Didn't DF find regressions with the transformer model vs cnn as well as improvements?
Correct, it's not all gravy, the new transformer model has some issues that need to be worked out
Last edited:
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Didn't DF find regressions with the transformer model vs cnn as well as improvements?
I actually don't like the Hogwarts 'improvements'. The stability is good, and that's a plus, but everything looks overly sharp, overly glossy and like it's got more texture because more is 'better'. Makes me think of super saturated TVs because more colour equalled better picture.If performance is better then of course quality will be better than the old quality on CNN yeah
![]()
"Better than native" - Finally nobody can hate on that claim.
The DLAA testing should have been with DLSS4. It would have been (normally) better.If performance is better then of course quality will be better than the old quality on CNN yeah
![]()
"Better than native" - Finally nobody can hate on that claim.
+1I must be living in crazy town because a native 4k image without any TAA looks horrible. Nostalgia for jaggies, shimmering and image instability isn't really a thing in my household.
And in current times all the transparencies were replaced by lower resolution transparent textures with forced TAA for temporal accumulation which blurs the hell out of them - in effect you get often smeary mess instead of proper image (unrelated to upscaling or AA used in the game). I don't call that progress, it's they opposite. Some Devs are able to make such things good, most have no clue how to, it feels+1
And MSAA was ok as long as you didn’t have transparencies. I remember there used to be an additional MSAA for alpha but was very costly and had to run it maxed out to notice any difference. Plus, no idea if it would have worked on animated transparencies ( think **** used to be static back then, like no moving ‘grass’ or ‘fences’ or anything like that ).
Anyway, back to current times.
Start IJ in the forest with transformer model, look at the foliage at the edges of the view (about 1cm thick area for me) and move forward, back, left and right. Come back and say that again."Better than native" - Finally nobody can hate on that claim.
I've found a few cases of very buggy, garbled mess regression too. It seems global turning it on missing in Nvidia app might be missing because of that - Nvidia is likely aware it's not always better, so they do it case by case only.Didn't DF find regressions with the transformer model vs cnn as well as improvements?
Still garbling in IJ foliage in movement, though.Dunno if many of you guys know but theres a new DLSS files and Preset K, its what the override uses if you set it too latest, IMO it looks better then Preset J and in HFW I'm not longer having issues
I must be living in crazy town because a native 4k image without any TAA looks horrible. Nostalgia for jaggies, shimmering and image instability isn't really a thing in my household.
So you guys haven't played Forza Horizon 4-5 I take it.+1
And MSAA was ok as long as you didn’t have transparencies. I remember there used to be an additional MSAA for alpha but was very costly and had to run it maxed out to notice any difference. Plus, no idea if it would have worked on animated transparencies ( think **** used to be static back then, like no moving ‘grass’ or ‘fences’ or anything like that ).
Anyway, back to current times.
Start IJ in the forest with transformer model, look at the foliage at the edges of the view (about 1cm thick area for me) and move forward, back, left and right. Come back and say that again.For people who do not have that game: it's a garbled mess of artefacts. FG is off.
I've tried to capture it on video, but there's nothing on it - I can see it very clearly on the screen, but I can't capture it with NVIDIA software. Interesting!Seems fine to me??
Well, I see it on the screen but on the av1 capture 100mb it's just not very visible there at all (I see it even on yours but if one didn't know what to look for, one wouldn't spot it easily). On my mobile video it's visible enough (still not as clear as in person but it's there). That said, I'm not sure how you can't see it in your game, it's there, even in your own video (just much less visible than in person). Maybe we just have eyes sensitive to different things.I see it exactly as my video shows^^
The capture is raw too so whatever is on screen is captured (100Mbps AV1 too).
If he's moving very fast, you stay very stillRaw as in direct from the capture unedited/touched.
As for in person, like I said it was the same as shown in the video, as in no real issue at all, nowhere you'd mentioned it was a garbled mess of artefacts which is not accurate as it's nothing like that at all. I could record record the screen off-screen with blackmagic camera at 200Mbps and still show the same results after uploading the video to google drive but that's a whole lot of effort to show the same results as the youtube video does anyway which would have shown a garbled mess regardless of youtube's own compression if there was a garbled mess as youtube compression only makes existing artefacts look worse, not the other way round.
Edit* From your phone video you are moving the camera around really fast which goes far and beyond "normal" gameplay camera movement, deliberately doing this to highlight any potential motion artefacting is kind of unfair as that kind of use goes beyond what is considered normal gameplay. If we all just cherry picked things this way then yes even the most native of native renders will show artefacts or error, as no mode is ever perfect, it's all about the best balance, which this is.