What are the delays of the GPU?

Question

I can find delays in terms of both ns and CPU between the processor and its cache, main memory, etc.

But it’s so hard to find similar information about modern GPUs.

Does anyone know about GPU delays, especially the delay between the modern nvidia GPU (GF110 or newer) and their memory, thanks.

GPU memory has a lot more bandwidth, but what about their delays?

I heard that delays for the GPU are as high as for the processor, so basically making the large bandwidth pretty much pointless for many general purpose computing tasks, I just need to confirm this.

Answer 1

Since suppliers did not reveal all the architectural details, the researchers used reverse engineering to demystify the GPU architecture. See This article Demystifying the GPU microarchitecture using microfunketing and other documents that cite this (note that this is not my article). I copied their results in the image below.