nbody-Simulation problem


I slightly changed the code of the n-body interaction problem (i actually used the functions from the pdf-file http://developer.download.nvidia.com/compu…gems3_ch31.pdf), but i have problems with running the code for many particles (about 100.000).
I tryed to calculate the amount of memory that is needed on the card, but that should be no problem.
It should be 44Byte for the velocities and 44 Byte for the accelerations (respectively 32Byte per particle), so 32 Byte times 100.000 particles is about 3 MByte plus some bytes needed by the threads.
For just 50.000 particles everything runs fine.

Any ideas, what the problem could be?

I forgot to describe the error… The system halts for a second, then the display gets black and a windows error message says that some graphic problem occurd.

Your kernel is running too long, so the watchdog kicks in and resets your graphics card.

Ok, but the run for 50.000 particles needs like a few milli-seconds, so the runtime should not be a problem, right?

First-chance exception at 0x774242eb in matrixMul.exe: Microsoft C++ exception: cudaError_enum at memory location 0x0012fbc0…
The program ‘[5264] matrixMul.exe: Native’ has exited with code 1 (0x1).

That’s the error message i get in the Microsoft Visual C++ console, when i start the programm via the “F5”-hotkey…

In my programm i just calculate the forces, so there is no graphical output like in the “nbody” sample programm - just a bit command promt output.