Image Quality and Framebuffer Speculations for WIP/alpha/beta/E3 games *Read the first post*
- Thread starter Farid
- Start date
Looks like it.
Yeah, though I think you can tell that it's not doing the greatest job of detection.
How do you mean?
It just looks very rough and ugly in those screenshots N_B has linked. I realize that some of it is because the lack of 2xMSAA (or was it Quincunx?) but still, it hurts the eyes a bit compared to UC2. Further away from offline CG image quality.
2x, but it was pretty broken since they only enabled it in the material pass and lighting was done on the SPUs.
I'd be curious to know why they bothered switching, but anyways.
Yeah, probably just tweaked the edge detection. It's still more blur than the previous two games had just by nature of the algorithm, and there are still a fair number of jagged edges left. I suppose the resulting blur is part of the weird look the beta had and in the screens here.
I'd be curious to know why they bothered switching, but anyways.
That's strange because the word I'd use for these screens is absolutely not blur but noise instead. A LOT of noise and not even the film grain looking kind. It's like the texture LOD is set to far too high.
Can't recall specifically, I'll have to look through the GDC slides again, but I don't remember them mentioning the specific cost of enabling it.
edit:
Ah ok, it was just in the graph. So 7ms for the depth/normals @ 2x, which should be same speed as 1x if I'm not mistaken (ROPs are 2x samples per clock, and this is Z+colour) The resolve from 2x RGBM to 1xFP16 is nearly 2.5ms.
And/or the edge filter they're using.
That's what it looks like, giving up on image quality related features in order to add more content, more action, more set pieces, while hoping that they can get away with it.
Goes to show how utterly efficient UC2's engine had to be, they can't push the system any further - so for every single thing they add, they have to take away something too.
Goes to show how utterly efficient UC2's engine had to be, they can't push the system any further - so for every single thing they add, they have to take away something too.
Yep. PS3 and 360 with hd pack are the same.
It definitely does not look like they got rid of it. There's quite a bit of the enhanced contrast/saturation/crushed blacks that's signature of filmic tone mapping curves.
I'm not so sure if you remember exactly the past unchy (no offence) but the IQ was the 'worst' part of unchy engine because msaa was almost useless in the most of places imho ... probably the CG offline (a lot more smoother than the previous) gives the feeling of worse IQ ingame (& shimmering gives it's 'best' contribution here ), but essentially because post processing filter works better in the close camera views & cinematic.
(((interference)))
Veteran
As far as I know. It's on DF's plate as well, but there's a lot to do.
What's the verdict leaning towards so far?
That's exactly what console graphics technology development is all about. For the last half year I have been optimizing and optimizing our GPU code just to get the last features in. Basically at the end you have to move single shader instructions from one shader to other to shift bottlenecks a little bit (extra alu is free if samping or texture cache are bottlenecks and vise versa). It's really challenging, since everyone expects you to get the features complete, but at the same time you have to frantically optimize to fit everything to the 16 ms (60 fps is a bitch on current consoles). It's really surprising how large tech refactoring I find myself willing to do, just to get a 1-2% GPU perf boost, if that 1-2% allows us to add a new feature
How much of that optimisation would you incorporate into future engine designs? Will your next-gen games use what you've learnt squeezing cycles out of this gen, or are the optimisations so specific to certain code that it bare no use going forward?That's exactly what console graphics technology development is all about. For the last half year I have been optimizing and optimizing our GPU code just to get the last features in. Basically at the end you have to move single shader instructions from one shader to other to shift bottlenecks a little bit (extra alu is free if samping or texture cache are bottlenecks and vise versa). It's really challenging, since everyone expects you to get the features complete, but at the same time you have to frantically optimize to fit everything to the 16 ms (60 fps is a bitch on current consoles). It's really surprising how large tech refactoring I find myself willing to do, just to get a 1-2% GPU perf boost, if that 1-2% allows us to add a new feature