Cuda and 3D blocks

I would like to share some questions with the community.

To begin, I have a GTX 260 896Mo and I need to work on 3D blocks of 100 Mo Max size.
I want to take a 3D block, to read it, perform a calculus, and write in an other 3D block.

First, I tried a simple CudaMalloc considering this block as a 1D array.
But, it seems that this allocation is not working. Is this because the 3D block size is too big for linear memory ?

Secondly, I tried cudaArray, but this type is not defined in cuda documentation. So, how can we access and write to data from cudaArray ?

  • am i obliged to cut my block because of his size ?
  • what kind of structure can i use to read and write 3D blocks ?

If someone have a clear explanation on how can i start with 100 Mo 3D blocks, it would be really great !

Thank you for reading.

There should be no problem with allocating 100 megabytes of linear memory. What error message do you get, and do you have multiple CUDA-capable devices in your computer?

Thank you for your concern. I have just one GeForce GTX 260, beta2 sdk and 177.41 driver.

In fact, when i execute my program, i have black screen and vista say that my drivers have been restaured.
That’s weird because sometimes, it works perfectly.

I thought it was the bloc size because with a smaller one, there’s no problem at all.

Moreover, i have 3 overlapped loops in the code and i read there was some problem with flow control instructions.

Any idea?