Control of GPU powermanagment

[caseycullen]

@caseyculen

The problem which i find now is that swapping of Radeon1950PRO to RadeonHD6970 didn't change a single FPS in some of my test cases (and also in GLXgears too). But Xeno says that for him in doom3 he has speed increase for sure. What it can mean is that for some "small examples" power management just didn't raise resources (so we can't see the full speed of the gfx card in some tests). Why i am so sure about that because we have had the same problem on AmigaOS4 before: i installed Radeon RX instead of RadeonHD and things start to be even worse. And in end, we find out that this is power management was on "base" and didn't raise resources (dunno why maybe no proper DMA or something). And this very much familiar with what we have.

Now what we need to be 100% sure that is to be able to control power management manually as we can now on AmigaOS4 via some prefs app. Need to find how to do it on our side on Linux. There should be 3 states: low base and performance. So we should even "hear" how "fan" will move less and fast.

For example, i never heard on all my tests that fans start to be louder they still same speed.

Besides testing, I'm not sure what the everyday purpose of restricting the graphics card performance on a desktop system would be. This kind of function is usually used with laptop GPUs to increase battery life.

I don't think using fan loudness or RPM speed is a good indicator of GPU performance. Even on my gaming PC when playing a graphically-intensive AAA game, the graphics card fan does not constantly spin up to 100%. If AmigaOS does this for a "high performance" mode, I'd suspect it was to avoid complexity in the driver, like programming thermal sensor interfaces and fan performance curves.

I see clear performance differences using different graphics cards in my X5000 running Fienix. For example:
Quake HD Redux at 1280x720 on "High" with "RT World" on
RV710 (HD4550) - 7 FPS minimum / 20 FPS average / 33 FPS maximum
RV940 (HD6870) - 29 FPS minimum / 59 FPS average / 116 FPS maximum

It would be helpful to know what specific benchmarks you've used and what the results were.

[kas1e]

@caseycullen

It would be helpful to know what specific benchmarks you've used and what the results were.

For the moment i tested only Irrlicht Engine examples, and glxgears.

The FPS output i have in Irrlicht Engine examples for both, X1950 and HD6970 are almost 1:1 the same, with no speedup. And HD6970 is even slower in comparison with RadenHD 250 some old mobile variants (and i mean MUCH slower) than on amigaos4.

So i think there is something wrong, and to understand what, i need to find out the way to control power management manually.

[kas1e]

@All
is it possible to somehow measure the maximum possible speed of CopyToVram and CopyFromVram on our PPC Linux? On amgaos4 we have GfxBench which can show all the details for us. For testing purposes, we also need to measure it under Linux as well.