Why does GPU performance tail off/decline with an artifact-free higher core speed?

[3t3P]

Another random thought popped into my head today that i've not seen an explanation for.

I'm sure we've all seen this, as well as reviewers comments.

You have the GPU at the limit of its overclocking headroom and you push it a bit more.

Your temps are ok

You have no artifacts

Yet game/benchmark performance is the same or even a bit worse

Why does this happen when the core is running faster?

 

[James24]

Limited by other factor? Memory bandwidth or CPU clock speed for example.

 

[eddyr]

It may be operating at a higher frequency but just not successfully or usefully. A selection of possible reasons from things I've picked up reading about over the years: You may be unable to feed the core quickly enough. Error correction/resolution in core memory/logic could be lowering useful performance. Gremlins. An architecture also tends to be fine tuned to operate at certain frequencies and complex cores tend to have many domains so you could be getting timing mismatches.

 

[Jakus]

Don't forget they have been using error correcting memory for a while, you might have got that right on the edge.......easy to test