Maximum Texture size

Accelerated Computing CUDA CUDA Programming and Performance
1

I read in cuda programming guide 2.0beta
A.1.1 Specifications for Compute Capability 1.0 page 78

1. For a texture reference bound to a one-dimensional CUDA array, the maximum
width is 2^13;
2. For a texture reference bound to a two-dimensional CUDA array, the maximum
width is 2^16 and the maximum height is 2^15;
3. For a texture reference bound to a three-dimensional CUDA array, the
maximum width is 2^11, the maximum height is 2^11, and the maximum depth is
2^11;
this is my questions.
what is the unit of width, height, depth? [byte???]
For a texture reference bound to a one-dimensional CUDA array, the maximum
width is 2^13 ->the CUDA array size is 8KB???
if the unit is [KB]so the CUDA array size is 8MB???
in two cases is not fit with my test, if once of two cases is exactly, i think that my program has a problem.

in my test
VS 2005, cuda2.0, geforce 8800GT
the maximum size of CUDA array I can bound with texture is 64KB. it is too small isn’t it?

2

As far as I know the unit is element. So it might be float, float2, float4, etc

3

I had assumed it was bytes - it seemed safer.

4

hello, i’m quoting this because i’m having the same trouble…

in the programming guide, the max width for a one dimensional texture bound to a cudaArray is 2^(13).

If we assume it’s the max number of elements, and my cudaArray is of float4, my max dimension should be 8192 x sizeof(float4).

My code works without probem until 2^(14) = 16384 elements, using float4. If i try to define a bigger cudaArray, it says me “invalid argument”.

Is this the bound?

my card is a geForce gtx295

CUDA Driver 2.3