65k+ 1D texture

Accelerated Computing CUDA CUDA Programming and Performance
1

I have a 200M elements of read-only grid data. Obviously the “texture memory” model is the best for me, because global device memory is not cached and has a brutal access time… also I don’t wanna write values, just to read them… so I need a “texture”.
The problem is that I wanna access as an 1D array but I see the maximum grid size blocks is 65k x 65k. Each block has 512 elements maximum… so I can store only:

65k * 512 = 33M elements in my “texture grid”

Any way to do this without dynamic streaming the grid data(PCI-Express)? Must be a solution because if my gf8800 has 768Mb I should be able to access a brutal 1D 768M data array!

2

er…
Excuse me, but how is grid size limit related to texture size?
1D texture’s maximal size is (1<<27) elements or sth, that should be enough for you.