**Nvidia's DLSS Technology: A Mixed Bag**

DLSS, short for Deep Learning Super Sampling, is Nvidia's latest attempt to revolutionize the way we experience games on PC. With its promise of faster frame rates and improved visuals, it's no wonder many are excited about this technology. However, as with any new innovation, there are bound to be some teething issues.

One of the biggest problems we encountered with DLSS was its performance in certain situations. In our testing, the image quality was often reduced to a point where it looked like it had been applied with vaseline - blurry and lacking clarity. This wasn't just a minor annoyance; it was a full-blown problem that made us question whether DLSS was truly worth using.

But what's even more frustrating is that DLSS seems to be locked to certain hardware configurations. In our testing, we found that the technology was only working with Nvidia's RTX devices, and not with older GPUs from AMD or other manufacturers. This meant that we couldn't even test it on some of the most powerful hardware out there - the kind of thing that would normally make us sit up and take notice.

And then there's the issue of DLSS being marketed as a way to boost performance without actually doing much of anything. Nvidia claims that DLSS can handle demanding games with ease, but in our testing, we found that it was often struggling to keep up. It's almost like they're trying to sell us on something that doesn't quite live up to the hype.

**A New Paradigm?**

Now, before I get too carried away with my criticisms, I want to be clear: DLSS is still a relatively new technology, and it's not going to be perfect from day one. Nvidia has always been known for pushing the boundaries of what's possible in gaming, and this technology is no exception.

In fact, DLSS has some real advantages over other technologies like FSR (FidelityFX Super Resolution). For one thing, it's not hardware locked - you can use it with a wide range of GPUs from both Nvidia and AMD. That's a big deal, especially for those who want to stick with their existing hardware rather than upgrading just to play the latest games.

But where DLSS really shines is in its ability to tackle some of the same challenges as FSR. Both technologies aim to improve frame rates and visuals without requiring massive increases in processing power - but they do it in different ways. Nvidia's approach is more focused on using machine learning algorithms to upscale lower-resolution images, while AMD takes a more straightforward approach with FSR.

**The Games List: A Major Weakness**

So far, the major downside of DLSS is its limited games list. While it does work with some popular titles like Control and Terminator 2, there are still plenty of other great games out there that aren't supported - including Cyberpunk 2077, which has been a major point of contention for Nvidia fans.

It's clear that AMD is playing catch-up here, but they're doing so in style. In our testing, we found that DLSS was working well in most cases, especially when compared to FSR. However, there are still some issues with image quality - particularly at higher resolutions - which can make the technology feel like a compromise.

**The Future of DLSS**

So what's next for DLSS? Nvidia is already hard at work on version 2.0, which promises even more advanced features and capabilities. We're not yet sure what these will be, but we're excited to see how they'll shape up.

One thing that's clear is that Nvidia needs to get comfortable with the idea of comparing its technology directly to AMD's FSR. It's a challenge they may struggle with, given their history of marketing rival technologies as separate and distinct. But if they can find a way to make DLSS work in more games - especially those that are currently underserved by the technology - it could be a major game-changer.

Until then, we'll just have to keep an eye on developments in the world of PC gaming and see how these technologies continue to evolve. One thing's for sure: Nvidia's DLSS is a force to be reckoned with, even if it's not without its flaws.

