It is common optimization technique to use shared "local" memory inside a compute shader to avoid unnecessary write to global memory (slow!)
https://developer.arm.com/documentation ... red-memory
upon reading this article, it claims that many ARM SOC's do not contain this high speed shared memory on the gpu, and that it is mapped onto system ram.
Is this the case with Pi5? Pi4? If so, what other optimization techniques are developers using compute shaders doing to overcome this?
https://developer.arm.com/documentation ... red-memory
upon reading this article, it claims that many ARM SOC's do not contain this high speed shared memory on the gpu, and that it is mapped onto system ram.
Is this the case with Pi5? Pi4? If so, what other optimization techniques are developers using compute shaders doing to overcome this?
Statistics: Posted by cpdev — Sat Nov 23, 2024 4:23 am — Replies 2 — Views 71