An observation cannot be untrue. It is true reguardless or any argument. If the person viewing the event, makes a statement. That is his truth of events and cannot be disproven.
-
Competitor rules
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.
Fidelity Super Resolution in 2021
- Thread starter Grim5
- Start date
An observation cannot be untrue. It is true reguardless or any argument. If the person viewing the event, makes a statement. That is his truth of events and cannot be disproven.
AA is normally taking multiple samples per pixel, to smooth out the jaggies. Upscaling you are filling in missing pixels.
My point was, how we name things isn't dependant of how they work on a basic level but what function they actually have and what the outcome of it is. FSR's function isn't to anti-alias edges but to preserve them when upscaling - otherwise we would have gaps in pixels and edges wouldn't be edges anymore but dotted lines. AA function is to smooth edges, adding more information than existed in native resolution. DLSS is much closer to AA (it was its primary function, initially) that FSR is, yet I don't see people calling it AA, even though it actually does AA as part of the process - but it's not its main goal and outcome anymore.
But in the end, it's all just semantics and not really that important here.
I actually updated my post to say some of what DLSS does is TAA although using a very different approach.
Soldato
Its DAMAGE Control that is all... That High premium price to get into DLSS is under threat by a free option that in its first 1.0 release is doing better than DLSS 1.0...
Comparing DLSS 2.0 to FSR 1.0 is abit silly really DLSS has had a few years to get it right.
The fact is you have a paid software that does a better job vs a free software that does a good job!
I am a fan of open standards FSR get my vote
Comparing DLSS 2.0 to FSR 1.0 is abit silly really DLSS has had a few years to get it right.
The fact is you have a paid software that does a better job vs a free software that does a good job!
I am a fan of open standards FSR get my vote
TBF though you aren't really having to pay any more to get dlss, like most of nvidia software features, it is a nice to have, sure you could say it is bundled in the overall price of the card but then couldn't we say the same about any of amds specific features? Plus it's not like you're paying considerably more even at MSRP values, most people are probably paying extra for the better ray tracing perf, to use their gsync screen, nvenc streaming features and things like that than solely just for dlss.
Also, given the way things went this year, plenty of people this time round paid far less for nvidia cards than what a lot would have paid for amd cards because of amds UK ban then their site not having any bot protection and stock being rare as ****....
On the topic of FSR, I think it looks very good for 4k and the perf. gains are lovely, don't think I would use it for anything less than 4k though (and only ultra quality or perhaps quality setting) as the results aren't quite there yet, I fully expect this to improve though.
What I like the most about FSR (or in general fidelityFX) is that it provides free alternative to older tech that's already included in engines like Unity. Devs can use it without working on their own upscaling methods or TAA (Unity's one is really bad, AMD's version much better etc.). This benefits everyone, irrelevant of the GPU brand they're using. It gives players choice too, without locking them to any particular brand. It's just sad that some people seem to hate the having a choice or things improving unless it's done by one particular brand they love and works only on the particular card that they own.
I switched lots between NVIDIA and AMD GPUs over time, I had GSYNC monitor, currently FreeSync/G-Sync compatible, I had Intel and AMD CPUs over time too, etc. I don't care much about the brand my GPU or monitor carry; I care about choice and improvements in games. I don't like tech that is locked to just one particular vendor/model of the hardware.
I switched lots between NVIDIA and AMD GPUs over time, I had GSYNC monitor, currently FreeSync/G-Sync compatible, I had Intel and AMD CPUs over time too, etc. I don't care much about the brand my GPU or monitor carry; I care about choice and improvements in games. I don't like tech that is locked to just one particular vendor/model of the hardware.
Soldato
https://www.overclockers.co.uk/forums/threads/graphics-cards-strikes.18766760/
Rules, mentioning as i reckon mods will start coming along soon to sort the thread out and get it back on topic soon.
Rules, mentioning as i reckon mods will start coming along soon to sort the thread out and get it back on topic soon.
