Games

Frame generation for every game? Lossless Scaling vs FSR 3 and DLSS 3

What if you could double or even triple almost any game’s frame-rate, just by installing a small program onto your PC? It has echoes of downloading additional RAM, but this is exactly what the app Lossless Scaling can now provide using frame generation technology. We’re already familiar with the effects of frame gen with Nvidia’s DLSS 3 and AMD’s FSR 3, so how exactly does Lossless Scaling work by comparison, how do the generated frames look and feel, and what are the best use cases for this new-found functionality?

First of all, it’s important to note that Lossless Scaling’s frame generation is fundamentally different to how Nvidia and AMD’s frame-gen tech works. DLSS 3 and FSR 3 leverage the final colour frame of two rendered images before the HUD is drawn, with motion vectors for both frames used to generate the frame between. Motion vectors are essentially a cheat sheet that tells the frame generation where and how objects are moving, and they’re deliberately exposed to DLSS and FSR frame generation by game developers.

In contrast, Lossless scaling is more like a mod, without any access to the game’s internals. The frame generation step is instead performed as a post-process, giving it access to two fully rendered images – including the HUD and not including the motion vectors. That means the app has to guess what is moving in an image, rather than having that hard data to hand. According to Lossless Scaling developers THS, this guesswork is performed using machine learning to more accurately generate those extra frames.

Here’s the full look at Lossless Scaling’s frame generation tech in motion. Watch on YouTube

Interestingly, the latest version of Lossless Scaling can generate two in-between frames, rather than just one, if you so choose. That means this app can generate more frames than DLSS 3 or FSR 3 can – though it still has the characteristic limitations of other frame-gen examples and additional hurdles from its status as a post-process effect.

Comparing Lossless Scaling against DLSS 3 and FSR 3 in slowed-down footage, it’s possible to see a hierarchy of quality. Lossless Scaling’s generated frames have more visible errors taking up more of the screen; FSR 3’s use of motion vectors ensures it looks better by a significant margin; while DLSS 3 with its machine learning has even fewer errors than FSR 3. The larger the error, and the more this error is repeated over multiple frames, the more likely you are to notice the discontinuity with the naked eye. Even quite large errors might not be noticeable due to the nature of the stroboscopic effect – where generated frames are bookended by traditionally rendered ones – but they are still there.

Using Ghost of Tsushima as an example, you will often see an outline effect around the main character or moving objects that persist in your mind’s eye when things move – almost like a layer of displaced air suddenly becoming visible. Another error that you can sometimes see is a subtle flickering of transparent objects like shadows or particles. These disappear every other frame, making them look almost faded or more raw than they otherwise would. Another issue can be found at the edges of the screen. Generated frames are going to be faking the screen edge quite a lot when a camera moves, so you’ll notice the same bubble of displacement at the screen edge if you consciously look there. The less movement in-game, the less obvious these errors are, and this applies to all frame generation techniques in general.

In motion, Lossless Scaling’s frame generation can look indistinguishable between that of DLSS 3 or FSR 3. Catch it on the right single frame though, and you can that visual errors can be more numerous. | Image credit: Digital Foundry

The last error you see often with Lossless Scaling’s frame generation concerns HUD elements or the mouse cursor. As we covered earlier, Lossless Scaling has no understanding of what is a part of the game world and what is being drawn on top of it, so HUD elements can often appear distorted as frame generation is unnecessarily applied to them. For example, in Turok Dinosaur Hunter I noticed the weapon wheel looked weird whenever I brought it up, and if you pause on a generated frame you can see the errors. This comes with the territory for post-process effects, but it is still jarring.

Moreover, errors increase in severity at higher scaling factors – ie when generating two frames between traditionally rendered frames rather than one. This can be the difference between these mistakes flying under the radar or being quite obvious – something I noticed in Ghost of Tsushima, where camera movement is fluid but larger character movement can break down. Despite this, 40fps to 120 fps frame generation still works, but I think more people will notice issues – so it’s best to use this option when you’re already at a high frame-rate and want to max out a high refresh rate monitor. At 240Hz or more, I think this feature could really sing.

