Announcement

Collapse
No announcement yet.

Valve's Dota 2 Adds AMD FidelityFX Super Resolution

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • oiaohm
    replied
    Originally posted by sandy8925
    All newer GPUs have compute capabilities, it can compute on AMD and Intel GPUs too.
    The problem here no matter how you look at it power of a GPU is limited. Remember as compute in vulkan gets sorted out we could get more games using video card compute for game logic stuff. So the method used to upscale need to in fact justify its GPU usage. DLSS being the only major talked about option has kind of been a mistake.

    Nvidia does have it own sharping filter and upscale that can be combined with games instead of DLSS as well.


    Yes the Nvidia sharpening filter and upscale uses less GPU resources than enabling DLSS and can give Nvidia users better performance than using DLSS as well.

    There does need to be the question with DLSS 2.0 for the GPU cost is the quality boost over updated classical algorithms that have lower GPU cost worth it. If it not worth it nvidia need to fix their limitations on their own classic sharpening solution.

    Yes Nvidia have never themselves showed a proper compare of the Nvidia scaling options and parties reviewing DLSS have also failed in most cases to put DLSS head to head with the classical solutions include in the Nvidia drivers. Pushing DLSS has you buy a new card the updated classical algorithms sharpen and upscale that Nvidia has also works on older Nvidia cards.

    The reality here the question of does DLSS provide enough benefit for it cost even using Nvidia parts has never been attempted to be answered properly. Yes just because something has artificial intelligence does not mean it will be providing a cost effective solution or be quirk-less. There is case after case after case where artificial intelligence is worse than classical solutions. We have the same thing with TAA where there is case after case that it loses to classical solutions and some cases newer solutions like sony one.

    Yes there is a chance that a person has bought a new Nvidia card to use DLSS when the games they play would have been better handled by the Nvidia classical solutions instead.

    Leave a comment:


  • zzarko
    replied
    I guess that somebody already posted, but if not, Shadow of Tomb Raider in its Linux build has FidelityFX option in the settings. I haven't tried it yet, just noticed today.

    Leave a comment:


  • oiaohm
    replied
    Originally posted by aufkrawall View Post
    It's TAA: https://blog.playstation.com/2018/09...ls-spider-man/

    What kind of syndrome is this to constantly make up the weirdest BS and even believe it yourself?
    No this is BS from you. Please learn to read. The sony method there is called "temporal injection" this is not TAA at all. There is a sony patent on this technology. Yes it also says over 50% of the frame pixels were replaced.
    We render slightly more than half the pixels, then “inject” them into a full-sized 4k image.
    This not TAA at all. This is using the same kind of method that is used in H264 video compression of only storing what has changed between frames. What "temporal injection" is doing is areas of the frame that have not changed not re-rendering. "Temporal injection" is not antialiasing. If you look up the sony patent they are still using the old MSAA anti-aliasing. This sony method is ghost free.

    Yes the way the person talk about it in that write up does exactly match what the patent says or what the application is really doing. Temporal injection in the patent includes a pixel interleave so that every so many frames all pixels have been replaced at least once.

    By slightly jittering the locations of the rendered pixels from frame to frame, we can reconstruct a very accurate anti-aliased full 4k image over a handful of frames.

    This bit of text is highly deceptive. The way this is written seams to match TAA but its not. When they say rendered pixels they are not mixing the new pixels with old and they are talking here about the "temporal injection" total image replacement system so that every so many frames the complete image has been replaced. This is a very carefully word true statement that if you read the wrong way you could been sucker punched into spending massive resources working on TAA attempting to copy them when that is not what they are using.
    Last edited by oiaohm; 25 June 2021, 10:34 PM.

    Leave a comment:


  • oiaohm
    replied
    Originally posted by artivision View Post
    Weird dispute you have started people. We all know what NV solution was and what it is now and we all know for patents. The thing here is that Unreal Engine 4.19 and above have Temporal Upsampling activated by config file wile version 5 will use it by default. Temporal Upsampling is good but: A) It's not really better than other solutions, B) It's not more available than other solutions, C) It's not an exclusive technology and other engines have it.
    artivision you have made a very big jump with point C. Just because X engines do it don't mean all engines can. Epic Games has a patent agreement between them and Nvidia this is why Epic Games made engines like Unreal Engine can use TAA technology without fear from Nvidia patents. If you don't have a patent agreement with Nvidia implementing TAA you are taking a serous risk. The patent issue makes using TAA for smaller developers who don't agreement already problematic.

