May 2019
Intermediate to advanced
452 pages
12h 16m
English
Efficiently manipulating the number of threads, blocks, and grids while addressing a GPU program kernel allows us to make the most of the GPU device's computational throughput. The following diagram demonstrates the components of a grid. We can see that it consists of blocks. A single block has also been zoomed into and illustrated to show what it contains as well.
The components of Block (1, 1) are 12 threads, each represented as arrows. Note that each block in the grid (shown in green) also include 12 such arrows to represent the individual threads. Every block in the grid is structured in this manner:

Read now
Unlock full access