walsufnir said:
SvennoJ said:
The problem with dynamic resolution is that it introduces extra lag. You need to buffer extra frames to make up for time lost on a frame that went over budget. It's not simple to calculate how long a frame render will take with different resolutions. That's probably why the 1360x1080 is fixed in multiplayer and 1920x1080 is only used in some places in the campaign. Probably hard coded to avoid lag from just in time adjustments.
Wipeout HD dynamically reduces resolution but drops frames first and then goes back to 60fps while lowering resolution. GT4 had a different approach, first rendering the even lines then the odd lines. In busy scenes in GT4 the resolution basically halves as only half the lines are updated for a small time. KZ:SF multiplayer does that with columns, except every other frame and tries to interpolate the missing pixels.
All approaches have disadvantages and introduce extra overhead.
Another approach could be to render the scene at half resolution first. Then fill in full resolution blocks starting from the center or where the action is at. That would work especially well for racing games as you're always looking down the road and the edges of the screen are usually smeared in motion blur anyway. Of course it's still more efficient to render 1 optimized pass. Or do what it seems AW did, lock the reolution based on where you are.
|
If we didn't have all the repeated talk about how important resolution is, devs probably wouldn't look for solutions like this. I think in general this is a nice approach and some research will be spent into this this gen. Again, I am more ok with this than dropping frames. How difficult it is for devs to make good predictions and other stuff I am glad to read when they release ppts of their stuff :)
|
Blame LCD screens for the resolution debate. Native resolution simply looks better than scaling. I would be fine with dropping effects to maintain frame rate instead of resolution. Ofcourse smoke, particles or distant shadows / detail suddenly disappearing would be ugly too.
It's a paradox, the more time you spend on accurately calculating and predicting how much time the next frame render will take, the more time you take away from actually rendering the scene. Last gen the problem was tackled by removing geometry and details beforehand while profiling the game. That got just as much repeated talk, ps3 missing extra grass, less items in the background, less alpha effects etc. Now it's more focussed on resolution.
Lowering the resolution affects the whole scene, rendering a few things less is only noticeable if you look at them side by side. Ofcourse the gap is a lot bigger this time around and simply removing a few objects might not be enough. And a less busy scene would not look next gen enough on screenshots :/