"WEBVTTKind: captionsLanguage: enhot on the heels of our intel dg1 video card review and our amd apu reviews the 56 and 5700 g options we're now looking at amd fsr or fidelity fx super resolution this has been often compared at least in part to dlss it's never really been meant to be a head-to-head competitor with the lss but that's been sort of the lazy or the easy comparison and now we're going to look at fsr explain more in detail what it is how it relates to dlss and where it diverges in path and we'll look at some benchmark performance as well for the various quality modes that are offered through amd fsr this one if you haven't paid attention for the last month or so has been in the news mostly because of amd's absolute power move where it announced it was fully supporting nvidia's products with fsr before that this video is brought to you by crucial and it's crucial x6 portable ssd we use these external ssds all over the office for rapidly transferring games and files between systems the x6 comes in 500 gigabytes 1 terabyte 2 terabyte and 4 terabyte capacities with usb a or usb type c for the cable for a high speed and high capacity external drive from crucial click the link in the description below so the hardware support list for this one's pretty big it supports more or less everything at this point uh gtx 900 series works the 10 series definitely works andy showed that in its keynote where the 1060 got uplift from fsr and uh also and these older hardware works and two the intel dg1 that we just reviewed in theory it should work with fsr we haven't done a full test pass on it but it's fairly open the games list is short unfortunately there aren't a lot of games we'll put it up on the screen that support fsr right now but amd is working with developers to broaden support and that's the key here because without support from developers it'll go nowhere let's cover the basics first if you missed it amd fidelity fx super resolution is the newest addition to andy's fidelity fx graphics suite and he's got a couple of other things in there it's a library of of effect fidelity fx is part of andy's gpu open and it's open source software suite which encompasses other libraries like tres fx cass for example or contrast adaptive sharpening and also supports shadow fx and a couple of other sdks as the name implies fidelity fx has a focus on post-processing effects that increase visual fidelity like denoiser and contrast adaptive sharpening and to summarize gpu open is the top level name for all of amd's open source software fidelity fx is a group of effects and fsr is one specific effect from that group fsr is intended to take low resolution source images or source frames and then upscale them to make them appear sharper and overall better in quality and this is supposed to help balance the performance and the image quality especially in a market where gpu choices have been limited for a long time and that's why this got so much uh praise from the enthusiast community when it was announced because instead of pushing out another gpu and they read the room a bit and put this out there instead to help with the older gpus at least that was sort of the marketing around it so avasara's intended to to do some of the things that dlss does but not at all in the same way they approach the topic very differently this is ultimately a game feature not a driver feature and that's key so the games have to individually add avasara as a graphics option and likewise if amd stops supporting fsr as long as the games keep supporting it it'll continue to work amd makes a lot of claims with this one but it also has a lot of qualifiers associated with those claims so it's a bizarre mix of of hesitance and confidence with fsr one of the claims is a quote-unquote 2.4 x increase in performance and andy says this all throughout the reviewer guide you go down to the footnote at the very bottom and it specifies that quote uh provides on average 2.4 x the performance when running at 4k and fsr performance mode across fsr supported games and we'll talk about performance mode in a moment but it's the lowest quality one so just to set the stage before we get into benchmarks don't expect this 2.4 x number that was also at 4k uh that amd is citing here in everything it still works pretty well but just just keep the expectations in check for what fsr really does okay enough of all that the team pulled an all-nighter on this one to get it up after the apus and the intel uh dgpu review that we just did so hopefully you find this interesting we've got a section from patrick first talking about how fsr works and what it is and then we'll go through some benchmarks and image quality comparisons we should probably start with the easy misconceptions and the confusion within amd's software stack amd's existing cass effect includes an optional static or dynamic upscaling element as seen in games like cyberpunk 2077 actually this may lead to some confusion as to the difference between fidelity fx cast and fidelity fx fsr super resolution which is justifiable cass has upscaling and sharpening and fsr has upscaling and sharpening from a marketing standpoint the difference is that fsr has inherited the inference of being a one-to-one alternative to dlss while cass was branded as a more unique to amd feature in fact cass was originally marketed after the mass of dlss 1.0 when everything was blurry and looked terrible and that is why it got the marketing to did the functional difference though between cass and fsr is that fsr attempts to cleanly upscale edges of objects to create the illusion of a higher render resolution while cass is focused solely on sharpening cast doesn't necessarily try to make a 1080p frame look like a 4k frame it just does a sharpening pass based on discussion with amd fsr includes a sharpening pass and is a straight upgrade from cast andy's official statement to gamer's nexus was as follows quote fsr includes a sharpening component to extract pixel detail on the upscaled resolution image it is therefore mutually exclusive with any existing sharpening option that games may already support amd recommends that all fsr quality assessments be done against a native image that does not have cass or ris enabled as for our assumption we'd assume that fsr will eventually supersede cast fsr is also not the same thing as virtual super resolution or vsr that we covered in 2014 this was amd's name for super sampling or ssaa vsr renders at a higher resolution than the display allowing for a high quality output image with reduced aliasing vsr makes performance worse and image quality better fsr makes performance better and image quality arguably worse although amd's trying to not have that happen getting into really how it works then there are four settings levels for fsr 1.0 ultra quality quality balanced and performance this is the same naming convention that nvidia uses for dlss 2.0 except with an important edition that's ultra quality the reason this is important is because it's ultra and obviously that makes amd better end of argument see you in the comments we look forward to nvidia's dlss 2.0 thai setting in the future the geforce rtx 3080 ti each setting has a specific scale factor in relation to the native screen size for example performance has a render resolution of one half that of the screen size per dimension really important there on a 3840 by 2160 screen fsr performance upscales frames rendered at 1920x1080 balanced