Brochure

10
high performance embedded computing
applications
Radar
Developers of today’s radars demand that their processing systems be founded
upon the principles of Modular Open System Architecture. Designs must be
scalable, open architecture and capable of sustainment for the long term with
technology insertion plans.
In response, GE has adopted OpenVPX as the primary form-factor, using widely
adopted processors from Intel and NVIDIA, connecting processing clusters with
standard interconnects like Ethernet and InfiniBand, and providing support for
Linux, Open Fabrics Enterprise Distribution, MPI, DDS and so on.
Signal Processing
By using the rapid prototyping capabilities of AXISView along with
quick implementation of radar algorithms with VSIPL and VSIPL++,
a signal processing system can readily be modeled to determine
how many processors are needed. GE can produce standard or
custom radar backend processors scalable to dozens of TFLOPS.
Such systems can extend from sensor input via standard interfaces
such as serial FPDP and 10GbE, through processing on clusters
of Intel processors with optional NVIDIA GPGPU co-processing, to
output to a backend system via standard Gigabit Ethernet.
Electronic Warfare
Electronic Countermeasure Systems are typically defined by their need
for low-latency processing. To date this has often required FPGAs to
meet the latency constraints.
Now, GPUs can be considered for such applications. GPUdirect allows
sensor data to be transferred straight into GPU memory, bypassing the
multiple copies and host processor involvement that were previously
required. Testing has shown a reduction in latency of better than 25x,
opening up the use of GPUs in applications that were previously not
candidates. Development time and cost are also reduced.