I'm lucky to have a high end GPU which with the Graphics preset = Low is giving me 99% of the time 75 fps.
I understand this will just be personal opinions but wandered - as the DK2 has a low(ish) resolution does anyone notice much benefit of running at Mid or High? If yes I might try turning up any recommended individual settings
I've been planning on writing a long post about settings, DK2 and reliable benchmarking, but RL is getting in the way.
The first decision is whether you can live with slight dips or lower than 75fps, keeping in mind that in my opinion it is likely to be near impossible to avoid occasional dips, especially in a demanding game like ED. So really the question becomes can you live with say drops to 65 fps in certain stations/ship combinations or the like? The answer to that will probably guide you more along what settings to change.
Though i've yet to perform exact tests it is my opinion that shadows, ambient occlusion and anti-aliasing are the three biggest contributers to overall performance. I've seen a large difference on my GTX970 between shadows off, low, medium and high, I think from memory at 2560x1440 screen size as much as 30 fps difference between low to high when viewing outside of a space station. The other two settings are both processor intensive (shader) and due to being full screen very dependent upon your resolution, so with running at HD they are likely to take up a good chunk of time each frame.
Myself my current settings are
Textures: High
ModelDrawDistance: High
Shadow Quality: Low/Medium
Bloom: Low ( was High)
Blur: On
AA: Off
Ambient Occlusion: Off
Environment Quality: High
FX Quality: High
Reflections Quality: High
Material Quality: Ultra
Admittedly this does not appear to provide a stable 75Hz and I suspect some tweaking will help, but I do find coming from a GTX465 where I had to have most stuff on Low that visually it looks so much nicer. I was also of the opinion that i'd rather have great visuals and occasional stuttering, but that was in the GTX465 which could never really manage 75Hz in many cases and I'm starting to wonder if being close to 75Hz but not hitting it is actually worse than being far away, as it does feel different now i'm on my GTX970.
I can't give any other recommendations other than going through each setting in turn and seeing if there is a marked difference in performance. This is something I want to do when I have time, but likely i'll do it in 2D mode as the relative differences in performance will be applicable to all uses. In other words i'm interested in seeing what settings have the greatest effect on framerate and which have minimal impact as I suspect some will have little effect to setting them to low other than spoiling the visuals.
I also want to detail about each setting trying to understand what its impacts might be. Obviously without seeing the code that is going to be easy, so will have to be educated guesses. So for example Texture Qulaity, with a 4GB GPU, and full PCI-e 3.0 x16 lanes of bandwidth I see no reason not too go below High. However whilst available memory should never be an issue, bandwidth might, especially if any 1 frame or couple of frames need to upload a large amount of asset data to the card. That will I believe result in a few frames of stutter.
For reference a x16 PCI-e 3.0 supports 16GB per second bandwidth (and I think thats both ways), which sounds like alot, but divide it by 75Hz and you are down into the 200MB per frame. Considering a 4096x4096 uncompressed texture with mipmaps is around 80Mb, it wouldn't take much to saturate that bandwidth in any one frame. If you drop to x8 lanes or are using PCI-e 2.0 then you get half that bandwidth, which is much more likely to get saturated. Of course this is mostly speculation on my part, i've never done any tests involving uploading such large amounts of data to a gpu in a single frame, but it seems logical. The only thing to consider is that if you are uploading new data to the gpu, then actually grabbing or creating that data itself may take longer than a frame and thus cause a stutter too. So while bandwidth may very occasionally cause issues, the processes that ended up need to push that amount of bandwidth themselves may be responsible for more substantial performance drop that frame.
I found this in one of the beta listings too which may be useful to know.
Added support for grouping scalable render features, so that a quality setting can be specified for the whole group and that translates automatically into per-feature quality settings. Grouped background quality and planet quality into the new 'Environment quality' setting, and also grouped light cones, lens flares and debris limit into 'FX quality'