I'm running an OpenCL application on a GPU linux server with some HD7970 cards in it, and since it is completely ALU bound, I'd like to downclock the memory to 400-500MHz, since this would reduce power consumption. However, it seems that all methods I try fail when trying it to reduce below (shader frequency - 150MHz). Why is this? How should I work around this?
All methods that I find either cannot decrease it below the shaderfreq-150, or are windows-only (like MSI afterburner). And since my other topic here doesn't seem to catch any answer, I'm not able to run windows on this machine to modify these values.
Does anyone know any way to do this? It's a waste of energy to run OpenCL on a device that provides way higher clock speeds than what I need...
im also waiting for this feature in linux drivers. and as i don't like currently implemented behavior - im simply NOT buying 7970 cards now. why i must pay so much for hardware that didn't do what i want?!? when i can have much cheaper hardware that DO, and lose not so much in efficiency. anyway, i will only migrate from 5870 cards when AMD finally give us ability to downclock 7970 cards memory in linux - thats a mandatory option for me to buy this cards.
btw, i tried to write letters / create tickets to AMD support regarding this issue - with very detailed description, but still not received adequate answer - only 'captain obvious' answers, which looks for me like they don't want to answer or just not going to implement in 7xxx cards drivers frequency core / memory control like in 5xxx series.
however - people using windows and msi afterburner CAN downclock memory successfully in 7970.
still im very disappointed with this moment for a months also.
PS. AMD, please give us possibility to make this changes in card settings and monitor more parameters like vrm temperature in linux!