uses a render resolution that's 10 17 per dimension so that'd be x and y quality uses two thirds and ultra quality uses 10 13. again these are ratios per dimension 1080p by name is half of 2160p but in reality it's only a quarter of the pixel count ultra quality might sound close to native but the pixel count is still only 59 of the native screen resolution these ratios aren't tied to specific screen sizes or resolutions though and that's good because that means fsr should prove flexible in working with unusual resolutions or unusual aspect ratios andy has also stated that fsr can work with dynamic resolution scaling in games which support it a major difference from nvidia's dlss is that fsr has no temporal aspect fsr gets applied to one single rendered frame at a time with no knowledge of past frames held in the buffer or knowledge of future frames predicted by motion vectors these are both features of nvidia dlss this is a significant limitation in terms of the so-called smartness of amd's that is to say nvidia's technology is doing more in a more advanced way but it also makes fsr easier to implement and easier to scale as with the lss everyone should temper their expectations with fsr fsr is inherently lossy that's the nature of it even at ultra quality fsr is working with fewer pixels and less data than the native resolution and its reconstruction won't be perfect andy almost made it through actually the entire reviewer guide without once claiming otherwise but it did sneak in this line quote in some situations fsar at ultra quality can produce visibly better than native image quality at normal gaming viewing distances on titles that don't employ sharpening or use it sparingly okay sure sure amd there's a lot of qualifiers in that sentence but what andy's trying to say with all those qualifiers is don't get your hopes up this isn't really actually going to happen but you're going to trick yourself into thinking it did if you think that blasting sharpness looks better fsr is still per game in the sense that developers have to manually add the feature as a post-processing step after anti-aliasing is applied and before the ui and effects like film grain are added but it doesn't require per game image training like dlss 1.0 did or motion vectors provided by the game engine like dlss 2.0 does this is helpful but more so for developers than for us whether or not fsr requires image training doesn't really make much difference to the end user the aspect that does benefit the consumer is that fsr isn't strictly hardware dependent or driver dependent so it should work equally well with the non-amd gpus and the cards that don't have ray tracing hardware and it should continue to work even if amd chooses to stop actively supporting it outside of our own testing we've gotten word of fsr working on cards at least as old as the nvidia gtx 900 series and it could go back further it should also work on the intel iris xedg1 video card that we recently reviewed showing how wide of a support spectrum fsr offers time for the comparison screenshots we're going to put a couple of different fsr options on the screen alongside native right now but for now we're not going to label them the options we're showing are native resolution fsr balanced and fsr performance the point though is to allow you some time to try and identify the differences between them while we talk through this next section we're trying to help you identify your own biases so you can remove them hopefully when viewing the comparisons later for all of our comparison screenshots we maximize all the in-game settings except for post-processing effects which we minimized or disabled when possible anti-aliasing was also maxed out where available and our intention for this was to provide fsr with the best possible input and apply as few post fsr effects as possible capture was taken on our gpu test bench with a sapphire rx 6800 xd nitro and all of that explained time to show the labels for each of those settings comment below if you got it right or how wrong you got it and was up first the brightly lit foliage of no1800 presents a challenge for fsr with lots of noise in the form of one pixel shadows this is the kind of detail that's in the most danger of being lost if detail isn't visible at all in the low res source image that fsr upscales evasar obviously has no chance to reconstruct it even at ultra quality this loss of detail is visible in the grassy field at the upper left of the screen where the small stones and the blades of grass are unavoidably blurred the horizontal fence railings behind the building at the center of the screen are only partially visible even at 4k native and are promptly erased when even the highest fsr quality level is applied still ultra quality looks close enough to native 4k that the difference might not be immediately obvious without the side by side comparison and that's the main goal if the player can't notice the loss in gameplay then the output is good dropping down further to quality the trees become muddier with less distinction between leaves and branches and so all these things compared ultra quality appears to be the ideal mix for ana fsr is a clear improvement over the source resolutions that it works with comparing anno at 4k with fsr performance mode to ano running at 1080p native demonstrates that fsr can successfully highlight details in areas like the stone and the wood textures of the docks the second scene in anna is more favorable the brightly colored and clearly delineated houses and docks are better suited to upscaling with fsr than the fine details of the forest and the roofs of the houses look almost the same at quality as they do at native 4k notable exceptions are the tile roofs on the two buildings at the top of the village and the striped red and white tent at the center of the screen the closely placed parallel lines get blurred with lower source resolutions and at the balanced setting fsr actually reconstructs the stripes running in the wrong direction these artifacts aren't present when running the game at 1080p which is the source resolution used for fsr performance and the artifacts that were seen are similar to some of the ones we've seen with dlss in the past and that's because the root cause of the issue is the same which is limited source information the rifts breaker has some of the same loss of detail that anno did in areas like palm fronds and blades of grass and certainly youtube compression isn't doing it any favors ultra is mostly successful however in preserving image quality the strangest and most dramatic change is in the shadows the rift breaker is the only game we selected for comparison that supports ray traced effects with rt soft shadows and ambient occlusion both of which we maxed out as fsr drops down from ultra quality to performance each tier removes more of those shadows making them a lighter and less well-defined terminator a game that presumably exists and which has a thriving player base of about 26 players yesterday is one of the strongest examples of fsr that we've seen other than the fact that us running it on two computers increased its active player counts by eight percent the game is interesting for other reasons the beginning of the game takes place at night in the ruins of pasadena california we could hardly tell a difference between the post-apocalyptic hellscape and terminator resistance the dark concrete textures aren't incredibly sharp and detailed