This is heavily dependent on the type of application you are running. Basically, if you have a heavy computation that you’re repeating, and you use the same input parameters often, you may be able to calculate that value a single time and then look up the result for future calculations using the same inputs. This trades off compute for memory accesses.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| price for lookup tables | 7 | 7286 | February 12, 2008 | |
| Fast Implementation of (Small-)Table Lookup | 14 | 955 | April 26, 2024 | |
| Miscalculation in Nsight Compute 2025.3 | 0 | 25 | August 9, 2025 | |
| Using Nsight Compute to Inspect your Kernels | 2 | 1715 | August 31, 2020 | |
| How to use Nsight Compute Profiling Results | 7 | 1171 | December 17, 2019 | |
| Memory Table Missing in ncu-ui only for H100 | 5 | 206 | April 30, 2025 | |
| Question about Compute Workload Analysis | 3 | 1113 | October 12, 2021 | |
| optimazing the programming of cuda | 0 | 4370 | May 30, 2010 | |
| No results from Python tools in Nsight Compute | 3 | 436 | October 12, 2021 | |
| Can nsight compute be used for training stream analysis? | 2 | 457 | November 15, 2023 |