Load/Store UnitLSU
Units in a Streaming Multiprocessor that dispatch requests to load data from, or store data to, the GPU's memory subsystems.
Load/Store Units (LSUs) dispatch requests to load or store data to the GPU's memory subsystems.
For CUDA programmers, what matters most is which memory they talk to:
- Directly: the Streaming Multiprocessor's on-chip SRAM L1 data cache, the lowest level of the memory hierarchy.
- Indirectly: the off-chip, on-device global RAM, the highest level of that hierarchy.
Both levels belong to the memory hierarchy of the CUDA programming model.
Related#
- L1 Data Cache: the memory LSUs access directly
- GPU RAM: the memory LSUs reach indirectly
- Memory Coalescing: how access patterns affect memory traffic