In a marketplace where getting a graphics card at a fair price is an exercise in pain management, GPU manufacturers have been coming up with inventive new ways for gamers to get more performance out of the graphics cards they already own. These methods include sharpening filters, AI-assisted image-scaling options, and supersampling tech.
That’s a lot of jargon; get ready for another burst of it. The five major options available today use some combination of technologies, techniques, and innovative algorithms to, in theory, give you more performance out of almost any current GPU without sacrificing visual quality in the process. They are Nvidia’s DLSS and Freestyle; the open-source ReShade plugin; AMD’s Radeon Image Sharpening (RIS) software; and AMD’s newest approach (launched on June 22), dubbed “FidelityFX Super Resolution” (FSR).
But the only way to know which option is the best for you and your setup is to dive into testing and see how these five approaches hold up in quality when they go head-to-head. So check out our complete breakdown of all things image-sharpening, anti-aliasing, and DLSS-ing. This field is constantly evolving, so here’s a look at the state of play in the second half of 2021. Which one could be right for you and your hardware?
TLDR? Fewer Jagged Edges for Less Money
So what are sharpeners, upscalers, and supersamplers, exactly? And how have they evolved since they first hit the scene almost half a decade ago?
In a sentence, every feature mentioned here shares the same goal: Get higher frame rates out of your system on the same hardware, without sacrificing visual quality in the process. It’s a bold new approach to the major slowdown in Moore’s Law. These technologies apply software, AI, and algorithms where increased transistor density just can’t pick up the slack like it used to.
Before we get into our anecdotal testing results, which show how those efforts are playing out, a quick disclaimer. In the time since the first third-party sharpening tools, such as ReShade, were introduced, Nvidia’s DLSS tech, one of the best known, has emerged, then gone through a full revision. (Version 1.0 is now 2.2.) Meanwhile, AMD’s Radeon Image Sharpening gained almost all the features I complained were missing in our original breakdown of these technologies a bit more than a year ago. And open-source ReShade (with the FidelityFX CAS sharpening filter) has been integrated right into Nvidia’s Freestyle—something I don’t think anyone could have seen coming. And then there’s AMD’s FidelityFX Super Resolution (FSR), which emerged a week ago and was the trigger for this fresh look at these technologies.
To help keep track of where everything stands, let’s check out a quick primer on what’s out there, and who can use it…
In our original test run of some of these services last year, AMD’s Radeon Image Sharpening walked away with the first-place trophy, clearly beating Freestyle and DLSS 1.0 in both clarity and performance, with the least amount of artifacting in the bunch. At the time, DLSS 1.0 had a quirky problem of smudging lines and making the render look blurrier than it was with the feature off; that’s a no-go when it comes to gamers and their graphics settings. As of that writing, AMD had a clear lead over the rest of the sharpening pack in our “eyeballing-it” quality tests (though the relative market share of Radeon GPUs versus GeForce ones is whole another issue). And at least at the time, it didn’t look like it would be all that close for the foreseeable future.
But things have changed since then, more than most folks could have predicted. So did AMD maintain its sharpening lead, and even extend it, with its launch of FSR? Or has Nvidia made some proper headway in the fight? Let’s get into our quality assessments to find out.
Eyes-On With Faster Frame Rate Technologies
It’s time to pit each technology against one another. In some cases, that involved, in essence, “simulated resolutions”—for example, displaying at 4K resolution, but actually rendering the scene at 1440p then upscaling to 4K.
I looked at each technology, where relevant, across 4K gaming results that were both natively rendered (that is, at actual 4K), as well as scaled from various resolutions below that. I’ve assembled a trove of images below (they’re hosted in sliders; click the arrows!) that show the same scene in various games (support-dependent), processed by each individual sharpener.
We used PC Labs’ graphics-card testbed to grab all screenshots. It uses an Nvidia GeForce RTX 3080 Founders Edition card (or an AMD Radeon RX 6800 XT, depending on the needs of the test), an Intel Core i9-10900K processor, 16GB of Corsair Vengeance DDR4 memory, a solid-state boot drive, and an Asus ROG Maximus XII Hero (Wi-Fi) Z490 motherboard.
