science:physics:nuclear-physics:shell-model:gpu-acceleration-of-the-tbme-calculation
Differences
This shows you the differences between two versions of the page.
| Both sides previous revisionPrevious revision | |||
| science:physics:nuclear-physics:shell-model:gpu-acceleration-of-the-tbme-calculation [2024/11/06 19:50] – Initial commit jon-dokuwiki | science:physics:nuclear-physics:shell-model:gpu-acceleration-of-the-tbme-calculation [2024/11/06 21:25] (current) – Add discussion jon-dokuwiki | ||
|---|---|---|---|
| Line 131: | Line 131: | ||
| As you can see, the speed-up is huge. The GPU code is 95 times faster than the CPU code. This is of course dependent on how good (or bad) I was at writing the code for the CPU. However, I have done several optimisations on the CPU code before using it as a basis for the GPU code. For example, using primitive data types for the basis states and pre-generating a bunch of indices. Optimising the GPU code further so that it only calculates the upper triangle improved the performance even further, now at 148 times faster than the CPU code. | As you can see, the speed-up is huge. The GPU code is 95 times faster than the CPU code. This is of course dependent on how good (or bad) I was at writing the code for the CPU. However, I have done several optimisations on the CPU code before using it as a basis for the GPU code. For example, using primitive data types for the basis states and pre-generating a bunch of indices. Optimising the GPU code further so that it only calculates the upper triangle improved the performance even further, now at 148 times faster than the CPU code. | ||
| + | |||
| + | ~~DISCUSSION~~ | ||
science/physics/nuclear-physics/shell-model/gpu-acceleration-of-the-tbme-calculation.1730919032.txt.gz · Last modified: 2024/11/06 19:50 by jon-dokuwiki