Significant progress porting the SGP4 algorithm to a CUDA implementation. I can read in TLEs, initialize with all the SGP models and propagate forward in time.
I was able to propagate 29176 Satellites 1000 times with 30 second time steps in 10 seconds.
One of the things I must do next are optimizing the code. I ran it through the NVIDIA Profiler and received a terrible compute to memcpy ratio. That's mainly due to copying from device at every time step.
Compute Utilization: 0.2%
I plan on dramatically improving these scores by minimizing the memory copies to host and just drawing the points to screen instead.
Here's an example of one satellite. This was drawn using MATLAB.Share on Twitter Share on Facebook