Because we had to use two different cards (and multiple games) to test these features to their full extent, this won’t be so much a “performance” comparison as it is strictly a look at the output image quality of each technology. We chose the game Control to show off what DLSS and the three sharpener technologies can do, while compatibility issues forced us into testing with the titles The Riftbreaker and Godfall for AMD’s FSR. (Control does not support FSR.)
Control also gives us the most amount of…ahem, er…control, in testing. That is because of the higher level of flexibility it offers when scaling up or down a specific resolution, as well as the intensity of ray-traced light in the scene.
AMD Radeon Image Sharpening (RIS): The Broad Approach
RIS PROS
- Best visual-clarity results with the least artifacting
- Works in any game running on DirectX 9, DX10, DX12, or Vulkan
- One-toggle setup
- Works with most recent AMD Radeon cards
RIS CONS
- Not supported by very lowest-end Radeon cards
- Doesn’t work on DirectX 11
- Doesn’t work with Nvidia-branded GPUs
AMD’s Radeon Image Sharpening (RIS) came a bit later to the party than both Freestyle and ReShade. (The feature made its public debut with the release of the Radeon RX 5700 and Radeon RX 5700 XT in June of 2019.) AMD says that its RIS technology does everything that Nvidia’s DLSS can do…though these claims have since been superseded by the marketing around AMD’s new-for-2021 FSR. (More about FSR in a bit.)
AMD is (rightfully) cautious to call RIS an “anti-aliasing” technology, despite the fact that it was developed by the same person who created both Temporal Anti-Aliasing (TAA) and Fast-Approximate Anti-Aliasing (FXAA). TAA and FXAA are both methods designed to smooth out the jagged roughness that is produced around the edges of objects or characters in 3D video games. In contrast, RIS is a post-processing sharpening filter that is applied at the API layer, and the effect appears only after every part of the image has already been rendered by the GPU. Based out of a larger open-source developer toolkit that AMD calls FidelityFX, RIS is an automatic process that sharpens the image of a game and removes some of the “fuzziness” around the edges of models that can occur at lower resolutions.
RIS accomplishes this through the use of a technique known as contrast adaptive sharpening (CAS). CAS tells your GPU to look for areas of stark contrast (say, between the outline of your main character and the background of a jungle), and it uses that contrast to inform which areas of the image should be sharpened. The math that makes this all work is extremely complex, but like DLSS, the main idea of RIS is simple: Make games running at lower resolutions look as close to, or just as good as, a resolution tier above it. (See our hands-on guide to running RIS.)
We’ll get into how it looks to us in a moment. But first, we should mention that RIS takes the compatibility cake versus DLSS. RIS runs on any game that is based on the DirectX 9, DX10, DX12, or Vulkan APIs, on just about every AMD video card from the last three generations. However, unlike FSR, which is compatible with both Nvidia and AMD GPUs, RIS works only with AMD-branded cards. The parameters of RIS open up the technology to a much, much larger number of titles (numbering in the thousands), with more being added every day as new games debut.
RIS: Quality Testing With the Game ‘Control’
We tested with the AMD Radeon RX 6800 XT installed. Note: In this test and all following screenshots of Control, try and look at how the character’s hair blends with the American flag on the left for the best indication of how an upscaling technique affects image quality. Enter “Fullscreen mode” to see more detail by clicking the icon in the top right for a full-size image.
Just like when we tested RIS the first time, RIS continues to hold up well under scrutiny. With the game rendering at a resolution of 2,560 by 1,440 pixels (downscaled from 4K), I set the image sharpening to the 30% and then 50% levels, using the slider in AMD’s Radeon Settings application, and launched into Control…
Although the image wasn’t as refined or as crisp as a DLSS image rendered at the same resolution, I’d say it was close enough for my purposes. As for an increase in performance, I saw a ratio gain similar to that of DLSS, but still a bit short: a 128% improvement (14 frames per second at native 4K resolution, versus 32fps with RIS enabled).
Overall, the number of artifacts remained low both in screenshots and during high-motion action scenes, which was RIS’s biggest win over Nvidia’s Freestyle the first time we tested these features back in 2019. Has anything changed for Nvidia’s sharpening tool this time around?
