX (VITA 46). Intel’s Core i7 processor family has become widely adopted by these COTS vendors for its ease of use, high performance, and good thermal characteristics, specifically in its mobile and embedded forms. The latest Core i7, based on Intel’s Sandy Bridge microarchitecture, heralds a new level of capability for embedded military computing applications. In addition to improvements in processor performance, it incorporates a high-end, on-chip graphics processor plus the first mainstream implementation of a 256-bit Single Instruction Multiple Data (SIMD) vector processor, known as Advanced Vector Extensions (AVX).
Sandy Bridge is the second generation of Intel’s Core i7 processor. It uses a different microarchitecture from its Arrandale predecessor, with Sandy Bridge typically offering a 20 percent increase in processing performance, improved power/system management, plus a big boost to the on-chip graphics capability. Sandy Bridge uses a highly parallel internal ring bus architecture to interconnect the processor cores, L3 cache, and the new graphics processor. Also on this ring is the system agent (effectively replacing the Northbridge) that provides the external DDR3 memory interface and 16x PCI Express 2.0 for all off-chip I/O connections.
Unlike NVIDIA’s or ATI’s separate, high-performance Graphics Processor Units (GPUs) using massive arrays of programmable SIMD processor cores, Sandy Bridge’s graphics engine implements the majority of its complex graphics functions in dedicated hardware. However, it also includes an array of SIMD shader processors plus larger register files to enable the dispatch of more instruction threads without starvation by the register file. Key to improved performance is the graphics processor’s equal L3 cache access rights, equivalent to that of any processor cores on the internal ring bus.
The 256-bit vector processor
By processing two vectors in a single execution cycle, each Sandy Bridge core’s 256-bit vector processing engine offers twice the data processing capability of the earlier Core i7’s AVX processor. This performance boost for floating-point applications will renew pressure on vendors such as Freescale (which recently announced reintroduction of the AltiVec to their QorIQ line) and others. These vendors will need to keep up, as vector processing finds more applications in wireless base stations, communications, security, and signals intelligence. In addition to these high-volume applications in relatively benign environments, there are many rugged, real-time military applications – such as ground mobile or airborne radar and video sensor suites – that will benefit from the performance of AVX-based Digital Signal Processing (DSP).
Despite the performance potential of AVX, Core i7 was not designed solely for vector processing applications. Its primary role is clearly for general-purpose computing, server, or embedded applications across the entire spectrum of potential uses. In rugged applications these roles might range from equipment control and monitoring, fire control, and communications, to complex mission computing and sensor processing. SBCs for rugged, embedded applications are equally general purpose in nature, typically offering maximum flexibility and customization through the addition of PMC/XMC mezzanine modules, networking, and many onboard I/O interfaces.
Core i7’s system agent (Northbridge) provides PCI Express 2.0 to simplify the connection of external interfaces such as Serial ATA (SATA), Universal Serial Bus (USB), serial ports, and network ports. PCI Express also offers SBC vendors the opportunity to differentiate their product offerings by providing, for example, avionics interfaces or multiple network or serial fabric choices on a single board. GE Intelligent Platforms’ SBC624 6U VPX SBC (Figure 1) uses the dual- or quad-core, embedded 26xx or 27xx Core i7 device running at 2.1 GHz. The SBC also adds InfiniBand to the onboard I/O options as a low-latency, high-bandwidth, switched-fabric alternative to Ethernet for multicomputing and server applications.
Range of DSP capability
The introduction of Sandy Bridge with its 256-bit vector processing capability to a broad range of SBC types, from the physically very compact 3U VPX profile upwards, will further encourage the assimilation of many DSP applications directly onto the SBC itself. Similarly, at the other extreme, multicompute DSP engines will benefit from quad-core devices, either reducing board count or increasing performance, to create a seamless range of DSP capability using the same AVX technology and software tools. Currently based on 32 nm process technology, a further boost in performance and reduction in power requirements is assured by the planned introduction of 22 nm process technology for the next generation of the Core i7 family in early 2012.