VPX is one of the fastest growing embedded systems form factors today. Generally, when I think of VPX, I envision applications like a tactical airborne radar or an advanced ship-borne navigation system.
However: the possibilities of such a flexible platform as VPX are near limitless. Coming from an instrumentation background, when I was introduced to OpenVPX I was somewhat in awe of the flexibility and capabilities of the standard. VPX features a backplane interconnect that is extremely user-definable and flexible, with multiple possibilities.
This flexibility has led some engineering companies to use the platform in some novel and unexpected ways. This is particularly true with FPGA payload modules, as their configurability stacks on the already flexible VPX platform. An example of a COTS FPGA product is the recently announced VP869, which is an upgrade to the Abaco VP868. The VP869 features two Virtex Ultrascale+ FPGAs and an embedded FPGA/ multi-processor system-on-chip (MPSoC). The design of the VP868 and VP869 products was intended to meet many applications for DSP and other applications.
Super Computers and 6U VPX
An interesting application we’ve seen is the use of the VP868 and VP869 for high performance supercomputing applications. In the space of big data processing, there has been a lot of research in the past five years around graph processor architectures. Graph processing is an efficient distributed computing solution that involves nodes and edges and that is particularly good at solving rapid data reduction, sparse matrices, and various search algorithms. Applications include calculating the fastest route on a map, routing autonomous vehicles, and DNA analysis.
The key architectural requirement of a graph processor is a balanced processor fabric with high speed interconnects. The VP869 can be used for just such an architecture when combined with the flexibility of OpenVPX and the also recently announced RTM630 from Abaco. The RTM630 is a COTS product that includes 48 bidirectional fiber optic links matched to Xilinx SerDes I/O. Each lane is capable of up to 8Gb/s for up to 384Gb/s for a single VP869/RTM630 combination. This COTS architecture could be used to build a large system of up to 128 streaming nodes.
Now, there aren’t many people in the world doing such high performance processing. However, it is inspiring to see applications such as these emerge as a result of the flexibility of VPX. As indicated in the title of my blog, OpenVPX has opened doors for engineers and scientists to solve some very tough problems that would have been economically prohibitive to build as a custom system.
Having a master’s degree in computer architecture and having worked at MIT in the CSAIL department in my early career, I confess that I’m a bit of a computer architecture fan. In my blogs, I often cite how Abaco is helping the allied warfighter with our technology, but I also find inspiration in the fact that we are also helping solve some of the toughest computer architecture challenges for scientists and researchers - and the VP869 and the RTM630 are a part of that.