Nvidia Freestyle: The Other Way to Go Green
FREESTYLE PROS
- Works with all Nvidia cards
- Easy implementation through GeForce Experience
- Now has integration with Reshade and FidelityFX CAS
FREESTYLE CONS
- Generates more artifacting than RIS
- Poorest visual results of the sharpeners
- Compatible with just over 900 games (but anything outside that list won’t work)
Shortly after AMD announced RIS (we’re talking a literal day between press briefings), Nvidia showed off a fresh iteration of its Nvidia Freestyle image-sharpening tool to select journalists at a closed-door event at E3 2019. (Freestyle first rolled out in January of 2018.)
Only part of what Freestyle does is sharpening; it also lets you apply filters to your game to change the overall look. Nvidia hasn’t shared much about how the technology behind Freestyle works, stating simply that it’s an “image post-processing tool” that sharpens the edges of objects in your games. (See our guide to running and using Freestyle.)
Nvidia’s list of compatible titles that work with Freestyle is smaller than AMD’s RIS list, but it’s not small: around 900 games and growing. That’s many more than DLSS can touch, but far fewer than RIS. So how does it stack up in quality?
Freestyle: Quality Testing With the Game ‘Control’
We tested with the Nvidia GeForce RTX 3080 Founders Edition at Ultra settings. The first shot is at native 4K, and the following two are at 1440p with sharpening applied. Reminder: You can enter “Fullscreen mode” by clicking the icon in the top right for a full-detail image.
Compared with the last time I tested Freestyle, I’d have to say there hasn’t been enough improvement in the technology to warrant recommending it over ReShade or RIS on its own, even with the newer “Sharpen+” filter applied, which integrates ReShade with CAS directly into (It was released as a part of a driver update from Nvidia that was pushed in late June 2021.) Freestyle still introduces a host of artifacts and jagged edges into scenes with lots of action, and although the intensity slider found in the Freestyle filter control panel can help with this, the only point where the artifacts disappear is when you set it around the 15% mark.
With the test game Control sharpened to 30% and down-rendered to 1440p, I was able to get a “4K” version of the game running at roughly the same frame-rate gains that I achieved with both RIS and ReShade. During testing, the game peaked at around 40fps in simulated “4K.”
Remember, though, that quality is a far more important factor in this battle than raw frame-rate performance. A game upscaled from a 1440p render to a 4K image will almost always run as quickly as a game natively rendered at 1440p, regardless of the hardware or any sharpeners applied on top. If the game looks like it’s being run at 4K while it’s being rendered at 1440p under the hood, that’s where the real difference lies.
Now that Freestyle supports an integration of ReShade with FidelityFX CAS, however, the results for Freestyle, Radeon Image Sharpening, and ReShade can be said to be just about the same since they’re all based on the same underlying CAS algorithm. What’s this ReShade we keep mentioning, you ask? Well that’s back where this whole thing started…
ReShade: The Open-Source Alternative
RESHADE PROS
- Works on all games and with all video cards
- Visual results were good (if not as good as what we saw with RIS)
RESHADE CONS
- Complicated setup process
- Must be installed individually for each game you want it to run on
The aftermarket, open-source post-processing software ReShade is an outlier, and it’s often changing. Since I’ve been investigating these sharpening tools and upscalers, ReShade added contrast adaptive sharpening—the same “CAS” mentioned, earlier specific to AMD—to its sharpening arsenal.
CAS is one algorithm implemented as a part of the larger FidelityFX developer shader toolkit, and it is primarily responsible for helping AMD’s RIS determine where the edge of a character in focus ends, and the background of an image begins. Because FidelityFX is open-source, the developers behind ReShade were able to implement CAS into their own software easily.
Unlike every other tool on this list, ReShade is API- and GPU-agnostic. If you have the time and the know-how, you can get it to work in the game of your choosing on the hardware you already own.
That said, the same thing that makes ReShade work on every game and with every GPU is ultimately its main drawback. Unlike RIS, Freestyle, FSR, or DLSS, which can be activated automatically at the flip of a switch, ReShade involves a somewhat complicated installation process.