to begin with so even performance mode looks good in this first screenshot with the most obvious giveaways being the usual ones a chain link fence for example in the center of the frame and the wooden vents behind it dropping the render resolution to 50 reveals that parts of the chain-link fence are aliased out at lower resolutions which explains why fsr has some difficulty reconstructing it looking at just this scene it'd be difficult to detect the differences between native and fsr quality and balanced would be an acceptable compromise for a higher frame rate we selected a second scene in terminator with an npc clearly visible so that we could see how fsr handled the fabric and skin it held up just as well as it did in the previous scene if not better balanced erases some detail around the npc's collarbone but all the detail in his face is still clearly visible again the fact that terminator's textures aren't especially high-res to begin with helps it here but that's still a valid result detail in wood paneling and beams in the background is also surprisingly well preserved there's a lot that we can benchmark with amd's fsr and we've only just gotten it recently so we'll be doing more follow-up pieces in the future we're keeping it simple for our initial route we were most interested in image quality which we already talked about and our secondary interest was in overhead cost and scaling between the settings types to better understand the performance benefit or the cost from each option we'll likely expand testing in the future but for now we start with something that we don't think any or many of the other videos will look at for today and that's the new amd ryzen 7 5700g apu that we reviewed recently and that technically isn't out for the diy market amd fsr could feasibly allow these types of parts to better serve as a stop gap in between dgpu purchases with our limited time for this one this made the most sense to provide a unique angle and a look at the truly low end below what you would get with the gtx 1060 or rx 580 that amd's already sort of showcased in its keynote we'll start with rift breaker in rift breaker at 1080p and with the mix of medium high and low custom settings that we applied we measured 46 fps average on the 5700g when run without fsr moving to ultra quality or a change to 1477 by 831 the 5700g then improved by 28 the improvement with quality mode or 1280 by 720 was 50 over the original stock result or an additional 18 over the previous ultra quality mode in exchange for 300 000 fewer pixels in the source material balanced mode boosted to 79 fps average with performance jaunting to 91 fps average but amd doesn't generally recommend performance mode to begin with and with these settings and the image quality we don't either at 1080p performance ends up working off of just 960 by 540 pixels or 518 400 pixels for the source information as opposed to over 2 million for native 1080p low scaled proportionally with all the results we didn't notice any odd behavior with frame time pacing from and the fsr so on the side of frame time consistency things look good for simplicity we also made this quick percentage scaling chart the blue line shows the percentage performance improvement versus baseline the red line shows the percentage decrease in pixels from baseline going from fsr off to ultra quality had the largest single increment in performance uplift at 30 percent tied with going from balance to performance at the same time going from fsr off to ultra quality resulted in the biggest single reduction in pixel count by percentage falloff we dropped 41 percent of our pixels in the source material but gained 30 percent in performance it's clearly not perfectly linear things do smooth out though with quality settings we gained 50 percent over baseline fps and dropped 55.6 percent of our baseline pixel data balanced is also fairly even but the performance line gain crosses back over again where we see performance mode rendering one quarter of the pixels in exchange for double the performance when tested with these settings that's not the 2.4x number amd cited but that number may have been based on something else 4k instead for example no 1800 at 1080p low with the r75700g had baseline at 74fps average with for tap aaa we ran with anti-aliasing on for this one despite initial tests with it off because the amd's fsr kept defaulting it to 4x you can force it off if desired but it's the recommended setting and after looking at it there's a good reason for that that results improved to 90 fps average then 102 in the bar chart then 116 then 132fps with performance mode we needed another one of the percent scaling charts to help out though no 1800 again shown here follows the same rules as the rift breaker chart the highest cost of source material is experienced at the offset with ultra quality but we only gained 21 more performance versus baseline with this configuration from there things scaled mostly linearly and at a steep curve getting steeper when moving from balanced to performance mode performance mode is a waste here you aren't going to need super high frame rates and nano to begin with but even if you did dropping 75 of the source material which is what's used to reconstruct the image for a 79 uplift just isn't a good trait because frame rate isn't that important in nano we also ran out 1080p medium settings on the r75700g and then we checked it just the quality mode against baseline off this change got us up to 65 fps average similar in scaling to the lower settings and that moves from 42fps to a more playable result lows improved in staff moving to 33 for 0.1 lows for the 5700g this is the more realistic scenario with fsr we've been able to enable higher quality settings going again up to medium instead of low for everything at the exchange of resolution but we get the improvement with fsr and so the end result is a more playable frame rate with a better image quality next a quick look at overhead testing at 720p native versus 1080p at quality should give us a like for like resolution comparison but super resolution in this comparison creates a better quality image that of course has a cost we can only speak for lower end hardware in this quick test but the 5700g ended up with a 15 overhead shown here between these two benchmarks with the end result being a better quality image of course so that's it for this one then overall fsr appears to be it's in an interesting position because when dlss came out the first time we had a lot of problems with it and the main problems were pretty simple one was image quality was not good it was very blurry it was like vaseline all over the screen it just didn't work very well and the second problem it was locked to hardware the third problem it wasn't even really using the hardware that it was locked to it was an effort to sell more rtx cards when rtx was a new initialism it didn't exist before that was gtx and so as nvidia tried to build hype and marketing and interest around rtx and this whole new software suite and paradigm shift in the company's marketing it locked down dlss to rt devices but it wasn't always really using the rt hardware that was kind of the tricky thing about it so it was a bit misleading dls 2.0 came out and actually in the games where it works like control for example is extremely good it does a great job at conquering a lot of the same challenges that fsr tries to conquer except in a way which is just