I've been working on an OpenGL program, and I've noticed something very odd.  About 1 in every 10 times I run the program, I get a significant performance increase, but if I exit the program and run it again, it's back to normal (slower) speed.  It seems to happen regardless of the program's actual load on the system, and without a change in the build - just running the program a second or third time sometimes triggers the great performance increase.

Looking at the task manager, during normal execution, the program takes ~50% of the CPU, as may be expected when running a core at capacity.  However, on the odd high performance run, the CPU is only running at 5% or so, indicating greater than 10x performance increase, since it will also run at a higher framerate.

This indicates to me that perhaps OpenGL's hardware acceleration query is often failing when it shouldn't, and it's reverting to a fallback most of the time, with the odd high performance sessions being OpenGL running as it should.

My system is:
2.4 GHz Intel Core 2 Duo with
2 GB of RAM and
NVIDIA 8800GTX graphics card running on
Windows Vista

we don't fallback to P3D, it's probably more likely to be a strange timing/frameRate issue--i.e. System.currentTimeMillis() being inaccurate and returning bad numbers to make things run slowly perhaps.

endShape() is where all the magic happens, so the majority of cpu will be burned inside endShape().

but if you can find a way to reproduce this reliably, i'd love to find it and track it down.