Why am I getting 10 fps in 400 point battle (graphicslib test mission) with GTX 1080 8Gb with graphicslib and gpu is utilized only 7%? Turning off every option in grahhicslib config brings back solid 60fps and gpu is utilized at 12%.
Specs:
- i5-2500k
- GTX 1080 8gb
- 16Gb ram
- ssd
vmparams are set correctly with 8gb allocated memory and starsector.exe and java.exe are forced to use gtx1080 in Nvidia control panel.
Your CPU, being 5 years older than your video card, is pretty heavily bottlenecked there. Note that Starsector is, for basically any modern video card, always going to be CPU-bottlenecked. You could be running a GTX 680 or something and see little to no difference in performance. On top of that, GraphicsLib basically tacks several additional renderers on top of the game (the ships are redrawn potentially three additional times each, not to mention all the shader buffers & other setup miscellany, and the lighting engine).
Between 10 and 60 fps (I kind of doubt those are the averages reported at the end of the run; you need to let it finish or the results are too unreliable), it's no surprise the one with 6x the frames pushes your GPU harder than the other.
I ran the benchmark at windowed 1080p with GraphicsLib and LazyLib alone, sounds enabled, with vsync off and the frame limiter set to 165. GraphicsLib had the deluxe preset.
Run length: 18.8 minutes
Average FPS: 51.6
Minimum FPS: 22.1
Average Frame Variance: 4.39 ms
Perceived Game Speed: 99%
Battle Size: 802
I ran again with GraphicsLib's shaders disabled (the other non-shader scripts were left on).
Run length: 20.1 minutes
Average FPS: 80.2
Minimum FPS: 28.1
Average Frame Variance: 3.15 ms
Perceived Game Speed: 100%
Battle Size: 762
If we consider the average frame rates and account for the slightly smaller battle size that the second run wound up with, we can conclude that GraphicsLib's Deluxe-preset shaders are eating up about 1/3 of my potential performance. A 33% reduction in FPS is more-or-less what anyone using GraphicsLib should expect. This is all CPU-limited, as I mentioned before.