Not only that, but each game must be configured individually for ReShade, a far cry from the “set it and forget it” ethos of AMD’s RIS and FSR, and Nvidia’s DLSS and Freestyle. That said, the benefits of universal compatibility outweigh the complexity of setup for really serious tweakers. The ReShade integration of these sharpening tools means they’ll work on any game, powered by any card—period—that you’re willing to put in the effort with. It also bears repeating that Nvidia’s Freestyle now comes with ReShade with CAS integration, so the two should be considered one and the same if you’re using the “Sharpen+” setting that was just introduced.
ReShade lets you choose from values ranging from 0 to 100 on a sharpening intensity scale, just like Nvidia’s Freestyle. Tune it just right, and in theory you can create a sharpened 1440p image that looks just as good as native 4K, all while gaining up to 30% in performance. What’s not to like? Well…let’s see!
ReShade: Quality Testing With the Game ‘Control’
We tested ReShade with the Nvidia GeForce RTX 3080 Founders Edition at Ultra settings. Note: Enter “Fullscreen mode” by clicking the icon in the top right for a full-detail image.
To my eye, ReShade was better than Freestyle in these tests, but just slightly under what RIS could do. That makes sense when you consider ReShade, with its CAS integration, is doing more than just flatly sharpening the image (like what’s going on in Freestyle). While a sharpener like Freestyle is applying a simple post-processing filter over the image, CAS actually goes in and algorithmically reads the image to see what elements can be sharpened and which can’t without leading to artifacts. That said, since ReShade can be integrated into Nvidia’s Freestyle menu, if you so choose, it’s kind of a moot point comparing the two.
The performance gains were nearly identical to the rest of the sharpeners, an increase of 18fps at native 4K to 46fps at 2,560 by 1,440 with a 35% sharpening intensity applied, which in any title, not just Control, is an astounding jump, all things considered.
Next up, Nvidia’s elephant in the room…
Nvidia DLSS 2.2: The Machines, They Are Learning
DLSS 2.2 PROS
- Increased performance vs. native resolution in Quality and Balanced mode with no visible loss in render quality
- Can actually improve the visual quality of certain elements onscreen when running in Quality mode
DLSS 2.2 CONS
- Limited current breadth of implementation, both in hardware and number of games supported (55 games, nearly three years post-launch)
- Requires a GeForce RTX video card
If you want the full specifics of how Nvidia DLSS works, you can read about them here. (It’s a lot.) In brief, the concept of DLSS is pretty brilliant: Take one of the most taxing tasks in gaming, namely anti-aliasing, and offload the workload to an AI supercomputer. (It’s so simple, why didn’t one of us think of it first?)
DLSS works at most major resolutions, including 1080p, 1440p, and 4K. However, the performance boosts will be most noticeable for those playing on the last two. For every step up or down in resolution, DLSS chooses a “true” render resolution that corresponds to the level of quality you select. So, say for example, you’re running the game Control in 4K with DLSS turned to Balanced mode; the game is actually rendering the engine in 1440p, and then upscaling the image back to approximated “4K.” When the magic trick works well enough, you shouldn’t be able to tell the difference between the 1440p render and what the game would look like while natively running in 4K.
In our last article, linked above, we tested the performance of DLSS 2.0 and found that at its most extreme, DLSS could offer absolutely staggering performance boosts of up to 184% in games that support it. (In that case, the game was Control.) This was substantially higher than any of the boosts we saw in testing DLSS 1.0, and it holds the promise to change much of what we know about the relationship between new cloud-based graphics rendering techniques and local discrete GPUs in the years to come.
DLSS 1.0, 2.0, and 2.2 all share one similar caveat, though: You have to own an Nvidia GeForce RTX card for them to work. Although DLSS is a feature that can be toggled within the graphics menus of your games, it will only show up as an option if you’ve got one of a select number of video cards offered exclusively by Nvidia installed in your desktop.
Every graphics card carrying the “GeForce RTX” badge has three kinds of GPU cores inside: the main GPU cores, the “RT” cores (which power ray-tracing), and the “Tensor” cores. It’s the last that handle DLSS. If you have a card in the GeForce GTX line, DLSS is not an option for you.
And with that introduction out of the way, let’s head into some DLSS testing…
DLSS 2.2: Quality Testing With the Game ‘Control’
We tested DLSS 2.2 with the GeForce RTX 3080 Founders Edition card at Ultra settings. Note: Enter “Fullscreen mode” by clicking the icon in the top right for a full-detail image.