Visuals are not everything though, as frame generation techniques also come with a penalty to input lag. This can drastically affect the perceived responsiveness of a game, and the post-process nature of Lossless Scaling means that it is heavier than other frame-gen techniques. In the best possible scenario – with Nvidia’s latency-reducing Reflex tech enabled in Cyberpunk 2077, running maxed in DLSS quality mode – I measured significant differences in input latency.

Including display latency from an LG C1 OLED, the 2x mode clocked in at 99ms of input lag, while the 3x mode clocked in at 92ms. By comparison, DLSS frame generation was much more responsive – 59ms – while disabling frame generation cut latency to 44ms. In a like-for-like 2x scenario, Lossless Scaling’s latency is nearly double that of DLSS 3.

If you look at all of these negative aspects in isolation, you might be start to think that the flaws would pile up and ultimately Lossless Scaling’s frame generation would be hard to recommend. In my experience though, I found it compelling, functional and genuinely worthwhile – most of the time.

For modern titles that offer FSR 3 or DLSS 3 support, I wouldn’t recommend Lossless Scaling, as it increases latency and artefacting for little gain. For titles that lack official frame-gen support though, Lossless Scaling is a new and often worthwhile option for boosting visual fluidity despite its drawbacks.

This is particularly true for games that are locked to a specific frame-rate – think Elden Ring or Resident Evil 4 at 60fps, or Turok and Freedom Fighters at 62fps. This way, you can get a much smoother-feeling game without breaking the game’s logic, which is tied to its internal frame-rate. Playing titles like this at a perceptual 120fps or even 180fps is brilliant, especially for gamepad-based or otherwise less precise games where the input lag penalty is less critical. There are also some PC games that are locked to 30fps, such as Command and Conquer 3: Tiberium Wars. I love this game, but it looks much less sluggish at 60 or 90fps – even if the added input lag is more noticeable with the fast-paced mouse-and-keyboard gameplay.

These are just some of the games I tried, but Lossless Scaling can be applied to pretty much any game that can run in a window, including emulated games – how about Killzone 2 running at 120fps? Sure, it adds even more input latency to a legendarily laggy FPS, but it’s still great fun.

Lossless Scaling's latency penalty vs FSR 3.1 and DLSS 3.7 - 99ms (LS 2x), 92ms (LS3x), 59ms (DLSS 2x), 44ms (no frame-gen).

In Cyberpunk 2077, it’s clear to see that Lossless Scaling’s frame generation comes at a hefty input lag penalty versus DLSS and native rendering. | Image credit: Digital Foundry

Enjoying Lossless Scaling’s frame-gen tech does require that you keep your expectations in check. Beyond quality and latency concerns, this app cannot fix a game’s poor performance. In Elden Ring, for example, it’s possible to boost the game to 120fps, but this doesn’t fix the game’s long-standing hitching and stuttering issues – you just have to put up with them.

Another thing to watch out for is that you need to configure a game correctly for it to present well. Lossless Scaling only presents fluid motion if the game is running at exactly your refresh rate or below it with VRR – so you either need to have the game running internally at less than half your refresh rate or you need to lock the game by some other means to exactly half your refresh rate. If you get this wrong, the game will look juddery and just isn’t fun to play. Using Special K’s frame-rate limiter or the in-game half-rate or third-rate v-sync option solves the issue. Interestingly, I found that Nvidia and RTSS’s frame-rate caps didn’t work well with Lossless Scaling – so I’d recommend Special K if you face similar issues.

Despite its limitations, Lossless Scaling is a great tool for PC users to take advantage of. The frame generation can be especially powerful when used with titles that have frame-rate caps, like Elden Ring, or in emulators. Even with a lack of visual quality in some areas and an increase to input latency, you may even be surprised by just how good it is. The app also offers other useful features, such as its titular lossless integer scaling, and given its £5 price point, it’s certainly worth experimenting with.




Source link

Related Articles

Back to top button