I recently updated my CUDA from 2 beta to 2 final, and of course i installed 177.84 driver. Then, my program became dead (i mean no response). Memory copy is ok, but global memory access in kernels would make my program dead.
Changing back CUDA 2 beta or 1.1 doesn’t make it work.
But when I installed old driver such as 175.X or 174.x, it finally worked well with CUDA 1.1.
Does anyone know why?
My equipment is 8800GT
Thanks a lot.
Unless you can provide a test app which reproduces this problem, its unlikely that anyone can speculate on what is causing this problem.
I can’t…
what i only can do is discription about that problem.
in our program, we combine MFC with CUDA, and use openGL to show our results.
both I and my colleague encounter the same problem (of course, she also update the driver)
i guess it only happened in kernel, because i did some simple experiment.
for exapmle
Load some bmp in h_memory1
copy h_memory1 to d_memory1 (device memory)
copy d_memory1 to back h_memory2
access some unrelated global memory (d_memory2) in kernel
show h_memory2 using opengl
it will work. we can see the image
but if i change its order
Load some bmp in h_memory1
access some unrelated global memory (d_memory2) in kernel
copy h_memory1 to d_memory1 (device memory)
copy d_memory1 to back h_memory2
show h_memory2 using opengl
then, we cannot see anything…
BTW before we used 177.84, we tried 177.89 first.
after discovering this problem, we also tried 177.92, but it didn’t solve this problem
only 169.21, 174.X,175.X can work in our program(with CUDA 1.1)
Now,our program only can work in cuda 1.1
my equipment is 8800GT, and my colleague is 8600GTS
i hope i provide some useful information.
BTW, we can run any sdk project, whether our program works or not.
That’s the most werid part.
maybe it is somehow related to this topic?
[url=“http://forums.nvidia.com/index.php?showtopic=75466”]http://forums.nvidia.com/index.php?showtopic=75466[/url]
could it be you are hitting some timeouts through the performance drops of the new drivers?