All geek questions in one place

Permanent memory versus texture memory and global memory in CUDA

I am trying to find the differences between permanent memory and texture memory versus global memory in CUDA.

I can find the following relevant articles, but I can not find the answer to my question

global and shared memory in CUDA

Using global and persistent memory in CUDA

An article that looks at the performance implications of all three: http://forum.beyond3d.com/showthread.php?t=52510

+6
source share
1 answer

Read only memory:

Kernel constants and arguments are stored here.

Slow, but with a cache (8 kb)

Permanent memory optimized for broadcast

Texture memory:

Cache optimized for 2D spatial access pattern

Reading has some advantages, such as address modes and interpolation, that can be used at no extra cost.

Global memory:

Slow and unencrypted (1.0), cached (2.0)

Requires sequential and aligned 16-byte read and write for fast (combined read / write)

Source: http://www.cvg.ethz.ch/teaching/2011spring/gpgpu/cuda_memory.pdf

+6
source

Source: https://habr.com/ru/post/902618/

More articles:


All Articles