I have build several test machines based around AMD A10-7800 and A10-7850 APUs targeting high density realtime graphics rendering.
We love the machines, except for one apparent issue which is driving my team crazy.
Our application always has issues with GPU throttling, as latency is critical for us (we output broadcast video in realtime with few buffers), and there are other processes that need time before and after rendering, so the GPU loads are very spiky.
The response in most throttling systems is to downclock as they are not fully loaded. This causes big latency problems for us.
With most of our AMD machines we can work around this by using an ugly modification in CCC utilizing a hand modified profile and AMD GPU Overdrive. This is not the best, however it works.
However with these new machines, we cannot get any access to GPU overdrive, it is simply missing from CCC. the Performance section contains only a 'CPU Power' Section.
This is a major issue for what would otherwise be a great system for us. IF we cannot lock the GPU clock, then they are unusable in low latency situations.
Of course what we would greatly prefer is a proper interface to allow us to control for GPU clock/throttling directly in our application, however we are not current aware of one for AMD.
We are testing on Win7 x64 professional SP1, and currently with drivers 14.4 and 14.7RC3 - and have checked usual obvious things such as .NET4.5, etc.
Am I being stupid and missing something to get this needed functionality?
Is there a better way to control these things?
On NVidia systems we simply create an application profile for our application, and tell it to prefer maximum performance, job done.
Please does anyone have a solution for this?