[leoben]

Not sure if this is the right place to open this topic up, so please move it if necessary.

I installed a fresh copy of FS2, media VPs a few days ago, and I noticed a really bad stutter. I uploaded a video here, that illustrates it quite well.

https://amd.box.com/s/tv6ye4cba2gq4avn3lwf

A couple of interesting points:

- It seems to appear randomly. Sometimes when I see the lightshafts, sometimes when big objects are on the screen, sometimes it makes no sense
- it also seems to go away if I hit afterburner. Sounds stupid, but this is consistently reproducible - every single time i hit the afterburner, the stutter goes away immediately, and returns immediately once the afterburner runs out of juice.
- This happens with 3.7.1 BP build. I didn't check other builds yet.

I have a Radeon R9 290X, with the openGL driver version 6.14.10.12614, running 25x16 with everything turned on.

Happy to provide more detail if it helps. Any ideas to start off what this issue is? Any suggestions which effect to turn off to debug?

[leoben]

Some additional info.

This new GPU uses fully dynamic clock switching. It will attempt to operate (and does remarkably well) at a set temperature and at a set fan speed, and it will accommodate that by dynamically switching between clockstates. Since FSO is not terribly hard on a very modern (and this being the fastest Graphics chip on the planet right now) GPU, it has no trouble maintaining 60FPS, even at very high resolutions. However, when the stutter occurs, I hear the GPU fan speeding up to the given limit that I define, which means that suddenly the workload has jumped magnitudes. When the stutter goes away (pretty random), fan turns itself back to normal, and the workload seems to be back to normal as well.

[The E]

This is something I have seen myself, but have so far been unable to track down. I am not even sure if it is a general issue or a GCN one; as you've seen, there is nothing in the workload being dispatched that can account for changes on such a scale.
Our built-in frame profiling code doesn't seem to be able to pick up the cause either, leading me to believe that it's at a lower level than we have access to.

[leoben]

I see. I just checked - with all features turned off, the stuttering is still occurring. BTW - would you guys consider building a 'Mantle' path for FSO?

[The E]

IF Mantle is cross-platform. IF Mantle can be integrated easily. Right now, given that we do not have any sort of reference regarding the API, there are no plans. That may change once we know more about it.

Please keep in mind that we a) do this for fun in our spare time, b) we have to support a wide range of hardware and OS combinations and c) Maintaining two code paths that do more or less the same thing is something we really have little interest in.

[leoben]

Fair enough. Next week there'll be a lot more info on Mantle. I'll get you some white papers if you're interested.

BTW this stutter is really bad. I'm looking at GPU-Z logs, and at times these stutters occur, my GPU load goes from 20% to 100%. The weird part is that while i afterburn in game, the stutter goes away, and as soon as it's out, the stutter comes back.

I also noticed that is has something to do with certain objects being in the field of view with an awkward combination of distance to those objects. Further away, the stutter seems to occur more frequently.