It's been a while since I've seen something like this, but with the advent of the new xbox I got reminded of an old SGI adage that stated "Sustained 60 is better than variable 300." The commment was in regard to framerates and how the best thing to do was to always maintain a framerate vs. varying it. The real question is if there's an easy way to do that with OpenGL? I know on my old Octane everything seemed to just default to that, but I never looked into how to code it. Thoughts?

Quote:Original post by TheSteve
It's been a while since I've seen something like this, but with the advent of the new xbox I got reminded of an old SGI adage that stated "Sustained 60 is better than variable 300." The commment was in regard to framerates and how the best thing to do was to always maintain a framerate vs. varying it. The real question is if there's an easy way to do that with OpenGL? I know on my old Octane everything seemed to just default to that, but I never looked into how to code it. Thoughts?

Depends on the situation, and what they are describing.


They might be referring to the number of frames per second with frames being divided up nearly equally over time. Your physics and AI should operate at a fixed rate. That was not always true of graphics systems, nor is it true of many non-professional games.

These days a good design separates the display from everything else, so that's not a problem.

As long as the 'variable' display is divided equally over time, then it's fine to render as fast as possible up to the refresh rate of the screen. Interpolate the display based on the time between the two physics steps and you'll be fine.



If they're referring to variable display as not divided equally over time, then they are correct. A worst-case example for variable 300 FPS display time is that one frame takes 0.999 seconds, then the remaining 299 frames are generated in the remaining 0.001 seconds. That's a BAD situation, but less extreme versions usually happen.

That situtaion is the reason you should be measuring the time of frames,rather than FPS. Your in-game benchmarks should have the average, min, and max over a given time frame. If your min and max get too far apart, you've got a problem.


frob.

Well, the framerate definetly wasn't based on the monitor. Almost all SGI programs on all different monitors ran at 60 fps, even if you had the lastest monitor and an incredible Onyx 2 (which was the bomb at the time). As far as it being variable with time, I'm not sure about that. All I know is that the big thing back in the 1995-1999 glory days of SGI was that they always maintained a 60fps sustained rate and nothing more/less. Their argument, which I believe is true from watching the xbox, is that anything variable is going to ultimately look worse than sustained. You can even see this in action to this day if you go and look at an xbox 360 running one of their new games. At 30 fps, it looks damn amazing. Not just in terms of "nice graphics," but in terms of overall smoothness and consistency. I think there's a more aggressive and perhaps a great deal more complicated way to do it. Hopefuly there's a big IRIX buff on this forum.

The general premise is entirely true. Constant 30fps 'feels' much smoother than a program that runs at a mean of 45fps, but has a high std deviation (> 5-8 perhaps?).

There isn't any point in rendering at 300fps. While it may make the owner of a 7800 GTX feel good, I think you're burning my laptop's battery down.

I disagree with most people at gamedev, in that I prefer to use Vsync. It keeps the framerate very consistant, while keeping resource usage down. But my programs are multithreaded, and use lots of background CPU time (on-the-fly resource loading for visualizing multi-terabyte data sets). The extra CPU freed up when the rendering thread sleeps (during a blocked buffer swap) makes a huge difference on single processor machines.