Hello,

I am currently using one compute shader, which calculates everything, i can calculate all deferred lighting with stable cascaded shadow mapping in one pass. I am using UAV texture to save results.

Because UAVs are 5.0 only ( RWTextures, also i heard, they are somehow slow becuase it's doing some checks ... ), i am thinking of using screenquad. My question is, is screenquad faster and better solution ? Perhaps i was just overloaded with expectations that CS is cool and fast :)

Also i had a cool feel when i can control thread count :) but pixel shader works same right ? every pixel runs parallel ?

thank you for answers, opinions ;-)

pixelshader tends to be faster for me (~10%), when you don't use anything special from the compute part. but compute shader can be way faster if you use shared memory.

Ok, thank you.

So, i will test it then. I just did not want to do blindly something, so it's easier to ask :)

In my experience a compute shader is always slower unless you're doing something that makes use of a compute shader's advantages (extra flexibility, shared memory, etc.). I have seen the IHV's recommend this as well, and they also recommend against switching between compute and graphics many times during a frame.