AMD have a patent for a hybrid upscaling tech. One that adds a AI network into the mix. Also Unreal Engine 5 shows that a Temporal upscaler is possible as well. You can get good results without an AI network.
AMD clearly went for speed with FSR 1.0 by using a spatial upscaling method. From the source code the upscaler is based on the Lanczos algorithm, which was invented in 1979. It skips the expensive sin(), rcp() and sqrt() instructions. Adds additional logic to avoid the ringing effects.
Last edited:
That is true sometimes but you lose other details, depends on what's your preference. For example that hero's beard at 3:30 looks better on DLSS even if it is blurred. So you get a much crisper image with SMAA but also a lot more noise.FSR looking crisp AF next to a blurry DLSS. The advantage of being able to choose your own AA
It's not just with SMAA tho, it's also with TAA where there's a loss of details. All the newly self-discovered wire-sophists forgot to check for those (by accident ofc). Better than that blurry DLSS mess in any case
DLSS has the highest quality in all images. The speakers are night and day better than even native. The hair and all other fine details are better as well in DLSS. The overall effect is better looking than native. FSR oversharpens compared to native, even areas that are ment to be blurred (out of focus) are also sharpened. Can be seen in the video with thor, his armour is massively oversharpened. The bridge scene the cables on the bridge the overall image is less than native with FSR. Even the water. DLSS looks better than native. All the fine details are better than native with DLSS. Hair, cables, speakers etc. Basically FSR is not there yet with image quality.
Last edited:
That image with the speakers is so freaking weird. I mean the mesh looks better in DLSS but idk if it's not looking too good. since you can't see what's inside anymore. If they created the interior it means it should be seen through the mesh.It's not just with SMAA tho, it's also with TAA where there's a loss of details. All the newly self-discovered wire-sophists forgot to check for those (by accident ofc). Better than that blurry DLSS mess in any case
That image with the speakers is so freaking weird. I mean the mesh looks better in DLSS but idk if it's not looking too good. since you can't see what's inside anymore. If they created the interior it means it should be seen through the mesh.
I'm assuming the model needs better training to properly interpret the speaker - it is just filling in mimicking what it thinks should be there and getting it wrong by the looks of it unless the lower resolution is causing some kind of LOD thing to happen and a different prop used or the texture looks different.
The second image DLSS is actually far superior - FSR is overly sharpening the jaggies and loss of text detail in the native image whils DLSS has correctly interpreted the text detail and trying to smooth out the jagged edges. (It could still be better but is the less worse outcome).
Curious that a mesh front that clearly shows internal details in 4K native becomes a solid box showing zero internal details in DLSS.
Amazing how that kind of decision is possible. Which then gets pulled up as the solid box looking better than the mesh effects from native and FSR.
The box does looks better but there's straight up no mesh in the DLSS image so what exactly has happened.
If everyone else can do a solid box it would also look better right?
Amazing how that kind of decision is possible. Which then gets pulled up as the solid box looking better than the mesh effects from native and FSR.
The box does looks better but there's straight up no mesh in the DLSS image so what exactly has happened.
If everyone else can do a solid box it would also look better right?
It's likely happening because the underlying resolution of FSR UQ is actually far higher.
It's a silly comparison as the DLSS Quality with 1440p will be performing far better than FSR Ultra Quality at 1662p. The latter has 32% more pixels to render with it's shaders.
DLSS doesn't currently have a level with the same underlying performance as FSR UQ.
If that example is actually DLSS is working on a solid box because that's what the lower resolution has then that example deserves to be punted out the window because its not an honest example of the tech working.
Also no, we saw FSR UQ at 1662p -> 4k pushing higher fps than DLSS quality 1440p -> 4k in necromunda so the extra pixels required for FSR UQ managed to be less work than sorting out DLSS on less pixels. Other examples may vary but it's certainly curious.
*put correct video, so many virtually identical videos spammed in here it's hard to remember which it was