objectively more advanced and you could say using computer terminology that marketers like to use smart or intelligent but that doesn't mean that fsr is bad it just means that nvidia is doing something in a much more complicated way so as long as fsr can avoid certain things like really bad blurring and the vaseline effect then it's in a good spot and thus far the the biggest downside to fsr is clearly the games list but it has a lot of upsides and the number one upside is that's not hardware locked it does work with older gpus from both vendors it's only been validated officially by andy back to gtx 1060 and rx 480 or 470 era for amd's devices but you can run fsr on older stuff than that we haven't tested too far back to see at what point it just doesn't work but at least the gtx 9 series should work and the older than rx 400 series should work as well so that's cool and that's really good place to be you do still lose quality with this that's always going to be the case so amd was really stretching the truth when it said in the reviewer's guide that sometimes the image quality is better than native but amd in that sentence you can you can read the guilt in the marketing bs from whoever wrote that because there's just one qualifier after another of when it might be considered better than native and ultimately it can't really be better than native with this approach you might perceive it that way subjectively just based on whatever specific game specific device specific settings maybe you'll perceive it that way especially if you like sharper images but uh that is getting pretty far into the subjective at this point so just to be really clear there is going to be data loss when you are rendering at a lower count of pixels and then bringing it up to something else it's not something to be really ashamed of unless it's terrible and in the instances of high quality and quality amd does pretty well to hold on to a lot of that detail you start losing the small shadow details the one pixel wide details go away if they're not rendered at the lower resolution clearly because you need something there in order to bring the data up into a a higher resolution but ultimately it's a good first start but it's very weak in the games department terminator 26 players yesterday and i'm pretty confident that most of them are people i know who are also reviewers i don't think they're that many real players sorry if you were one of them but uh hey you were online at the same time as a bunch of reviewers at least other than that anno seems like maybe the the most legitimate or the largest of them rift breaker looks like it has a little bit of interest around it so they're getting out there but amd really needs to get into stuff like cyberpunk cass is in cyberpunk and yet fsr is conveniently absent from it why is that dlss so amd is trying very hard right now to avoid dlss comparisons and that's the part we really want to see is as it starts spreading when will amd become comfortable enough for confident enough in its product that it is willing to put it into a game that dlss is also and that's the comparison we want to see even if they're not directly linearly the same they're sort of trying to tackle some of the same issues though and that's why it's interesting anyway good first start needs more games and uh until it gets them we can't really say much more about it but from a technology standpoint appears to be working more or less as it should be just performance mode looks bad probably don't use that one but high quality is not bad so that's it for this one thanks for watching as always subscribe for more you can go to store.gamingtexas.net or patreon.comgamersnexus helps out directly in these endeavors and we'll see you all next timehot on the heels of our intel dg1 video card review and our amd apu reviews the 56 and 5700 g options we're now looking at amd fsr or fidelity fx super resolution this has been often compared at least in part to dlss it's never really been meant to be a head-to-head competitor with the lss but that's been sort of the lazy or the easy comparison and now we're going to look at fsr explain more in detail what it is how it relates to dlss and where it diverges in path and we'll look at some benchmark performance as well for the various quality modes that are offered through amd fsr this one if you haven't paid attention for the last month or so has been in the news mostly because of amd's absolute power move where it announced it was fully supporting nvidia's products with fsr before that this video is brought to you by crucial and it's crucial x6 portable ssd we use these external ssds all over the office for rapidly transferring games and files between systems the x6 comes in 500 gigabytes 1 terabyte 2 terabyte and 4 terabyte capacities with usb a or usb type c for the cable for a high speed and high capacity external drive from crucial click the link in the description below so the hardware support list for this one's pretty big it supports more or less everything at this point uh gtx 900 series works the 10 series definitely works andy showed that in its keynote where the 1060 got uplift from fsr and uh also and these older hardware works and two the intel dg1 that we just reviewed in theory it should work with fsr we haven't done a full test pass on it but it's fairly open the games list is short unfortunately there aren't a lot of games we'll put it up on the screen that support fsr right now but amd is working with developers to broaden support and that's the key here because without support from developers it'll go nowhere let's cover the basics first if you missed it amd fidelity fx super resolution is the newest addition to andy's fidelity fx graphics suite and he's got a couple of other things in there it's a library of of effect fidelity fx is part of andy's gpu open and it's open source software suite which encompasses other libraries like tres fx cass for example or contrast adaptive sharpening and also supports shadow fx and a couple of other sdks as the name implies fidelity fx has a focus on post-processing effects that increase visual fidelity like denoiser and contrast adaptive sharpening and to summarize gpu open is the top level name for all of amd's open source software fidelity fx is a group of effects and fsr is one specific effect from that group fsr is intended to take low resolution source images or source frames and then upscale them to make them appear sharper and overall better in quality and this is supposed to help balance the performance and the image quality especially in a market where gpu choices have been limited for a long time and that's why this got so much uh praise from the enthusiast community when it was announced because instead of pushing out another gpu and they read the room a bit and put this out there instead to help with the older gpus at least that was sort of the marketing around it so avasara's intended to to do some of the things that dlss does but not at all in the same way they approach the topic very differently this is ultimately a game feature not a driver feature and that's key so the games have to individually add avasara as a graphics option and likewise if amd stops supporting fsr as long as the games keep supporting it it'll continue to work amd makes a lot of claims with this one but it also has a lot of qualifiers associated with those claims so it's a bizarre mix of of hesitance and confidence with fsr one of the claims