Comment on Anon makes a modern game

• addie@feddit.uk ⁨2⁩ ⁨days⁩ ago

  There’s two kinds of motion blur, really - camera based, and model based. Camera-based requires calculating one motion vector for the whole screen, which is basically free. Model-based requires projecting the motion of each vertex of the model in the projected view; one matrix multiply per vector is not ‘expensive’ on a modern graphics card. Depth of field requires comparing the depth buffer, which you’ll have already created as part of rendering, and then taking several ‘taps’ around each point on the screen to calculate the blur for the ‘focus distance’ compared to the actual distance. The final image post-processing will generally process the whole screen anyway, so you’re just throwing a couple of extra steps in for the two effects.

  Now, what does it save you? If your engine is using TAA (temporal anti-aliasing) then that’s performed by ‘twitching’ the camera a tiny amount (less than a pixel) every frame. If nothing’s moving, then you can merge the last several frames to get a really high-quality anti-alias; all the detail that wouldn’t be caught with a ‘completely static’ camera will be captured, and the result looks great. But things do move; if you recalculate ‘where things were’ then you can get a reasonable idea of what colour ought to be at each pixel. Since we need to calculate all the movement vectors to do that, then using the same info gives us the motion blur data ‘for free’ - we can add a little blur in post-processing to hide the TAA mistakes in post processing, and when implemented well(*) then it looks pretty effective. It’s certainly much, much cheaper to calculate that ‘proper’ antialiasing like MSAA.

  (*) It is also quite easy to not implement TAA well, and earn the ire of gamers for turning everything into a blurry mess. Doom (2016) does a fantastic job of it - it’s in the engine at a low level - and I’ve never seen anyone complain about that game being blurry or smeared.

  It takes time to load high-quality textures and models from disk, and it uses up the RAM budget for each frame. Using lower-quality textures and models for distant objects greatly helps rendering speed and prevents stutter, and a bit of depth-of-field hides the low-quality rendering with a bit of a smear.

  Now, if your graphics card greatly exceeds the design requirement (which was probably some kind of console) then you can switch these effects off and the game will look even better, which might make you question why they’re there in the first place. To help consoles look better with some ‘cinematic’ effects, is why.

  source