    Originally posted by artivision View Post
    No problem with a good TAAU implementation, the problem is how NV advertised this like supreme exclusive advantage and exclusive reason to buy RTX Gpus. That's fraud from the start when they had nothing in version 1.0 till they got saved by TAAU in version 2.0.
    Question does a good TAAU implementation reality exist. The horrible reality Unreal Engine 5 release so far has been put back twice due to ghosting in the TAAU implementation. Remember Unreal Engine developers are well funded by epic if you are a smaller game maker using a less funding development engine most likely you don't have the resources to have a TAAU as good as the Unreal Engine one and the Unreal Engine TAAU for the 5 version is still buggy.

    Do note I said 2030 the patents from Nvidia expires on TAA. DLSS was only released in 2019. Patent life is 20 year from approval of filing in the USA so a patent that expires in 2030 was submit before 2010 and approved in 2010. Yes if you follow it back the technology Nvidia uses in DLSS 2.0 they put in the first paperwork over in in 2005. Nvidia does have patents for AI and non AI implementation of TAAU.

    The names of "Temporal antialiasing" the first time this name is used is in the Nvidia documents for the patent submits in 2005. The next party to use that name was developers from Crytek giving a presentation over the CryEngine guess what Crytek had already got a patent agreement with Nvidia before that. This something that most people are not aware of that every major game engine that has TAA the company behind it has a Nvidia Patent agreement. Nvidia was in the antialiasing game well before the DLSS stuff and applying for patents covering it.

    Originally posted by artivision View Post
    After CAS was released i'm using it extensively (at 0.5 setting) on all my DXVK games via VKbasalt and by default with an export command in my environment file. This was the first time that i was happy that i paid a company in advance (buying a Polaris when they first came out). My money did go to a good cause. My old games like ACIII look better beyond the Ultra available option.
    I have come to the point of view that TAA is most likely the wrong path.
    1) TAA being way too time expensive for smaller development teams to get working reasonably correctly.
    2) TAA ghost problem introduces a artefact that is adverse to many types of game play.
    3) TAA you have the Nvidia patent problem.

    1) CAS does not generate ghosting.
    2) Yes CAS can cause some visual artefact these artifacts are normally not adverse to game play.
    3) CAS does not require on going developer hours and testing hours to get to the point of being decent.

    This is the thing the other options to TAA are more cost effective development time for game engine developers to use. Yes why you see games with old school bilinear upscaler the result may be garbage but its requires very little effort from the developer and does not introduce game play effecting artefacts.

    There is two different things game engine is after.
    1) playability
    2) graphics quality.

    Please note that order is critical you can make the best looking game on earth but if its not playable due to some artefact or issue its not a game this is why TAA ghosting is so bad because it can ruin a person ability to play areas of a game as the developer intended.

    Leave a comment:


  • aufkrawall
    replied
    Originally posted by oiaohm View Post
    Yes the CAS that is part of the FSR is designed to attempt to remove some of the TAA caused ghosting.
    Of course not.

    Originally posted by oiaohm View Post
    Sony for games with their upscaler is also doing reconstruction but without TAA.
    It's TAA: https://blog.playstation.com/2018/09...ls-spider-man/

    What kind of syndrome is this to constantly make up the weirdest BS and even believe it yourself?

    Leave a comment:


  • 900k
    replied
    Originally posted by leipero View Post

    I don't know if FSR is just a shader, I wouldn't think so, because if it was, it wouldn't need to be implemented in games directly I assume. I guess it would be something like Radeon Image Sharpening if that is the case, and that works relatively well, but I assume it's not comparable to the FSR (otherwise FSR wouldn't make any sense). Solution that works similar to Image Sharpening with "FSR method" would be a preferable way for both AMD and users, because it wouldn't be limited by game developers. How possible is that, that's another question.
    well the thing is that if fsr is a shader( which im pretty sure it is, given amd's slides), you wouldn't want it to run after a post-process like film grain, so being implemented in the game helps alot, also trying to force games to render at lower resolutions and adding a post-process via a driver hack is messy, and probably wouldn't work very well.

    Leave a comment:


  • artivision
    replied
    Weird dispute you have started people. We all know what NV solution was and what it is now and we all know for patents. The thing here is that Unreal Engine 4.19 and above have Temporal Upsampling activated by config file wile version 5 will use it by default. Temporal Upsampling is good but: A) It's not really better than other solutions, B) It's not more available than other solutions, C) It's not an exclusive technology and other engines have it.

    No problem with a good TAAU implementation, the problem is how NV advertised this like supreme exclusive advantage and exclusive reason to buy RTX Gpus. That's fraud from the start when they had nothing in version 1.0 till they got saved by TAAU in version 2.0.

