Skip to main content

Advertisement

Table 1 Differences between memory types in CUDA

From: Protein alignment algorithms with an efficient backtracking routine on multiple GPUs

Memory type Located on chip Cached Access Scope
Registers yes n/a R/W Thread
Local no no/yes* R/W Thread
Shared yes n/a R/W Block
Global no no/yes* R/W Program
Constant no yes R Program
Texture no yes R Program
  1. Differences between memory types in CUDA (n/a stands for „not applicable”, letter R for „read” and letter W for „write”, * - caching depends on compute capability). For more information see CUDA Best Practices Guide [24].