The attached graphs show a comparison of Shadow levels low, med, high and max settings with all other settings remaining unchanged. 2x GTX580 3GB Driver is Nvidia 296.10, 32xCSAA, Transparency Supersampling 8xSS, 16x AF MSI Afterburner 2.1.0 rFactor, all settings to maximum except for shadow. 25 AI. One lap of Spa with 67 F2 car 3rd gear @ a constant 7000 rpm. Suprisingly, all settings except "off" are almost equally costly in terms of fps loss. Performance hit appears to depend on the type or size of shadow with medium and high settings showing larger frame rate losses than max in some sections of the course. Generally, shadows seem to draw more slowly at higher settings. The effect is very pronounced at max setting even though fps may be similar to other shadow settings in some parts of the course. Further tuning of the graphics engine seems to be in order.
No Shadow FPS Performance Chart For completeness, attached is a comparison chart of no shadow and low shadow settings. The framerate hit even for low shadows is quite significant. Could someone from ISI comment on whether or not rendering methods have been optimized for multiple gpus? The NVidia SLI Practices Guide (Feb. 2011) describes some scenarios in which performance could suffer... http://developer.download.nvidia.com/whitepapers/2011/SLI_Best_Practices_2011_Feb.pdf
Interesting. Cross it off the list than. You should do this with all the settings. Texture quality and such. Something must be murdering frames. It would help if it wasn't limited to 60.
But this posts whole purpose was to show that it wasn't the case and that changing the shadows made very little difference.
Shadows, reflections and mirrors all cost surprisingly large amounts. Basically anything that involves multiple passes over the image before it's rendered, I suggest.
Yes, that seems to be what the graphs show. In fact, any shadows are costly in performance terms. The gpu graphs confirm this. With any level of shadows the gpus are maxed out on certain parts of the circuit where there are more shadows. This could be because of the way render targets are coded. The speed of multiple renders becomes visibly slower when increasing shadow levels. The same holds true for anti-aliasing. Any use of sparse grid super-sampling is almost as costly as turning on low level shadows.
I've never used supersampling in any game with any card. The performance hit is just too much. In my opinion you might want to try multi sample in another test. Again, just my opinion.
It looks as though the low shadow is offset to the right relative to the other shadow-types. Offsetting the low shadow back to superimpose the max etc shadows, the peaks and valleys appear to be approximately the same type although different magnitude although not nearly as low as I would have expected. Thanks for the info, always helpful to get real numbers