is a quote-unquote 2.4 x increase in performance and andy says this all throughout the reviewer guide you go down to the footnote at the very bottom and it specifies that quote uh provides on average 2.4 x the performance when running at 4k and fsr performance mode across fsr supported games and we'll talk about performance mode in a moment but it's the lowest quality one so just to set the stage before we get into benchmarks don't expect this 2.4 x number that was also at 4k uh that amd is citing here in everything it still works pretty well but just just keep the expectations in check for what fsr really does okay enough of all that the team pulled an all-nighter on this one to get it up after the apus and the intel uh dgpu review that we just did so hopefully you find this interesting we've got a section from patrick first talking about how fsr works and what it is and then we'll go through some benchmarks and image quality comparisons we should probably start with the easy misconceptions and the confusion within amd's software stack amd's existing cass effect includes an optional static or dynamic upscaling element as seen in games like cyberpunk 2077 actually this may lead to some confusion as to the difference between fidelity fx cast and fidelity fx fsr super resolution which is justifiable cass has upscaling and sharpening and fsr has upscaling and sharpening from a marketing standpoint the difference is that fsr has inherited the inference of being a one-to-one alternative to dlss while cass was branded as a more unique to amd feature in fact cass was originally marketed after the mass of dlss 1.0 when everything was blurry and looked terrible and that is why it got the marketing to did the functional difference though between cass and fsr is that fsr attempts to cleanly upscale edges of objects to create the illusion of a higher render resolution while cass is focused solely on sharpening cast doesn't necessarily try to make a 1080p frame look like a 4k frame it just does a sharpening pass based on discussion with amd fsr includes a sharpening pass and is a straight upgrade from cast andy's official statement to gamer's nexus was as follows quote fsr includes a sharpening component to extract pixel detail on the upscaled resolution image it is therefore mutually exclusive with any existing sharpening option that games may already support amd recommends that all fsr quality assessments be done against a native image that does not have cass or ris enabled as for our assumption we'd assume that fsr will eventually supersede cast fsr is also not the same thing as virtual super resolution or vsr that we covered in 2014 this was amd's name for super sampling or ssaa vsr renders at a higher resolution than the display allowing for a high quality output image with reduced aliasing vsr makes performance worse and image quality better fsr makes performance better and image quality arguably worse although amd's trying to not have that happen getting into really how it works then there are four settings levels for fsr 1.0 ultra quality quality balanced and performance this is the same naming convention that nvidia uses for dlss 2.0 except with an important edition that's ultra quality the reason this is important is because it's ultra and obviously that makes amd better end of argument see you in the comments we look forward to nvidia's dlss 2.0 thai setting in the future the geforce rtx 3080 ti each setting has a specific scale factor in relation to the native screen size for example performance has a render resolution of one half that of the screen size per dimension really important there on a 3840 by 2160 screen fsr performance upscales frames rendered at 1920x1080 balanced uses a render resolution that's 10 17 per dimension so that'd be x and y quality uses two thirds and ultra quality uses 10 13. again these are ratios per dimension 1080p by name is half of 2160p but in reality it's only a quarter of the pixel count ultra quality might sound close to native but the pixel count is still only 59 of the native screen resolution these ratios aren't tied to specific screen sizes or resolutions though and that's good because that means fsr should prove flexible in working with unusual resolutions or unusual aspect ratios andy has also stated that fsr can work with dynamic resolution scaling in games which support it a major difference from nvidia's dlss is that fsr has no temporal aspect fsr gets applied to one single rendered frame at a time with no knowledge of past frames held in the buffer or knowledge of future frames predicted by motion vectors these are both features of nvidia dlss this is a significant limitation in terms of the so-called smartness of amd's that is to say nvidia's technology is doing more in a more advanced way but it also makes fsr easier to implement and easier to scale as with the lss everyone should temper their expectations with fsr fsr is inherently lossy that's the nature of it even at ultra quality fsr is working with fewer pixels and less data than the native resolution and its reconstruction won't be perfect andy almost made it through actually the entire reviewer guide without once claiming otherwise but it did sneak in this line quote in some situations fsar at ultra quality can produce visibly better than native image quality at normal gaming viewing distances on titles that don't employ sharpening or use it sparingly okay sure sure amd there's a lot of qualifiers in that sentence but what andy's trying to say with all those qualifiers is don't get your hopes up this isn't really actually going to happen but you're going to trick yourself into thinking it did if you think that blasting sharpness looks better fsr is still per game in the sense that developers have to manually add the feature as a post-processing step after anti-aliasing is applied and before the ui and effects like film grain are added but it doesn't require per game image training like dlss 1.0 did or motion vectors provided by the game engine like dlss 2.0 does this is helpful but more so for developers than for us whether or not fsr requires image training doesn't really make much difference to the end user the aspect that does benefit the consumer is that fsr isn't strictly hardware dependent or driver dependent so it should work equally well with the non-amd gpus and the cards that don't have ray tracing hardware and it should continue to work even if amd chooses to stop actively supporting it outside of our own testing we've gotten word of fsr working on cards at least as old as the nvidia gtx 900 series and it could go back further it should also work on the intel iris xedg1 video card that we recently reviewed showing how wide of a support spectrum fsr offers time for the comparison screenshots we're going to put a couple of different fsr options on the screen alongside native right now but for now we're not going to label them the options we're showing are native resolution fsr balanced and fsr performance the point though is to allow you some time to try and identify the differences between them while we talk through this next section we're trying to help you identify your own biases so you can remove them hopefully when viewing the comparisons later for all of our comparison screenshots we maximize all the in-game settings except for post-processing effects which we minimized or disabled when possible anti-aliasing was also maxed out where available and our intention for this