Though I’d never be able to fully articulate how they managed to do it, the engineers at Nvidia have pulled off the seemingly impossible: make a game look even better than it does at native resolution, and run faster than native resolution, at the same time.
Looking at the image treated with DLSS, it’s clear that the new network has visibly improved over DLSS 1.0 in how it renders the image, and areas of focus that it determines are important. In testing, I found that DLSS’s Balanced mode, perhaps predictably, offered the best balance of performance and quality, managing to squeeze 147% performance gains (from 19fps at native resolution to 47fps in Quality mode) out of the same GeForce RTX 3080 Founders Edition card.
With DLSS switched on, the edges of objects in the game looked more refined, text was easier to read, and everything just looked that much cleaner and crisper than with it turned off. DLSS 2.2 finally delivers on the promise of better graphics with faster performance, and all it took was a little AI sprinkled in to make the whole thing work.
A DLSS Caveat: Mind the Tensor Tax
Before we finish out this analysis section of DLSS, though, we also need to consider the variable of Tensor cores, and more specifically, their cost.
Both current- and previous-generation AMD Radeon cards now run RIS, and Freestyle works across almost every card Nvidia has to offer. Reshade works on any card you throw its way. DLSS, on the other hand, works on just one tier of graphics cards from one manufacturer, which also happen to be the most expensive GPUs that Nvidia sells. (And that’s just at MSRP; it takes cunning to even buy one these days and not get fleeced.)
So does the Tensor core finally live up to its own value proposition, several years after its initial release?
DLSS is a highly innovative approach to the comparatively ancient problem of anti-aliasing, and one that could revolutionize the relationship between raw gaming hardware performance and cloud-based AI computing on the road forward.
But as of this writing in June of 2021, it works on just 55 games. (See the full updated list at Nvidia’s site.) That isn’t nothing, but it’s nowhere near the tens of thousands of independent and AAA titles that people play every year.
So here’s the skinny on DLSS today: If you play a lot of Control or multiplayer games like Call of Duty: Modern Warfare, you own an RTX GeForce card, and your game is on the DLSS list, DLSS will be the best thing to happen to your gaming in a long, long time. The value proposition of paying for the Tensor cores on your RTX card is huge. Plus, once the DLSS training network really starts to get off the ground, I could see developers flocking to Nvidia’s neural network for the free performance gains it affords their titles, and the increased visual fidelity in their engines that only a technology like DLSS can provide.
The gigantic elephant with DLSS, though, is that limited list of titles and the scarcity and cost of RTX cards. Otherwise, though, the other sharpeners are what we have available to us today in 99.99% of games and most GPUs.
Enter AMD to save the day…maybe?
AMD FidelityFX Super Resolution (FSR): DLSS for All (One Day, Perhaps?)
FSR PROS
- Works on a much larger list of GPU hardware than DLSS
- Visual results are strong (at least until you engage Balanced mode)
- Healthy performance uplift with little loss in visual clarity
FSR CONS
- FSR support in a mere eight titles at launch
- Performance mode has more visual degradation than equivalent DLSS setting
Finally, we come to the most recently released upscaling/supersampling tech: AMD’s FidelityFX Super Resolution (FSR). As DLSS has steadily trundled forward in support, AMD has been in the wings, watching for its best opportunity to release what could be, if adopted widely, the DLSS challenger Nvidia might fear.
Launched as an addition to AMD’s growing library of tools contained under the FidelityFX umbrella, FSR uses “spatial” aliasing (as opposed to DLSS’s “temporal” aliasing) to reconstruct lower-resolution source images and upscale them to your preferred resolution. Much like CAS, the game will automatically—depending on the developer’s input—select a resolution to render the game at on the GPU, depending on the quality setting, and then upscale that lower resolution to match the output of your monitor using a heck of a lot of fancy math.
So, for example, when FSR is running in its 4K Quality mode, it’s actually rendering the game at 1440p. Then, using its spatial aliasing algorithm to reconstruct the lost information around sharp edges and areas of the image with finer detail, FSR helps rebuild the image in a way that becomes less noticeable while characters or backgrounds are in motion (which they often are in a video game, barring, say, long-winded RPGs).
