Block and thread configuration

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I have tried to do the following and I get an invalid configuration error.

dim3 grid( 5, 1, 1);
dim3 threads( 10, 128, 1);

while this works fine

dim3 grid( 5, 1, 1);
dim3 threads( 10, 1, 1);

Why does the top not work? Is the answer dependant on the size of the kernel I run? The GPU docs say I need to try and make as many threads as I can to get the best perfomance, but this is not launching!

2

The maximum number of threads in a block (as written in the manual…) is 512.
You are using 1280 in the first case when you get the configuration error.

3

Oh okay, thank you.