March 30, 2010 at 7:14 pm
http://www.hpcsweden.se/files/RadarSignalProcessingwithGraphicsProcessors.pdf
In a near 130 page document, it is quite hard to condense everything to a few lines. But basically it looks at using GPGPU acceleration for radar filtering algorithms.
From the benchmark results, it is obvious that the CUDA GPU is substantially faster than the CPU and CPU code used, for radar relevant data sizes. For the most demanding benchmarks, a speedup of close to a factor of 100 is achieved.
It should be noted this work was completed on an FX4800, which has been replaced with the Fermi uarch. While crap for gaming, Fermi will accelerate the speedups to >200x vs. a CPU.
Thus, in the not too distant future, no manned aircraft will be invisible to cutting edge SAM radar within 50 miles, probably more like 100 miles. The aircraft in question will need to fall back onto the traditional strengths of a combat aircraft, namely speed and maneuverability.
The F-22 has both properties in abundance. The F-35 does not.
I am assuming the PAK-FA will be much closer to the F-22 than it will be to the F-35 (I think that is a fairly safe assumption).