While it is not strictly based on the same algorithm that powers CAS, according to AMD we should consider FSR an “upgrade” to that same tech, though it uses a different technique to extract the data it uses during the upscaling process. Integrated as a part of the shader queue into the rendering pipeline for a GPU, FSR will be compatible with most GPUs currently on shelves (including many Nvidia GeForce cards, which came as a shock), as well as several Ryzen-based APUs that utilize Radeon RX Vega integrated graphics. To see the full list of supported hardware, scroll about halfway down AMD’s reference page linked here.
While its hardware support is significantly broader than Nvidia’s DLSS, the number of games that support FSR at the time of this story (late June 2021) is…not many, just like DLSS in its earlier days. To be exact, it’s a whopping, ahem, eight…
You’ll see seven above in the left panel, and at this writing, DOTA 2 had just added it. (The list may be longer when you read this.) If you don’t recognize most of the other titles mentioned, don’t worry, most people wouldn’t. That said, AMD looks to be pushing forward aggressively, announcing upcoming support for big AAA titles like Far Cry 6 and Resident Evil: Village, two games that could benefit from having their beautiful visuals pushed a bit further in performance.
FSR, like CAS and many of the other FidelityFX features, will be made available as a part of the GPUOpen suite of open-source developer toolkit. This means it will be much easier for developers to integrate FSR into their games than what it takes to train up a game for DLSS. Using that slower training methodology, DLSS took years to get to its current supported list of 55 games; with support for GPUOpen, it could be just a matter of months before FSR surpasses that. We’ll have to see.
The possible clincher? Unlike DLSS, which requires specialized training per-game on Nvidia’s neural net, FSR integration is so easy to do that it’s also been announced as a part of the Xbox Game Development Kit. That will make it available to anyone who wants to publish games on various consoles in the Xbox family. (Could this finally be the tech that fixes Cyberpunk 2077’s frame rate on Xbox One?)
But enough about how FSR works. Let’s get down to the moment of truth: How does FSR look when we rolled it out with two of the supported titles provided to us by AMD: Godfall, and The Riftbreaker? Can it hold a candle to DLSS?
FSR: Quality Testing With the Game ‘The Riftbreaker’
We tested FSR with the Nvidia GeForce RTX 3080 Founders Edition at Ultra settings. Note: Enter Fullscreen mode by clicking the icon in the top right for a full-detail image.
FSR: Quality Testing With the Game ‘Godfall’
We hen tested FSR again in Godfall with the Nvidia GeForce RTX 3080 Founders Edition at Ultra settings. Again, note: Enter Fullscreen mode by clicking the icon in the top right for a full-detail image.
Ultimately while it’s our job to nitpick these images to death here, a couple of things to keep in mind. First, most of the shimmering and artifacting we noticed in the images above, especially once you get to Balanced and Performance modes, are far more obvious in screenshots than they are in actual gameplay motion. Second, FSR is using a less advanced reconstruction method than DLSS, so ultimately the quality of FSR will always be dragging slightly behind.
That said…you really have to be looking for it to notice it at all. FSR does a fantastic job of effectively coming within spitting distance of DLSS quality, without Tensor cores or fancy AI training required. It manages to capture that same sense of “magically” watching your frame rate go up while the visuals stay the same, although the illusion starts to fall apart if you engage Balanced and Performance mode.
Quality mode looked great, though, and still offered some serious performance boosts. In our testing on Godfall, FSR took our benchmarks from a native 4K render result of 56fps, up to 85fps when it was turned on, representing just over a 50% gain in frame rate!
Once we went down to Balanced and Performance modes, smudging became more of an issue. The algorithm would sometimes reconstruct distinct vertical lines into a single object, though this happened more often in Performance mode than Balanced mode.
Overall, FSR would seem by far the best option for the widest number of hardware owners, and the one to watch the most closely as 2021 progresses. Problem is, its currently tiny library of supported games will keep it out of most people’s field of vision until more developers start to integrate the FSR option into their titles. Luckily, anyone can help contribute to picking out which games get FSR ahead of the others by heading over to AMD’s FSR “Wish List” survey, and requesting the title of their choice to be considered.