    After CAS was released i'm using it extensively (at 0.5 setting) on all my DXVK games via VKbasalt and by default with an export command in my environment file. This was the first time that i was happy that i paid a company in advance (buying a Polaris when they first came out). My money did go to a good cause. My old games like ACIII look better beyond the Ultra available option.
    Last edited by artivision; 25 June 2021, 12:09 PM.

    Leave a comment:


  • LeJimster
    replied
    FSR may not be as good as DLSS, but its something we can play with on AMD and hopefully they continue to improve it. I have been holding off buying Horizon Zero Dawn and a couple of other games because the frame rate on my card is less than I would like, but now we have the option of FSR... I might actually buy the game now.

    Leave a comment:


  • oiaohm
    replied
    Originally posted by leipero View Post
    oiaohm I agree with what you said about laws and TAA, that specific implementation of TAA really is as good if not better than 4xMSAA with minor performance cost, I can't say it's always the case, I'm not a "gamer" (person who plays a lot of different games) so my experience is fairly limited. Considering the fact that LCD displays in general have (usually tons of) ghosting (I yet have to see one that doesn't - like good old CRTs), I don't find it to be an issue. I also did read somwhere that TAA implements some sort of "motion edge manipulation" (my term, can't remember what was the name) where it actually helps in case of LCDs and their issues with motion image presentation, or maybe I mixed it with something else, but I'm quite positive it was about TAA. So yeah, in general, I would argue that in most cases ghosting is a non-issue, especially on LCD panels.
    LCD panel ghosting make TAA ghosting worse not hide it. Why ghost from the TAA can end up reinforcing the LCD ghosting the result of this double ghost can be a ghost image that looks almost as clear as the correctly rendered leading to gamer error. This is why particular games with LCD ghosting+TAA ghosting is really bad and results in players of those games have a absolute hate of TAA. TAA is one of these things when it works it good when it does it really does not. Yes the CAS that is part of the FSR is designed to attempt to remove some of the TAA caused ghosting.

    Originally posted by leipero View Post
    As for FSR, I really don't know much about it, so that's a good info. One thing is important to note, that unlike TAA, FSR doesn't combine information from previous frames and current frame to apply it, hence why it isn't as good at image reconstruction (it basically doesn't do it at current form) as other methods such as DLSS. I saw other fidelityfx features, it's kinda nice to have things separate, so you apply things you want/need, as long as it isn't too much complications that would likely backfire (possibility for user error, hence not fulfilling whole potential of the methods combined).
    There is a trade off doing reconstitution across frames that is ghosting. Sony upscaler is interesting as it using z information and profile data from multi rendering of the same frames instead of using previous frames this avoids the TAA ghost generation problem. Yes video enchancement of damage footage you are forced to look across multi frames because due to damage you are unlikely to be able to generate z data and you cannot do a sony of have the game run though at multi different resultions to tune upscaler. Sony for games with their upscaler is also doing reconstruction but without TAA.

    Leave a comment:


  • leipero
    replied
    oiaohm I agree with what you said about laws and TAA, that specific implementation of TAA really is as good if not better than 4xMSAA with minor performance cost, I can't say it's always the case, I'm not a "gamer" (person who plays a lot of different games) so my experience is fairly limited. Considering the fact that LCD displays in general have (usually tons of) ghosting (I yet have to see one that doesn't - like good old CRTs), I don't find it to be an issue. I also did read somwhere that TAA implements some sort of "motion edge manipulation" (my term, can't remember what was the name) where it actually helps in case of LCDs and their issues with motion image presentation, or maybe I mixed it with something else, but I'm quite positive it was about TAA. So yeah, in general, I would argue that in most cases ghosting is a non-issue, especially on LCD panels.

    As for FSR, I really don't know much about it, so that's a good info. One thing is important to note, that unlike TAA, FSR doesn't combine information from previous frames and current frame to apply it, hence why it isn't as good at image reconstruction (it basically doesn't do it at current form) as other methods such as DLSS.
    I saw other fidelityfx features, it's kinda nice to have things separate, so you apply things you want/need, as long as it isn't too much complications that would likely backfire (possibility for user error, hence not fulfilling whole potential of the methods combined).

    Sort of off topic:
    In video upscaling, there are different methods used by different applications and people, there are methods that are more "to the core" and do not introduce artificial "anything" really to make image looks "non natural", however, when we talk about "image quality" of such methods, the goal of those methods should not be "What is the ost precise method?", but "What does have the biggest effect in observable experience by the user" IMO. Take for example "A.I." used by Topaz for their "Video Enhancer", they are far away from the "most precise solution", but, in general "user impression", they are far above other methods in most cases, maybe only not for trained eye who seeks some sort of "artistic form" or whatever, but, for general use, you don't really need that.

    Leave a comment:

Working...
X