CUDA: What is scattered write? -


The various CUDA demos in the CUDA SDK refer to "scattered writing" what is this scattered writing and why it is so good ? What is the opposite?

I am going to use the terminology of the CUDA here.

Scattered write means that you are writing from one thread to an arbitrary address (i.e. will not write threads of your warp constantly in memory, eg this frame- Conflicting with writing buffers, which are 2D-compatible, can be processed by hardware. They only wrote one unless long before the GPUs were available.

They Contrary to the code collected by a , there are jobs which Read the data from the place where it is stored, and collect them all before collecting the thread of the thread executed in a CIDD fashion on the collected data. However, the accumulated assemblies have become available on the GPU for long time through arbitrary texture.


Comments

Popular posts from this blog

MySql variables and php -

c# - How can I find out if there are windows above a control? -

javascript - Classic ASP "ExecuteGlobal" statement acting differently on two servers -