was to provide fsr with the best possible input and apply as few post fsr effects as possible capture was taken on our gpu test bench with a sapphire rx 6800 xd nitro and all of that explained time to show the labels for each of those settings comment below if you got it right or how wrong you got it and was up first the brightly lit foliage of no1800 presents a challenge for fsr with lots of noise in the form of one pixel shadows this is the kind of detail that's in the most danger of being lost if detail isn't visible at all in the low res source image that fsr upscales evasar obviously has no chance to reconstruct it even at ultra quality this loss of detail is visible in the grassy field at the upper left of the screen where the small stones and the blades of grass are unavoidably blurred the horizontal fence railings behind the building at the center of the screen are only partially visible even at 4k native and are promptly erased when even the highest fsr quality level is applied still ultra quality looks close enough to native 4k that the difference might not be immediately obvious without the side by side comparison and that's the main goal if the player can't notice the loss in gameplay then the output is good dropping down further to quality the trees become muddier with less distinction between leaves and branches and so all these things compared ultra quality appears to be the ideal mix for ana fsr is a clear improvement over the source resolutions that it works with comparing anno at 4k with fsr performance mode to ano running at 1080p native demonstrates that fsr can successfully highlight details in areas like the stone and the wood textures of the docks the second scene in anna is more favorable the brightly colored and clearly delineated houses and docks are better suited to upscaling with fsr than the fine details of the forest and the roofs of the houses look almost the same at quality as they do at native 4k notable exceptions are the tile roofs on the two buildings at the top of the village and the striped red and white tent at the center of the screen the closely placed parallel lines get blurred with lower source resolutions and at the balanced setting fsr actually reconstructs the stripes running in the wrong direction these artifacts aren't present when running the game at 1080p which is the source resolution used for fsr performance and the artifacts that were seen are similar to some of the ones we've seen with dlss in the past and that's because the root cause of the issue is the same which is limited source information the rifts breaker has some of the same loss of detail that anno did in areas like palm fronds and blades of grass and certainly youtube compression isn't doing it any favors ultra is mostly successful however in preserving image quality the strangest and most dramatic change is in the shadows the rift breaker is the only game we selected for comparison that supports ray traced effects with rt soft shadows and ambient occlusion both of which we maxed out as fsr drops down from ultra quality to performance each tier removes more of those shadows making them a lighter and less well-defined terminator a game that presumably exists and which has a thriving player base of about 26 players yesterday is one of the strongest examples of fsr that we've seen other than the fact that us running it on two computers increased its active player counts by eight percent the game is interesting for other reasons the beginning of the game takes place at night in the ruins of pasadena california we could hardly tell a difference between the post-apocalyptic hellscape and terminator resistance the dark concrete textures aren't incredibly sharp and detailed to begin with so even performance mode looks good in this first screenshot with the most obvious giveaways being the usual ones a chain link fence for example in the center of the frame and the wooden vents behind it dropping the render resolution to 50 reveals that parts of the chain-link fence are aliased out at lower resolutions which explains why fsr has some difficulty reconstructing it looking at just this scene it'd be difficult to detect the differences between native and fsr quality and balanced would be an acceptable compromise for a higher frame rate we selected a second scene in terminator with an npc clearly visible so that we could see how fsr handled the fabric and skin it held up just as well as it did in the previous scene if not better balanced erases some detail around the npc's collarbone but all the detail in his face is still clearly visible again the fact that terminator's textures aren't especially high-res to begin with helps it here but that's still a valid result detail in wood paneling and beams in the background is also surprisingly well preserved there's a lot that we can benchmark with amd's fsr and we've only just gotten it recently so we'll be doing more follow-up pieces in the future we're keeping it simple for our initial route we were most interested in image quality which we already talked about and our secondary interest was in overhead cost and scaling between the settings types to better understand the performance benefit or the cost from each option we'll likely expand testing in the future but for now we start with something that we don't think any or many of the other videos will look at for today and that's the new amd ryzen 7 5700g apu that we reviewed recently and that technically isn't out for the diy market amd fsr could feasibly allow these types of parts to better serve as a stop gap in between dgpu purchases with our limited time for this one this made the most sense to provide a unique angle and a look at the truly low end below what you would get with the gtx 1060 or rx 580 that amd's already sort of showcased in its keynote we'll start with rift breaker in rift breaker at 1080p and with the mix of medium high and low custom settings that we applied we measured 46 fps average on the 5700g when run without fsr moving to ultra quality or a change to 1477 by 831 the 5700g then improved by 28 the improvement with quality mode or 1280 by 720 was 50 over the original stock result or an additional 18 over the previous ultra quality mode in exchange for 300 000 fewer pixels in the source material balanced mode boosted to 79 fps average with performance jaunting to 91 fps average but amd doesn't generally recommend performance mode to begin with and with these settings and the image quality we don't either at 1080p performance ends up working off of just 960 by 540 pixels or 518 400 pixels for the source information as opposed to over 2 million for native 1080p low scaled proportionally with all the results we didn't notice any odd behavior with frame time pacing from and the fsr so on the side of frame time consistency things look good for simplicity we also made this quick percentage scaling chart the blue line shows the percentage performance improvement versus baseline the red line shows the percentage decrease in pixels from baseline going from fsr off to ultra quality had the largest single increment in performance uplift at 30 percent tied with going from balance to performance at the same time going from fsr off to ultra quality resulted in the biggest single reduction in pixel count by percentage falloff we dropped 41 percent of our pixels in the source material but gained 30 percent in performance it's clearly not perfectly linear things do smooth out though