You Gotta Look Sharp! So, Who Wins?
First, let’s dispel the notion of “winners and losers” in this comparison piece. With so many varying levels of support, both from a software and hardware perspective, which tech is right for you will ultimately depend on the hardware you own, the games you play, and your level of personal technical know-how. Anti-aliasing has come a long way in a short amount of time in the past few years, but it seems it took a kick in the “CAS” to get things moving in a new direction where two technologies—one old and one new—work together to give gamers better graphics at higher frame rates in 2021.
In quality testing for the sharpeners alone (ReShade, RIS, and Freestyle), I found that RIS still produces the cleanest-looking images, to my eyes, followed closely by ReShade. Freestyle is fine for what it is, but it introduces too many smudges and artifacts in fast motion scenes to recommend as the best pick.
That’s why it’s hard to call an objective, unequivocal winner on this battleground. Some techniques work on more games, others use more advanced technology. And others still might be more complicated to set up than most users are willing to bear. But, ultimately, none of these factors matters when the end result comes down to what hardware you own and your level of technical expertise. (Especially since video cards are so hard to come by, and so dear, these days.)
To my eyes, DLSS consistently did the best job with the game Control, but because of the limitations in support for that game (it doesn’t do FSR), we weren’t able to test that quality in a 1:1 comparison against AMD FSR. There are no DLSS and FSR titles in common at the moment. In some cases, a game can be tuned in DLSS to give you the same results from an RTX 2060 Super as an RTX 2080 Ti without it. And if you walked away from our tests scratching your head because you couldn’t see a difference between the native 4K image and the lower resolution versions with upscaling, supersampling, or sharpening applied? Good! That means they’re doing their job.
DLSS, in particular, paints one vision and approach to these technologies. It’s that of a closed-loop system where constant communication and coordination between Nvidia and game developers will be the only path through which gigantic gains in performance for AAA games (say, to run at 8K on a PC, or to run at 4K-like quality with lower-end hardware) might be achievable in the next few years. The launch of AMD’s FSR could start to shift the narrative away from that top-down approach, thanks to its GPUOpen compatibility. But at the time of this writing, if you want to achieve the same effect on 99.9% of games today, you’ll need to stick to a sharpening solution like ReShade, RIS, or Freestyle.
ReShade with CAS integration stands out thanks to the fact that it works universally across all hardware and software, and it comes in a close second on visual fidelity in the sharpener department. Our main caveats with ReShade: There’s still a bit of noticeable jaggedness that appears when you push above the 50% mark, and getting it installed is no simple feat. This is where RIS’s application of the CAS algorithm seems to be the most…crisp, if that’s a word that applies. It regularly won the battle of the sharpeners on every game and resolution we tested at. We’d use RIS where we could, but RIS works only on AMD Radeon hardware, which limits its broader appeal.
At most percentage levels, Nvidia’s basic installation of Freestyle came in last place. It introduced too many choppy edges into the test image that made it look worse than an image that had no sharpening applied at all, though these issues were least noticeable when the effect was tuned down to 25%.
The Wrap-Up: Sharpen ‘Em if You Got ‘Em
AMD and Nvidia have, independently of the other, each claimed that their technologies differ from their competitors in X, Y, and Z ways. But from a consumer’s perspective, we see them all as different methods toward the same madness: pushing as many frames out of your GPU as you can, and losing as little visual quality as possible in doing that.
If you take anything away from this piece, let it be this: If you aren’t already using a sharpening, upscaling, or supersampling solution (the one that best suits the hardware you own, and the games you want to play), you absolutely should try one. It’s the best way to get a free performance boost—anywhere from 10% to 70%, depending on your graphics card and quality tolerance—and there’s no reason not to have one running, with all the different options at your disposal.
As Moore’s Law continues to slow down and GPUs become more and more complex to manufacture (and maddeningly expensive to buy), it will take more than just raw horsepower alone to increase graphical fidelity, add ray tracing, and move your favorite games to lofty resolutions like 4K or the coming 8K. Sharpeners, supersamplers, and related technologies—FSR, CAS, and DLSS—help to bridge that gap. And if they’re already this good now, imagine what’s waiting for us just around the corner.
