I have two CUDA applications. When I run the first a certain number of times, the CUDA driver stops working. Specifically, when I try to allocate memory during initialization I get an “out of memory” error message. Querying the amount of available memory using “cuMemGetInfo” shows that I have 200 Mb of free memory and my program only needs ~1Mb in total. Running the the second CUDA application results in no errors, even though running the first program had just given me and “out of memory” error.
Restarting the computer also fixes the problem, however this is annoying as I run the program often to debug it.
I’m using a Geforce 9400M in a MacBook Unibody running 10.6.2 and the 3.0 beta driver (same problem with 2.3a driver).
Any suggestion for solving this kind of problem? Thanks!