with quality settings we gained 50 percent over baseline fps and dropped 55.6 percent of our baseline pixel data balanced is also fairly even but the performance line gain crosses back over again where we see performance mode rendering one quarter of the pixels in exchange for double the performance when tested with these settings that's not the 2.4x number amd cited but that number may have been based on something else 4k instead for example no 1800 at 1080p low with the r75700g had baseline at 74fps average with for tap aaa we ran with anti-aliasing on for this one despite initial tests with it off because the amd's fsr kept defaulting it to 4x you can force it off if desired but it's the recommended setting and after looking at it there's a good reason for that that results improved to 90 fps average then 102 in the bar chart then 116 then 132fps with performance mode we needed another one of the percent scaling charts to help out though no 1800 again shown here follows the same rules as the rift breaker chart the highest cost of source material is experienced at the offset with ultra quality but we only gained 21 more performance versus baseline with this configuration from there things scaled mostly linearly and at a steep curve getting steeper when moving from balanced to performance mode performance mode is a waste here you aren't going to need super high frame rates and nano to begin with but even if you did dropping 75 of the source material which is what's used to reconstruct the image for a 79 uplift just isn't a good trait because frame rate isn't that important in nano we also ran out 1080p medium settings on the r75700g and then we checked it just the quality mode against baseline off this change got us up to 65 fps average similar in scaling to the lower settings and that moves from 42fps to a more playable result lows improved in staff moving to 33 for 0.1 lows for the 5700g this is the more realistic scenario with fsr we've been able to enable higher quality settings going again up to medium instead of low for everything at the exchange of resolution but we get the improvement with fsr and so the end result is a more playable frame rate with a better image quality next a quick look at overhead testing at 720p native versus 1080p at quality should give us a like for like resolution comparison but super resolution in this comparison creates a better quality image that of course has a cost we can only speak for lower end hardware in this quick test but the 5700g ended up with a 15 overhead shown here between these two benchmarks with the end result being a better quality image of course so that's it for this one then overall fsr appears to be it's in an interesting position because when dlss came out the first time we had a lot of problems with it and the main problems were pretty simple one was image quality was not good it was very blurry it was like vaseline all over the screen it just didn't work very well and the second problem it was locked to hardware the third problem it wasn't even really using the hardware that it was locked to it was an effort to sell more rtx cards when rtx was a new initialism it didn't exist before that was gtx and so as nvidia tried to build hype and marketing and interest around rtx and this whole new software suite and paradigm shift in the company's marketing it locked down dlss to rt devices but it wasn't always really using the rt hardware that was kind of the tricky thing about it so it was a bit misleading dls 2.0 came out and actually in the games where it works like control for example is extremely good it does a great job at conquering a lot of the same challenges that fsr tries to conquer except in a way which is just objectively more advanced and you could say using computer terminology that marketers like to use smart or intelligent but that doesn't mean that fsr is bad it just means that nvidia is doing something in a much more complicated way so as long as fsr can avoid certain things like really bad blurring and the vaseline effect then it's in a good spot and thus far the the biggest downside to fsr is clearly the games list but it has a lot of upsides and the number one upside is that's not hardware locked it does work with older gpus from both vendors it's only been validated officially by andy back to gtx 1060 and rx 480 or 470 era for amd's devices but you can run fsr on older stuff than that we haven't tested too far back to see at what point it just doesn't work but at least the gtx 9 series should work and the older than rx 400 series should work as well so that's cool and that's really good place to be you do still lose quality with this that's always going to be the case so amd was really stretching the truth when it said in the reviewer's guide that sometimes the image quality is better than native but amd in that sentence you can you can read the guilt in the marketing bs from whoever wrote that because there's just one qualifier after another of when it might be considered better than native and ultimately it can't really be better than native with this approach you might perceive it that way subjectively just based on whatever specific game specific device specific settings maybe you'll perceive it that way especially if you like sharper images but uh that is getting pretty far into the subjective at this point so just to be really clear there is going to be data loss when you are rendering at a lower count of pixels and then bringing it up to something else it's not something to be really ashamed of unless it's terrible and in the instances of high quality and quality amd does pretty well to hold on to a lot of that detail you start losing the small shadow details the one pixel wide details go away if they're not rendered at the lower resolution clearly because you need something there in order to bring the data up into a a higher resolution but ultimately it's a good first start but it's very weak in the games department terminator 26 players yesterday and i'm pretty confident that most of them are people i know who are also reviewers i don't think they're that many real players sorry if you were one of them but uh hey you were online at the same time as a bunch of reviewers at least other than that anno seems like maybe the the most legitimate or the largest of them rift breaker looks like it has a little bit of interest around it so they're getting out there but amd really needs to get into stuff like cyberpunk cass is in cyberpunk and yet fsr is conveniently absent from it why is that dlss so amd is trying very hard right now to avoid dlss comparisons and that's the part we really want to see is as it starts spreading when will amd become comfortable enough for confident enough in its product that it is willing to put it into a game that dlss is also and that's the comparison we want to see even if they're not directly linearly the same they're sort of trying to tackle some of the same issues though and that's why it's interesting anyway good first start needs more games and uh until it gets them we can't really say much more about it but from a technology standpoint appears to be working more or less as it should be just performance mode looks bad probably don't use that one but high quality is not bad so that's it for this one thanks for watching as always subscribe for more you can go to store.gamingtexas.net or patreon.comgamersnexus helps out directly in these endeavors and we'll see you all next time\n"