V-Rally 3 on GBA Makes DLSS 5’s ‘Yassify’ Graphics Look Soulless

V-Rally 3 on a 16MHz Toy Should Not Feel This Good

V-Rally 3 on Game Boy Advance should have been a joke. When a friend shoved the tiny purple cart into my hands back in the day and said “no, seriously, it’s proper 3D,” I braced for the usual handheld smoke and mirrors. Mode 7 fakery. Wobbly lines pretending to be polygons. The kind of technical flex that impresses Digital Foundry types for five minutes but feels awful to actually play.

Instead I got a rally game that, on a screen the size of a business card, sold me a sense of speed and physical space better than some “realistic” racers on full-blown consoles. The tracks rolled up and down real hills. Corners crept up out of the fog. Weather actually changed the mood. It felt like PS1-era Colin McRae had been shrunk, distilled, and wired directly into my thumbs.

Fast forward to 2026, and on the other side of my setup NVIDIA is trying to convince me that DLSS 5 is the future of how games should look. AI-driven photorealism. Neural rendering. Every frame “enhanced” by a model trained on who knows what. And somehow, that tech showcase running on obscene desktop hardware impresses me less than a 2002 cartridge running off a 16MHz ARM CPU and three AA batteries.

I am not anti-tech. I love a good graphics breakthrough. But the contrast between V-Rally 3’s craft on GBA and the weird, plasticky “Yassify” effect of DLSS 5 says something uncomfortable about where visual tech is heading. It is not just about pixels or polygons. It is about who is in charge of the image: the artists, or the algorithm.

What V-Rally 3 Actually Pulled Off on GBA

The Game Boy Advance was basically marketed as a portable SNES. Sprites. Tiles. 2D all the way down. Officially, 3D was not what the hardware was built for. Yet V-Rally 3 rolls up with a full 3D rally racer built around a custom engine called V3D from Fernando Velez and Guillaume Dubail – two devs who made a career out of making Nintendo handhelds do forbidden things.

Here is what you actually get on that little 240×160 screen, all running at what contemporary players reported as a surprisingly smooth, near-60fps clip:

• Texture-mapped 3D tracks with real elevation, not just flat “fake” roads
• 2D sprite cars and effects layered into the scene to save precious cycles
• Weather variants like rain, snow, fog and night that change visibility and vibe
• A cockpit-style first-person camera that makes the whole thing feel grounded and physical
• Licensed cars from actual manufacturers instead of generic boxes on wheels

Velez and Dubail pick their battles. The cars themselves are sprites, because trying to do full polygonal vehicles on that chip would melt it. The tracks carry the 3D illusion instead, with enough detail and texture to trick your brain into believing you are threading through a real course. Add in clean, readable UI and sound design that sells slides and engine strain, and suddenly the GBA – a machine known for pixel art platformers – is putting on a rally clinic.

There are no technical whitepapers breaking down exactly how V3D works, no exhaustive dev postmortems full of charts and cycle counts. What we can see and feel is the output: a game that takes a system designed for 2D and bends it into something else entirely, through ruthless optimisation and clear artistic intent. Every compromise is obvious, but every compromise is theirs.

The result still looks great today, not because it is remotely realistic, but because it is coherent. It has an identity. You look at a screenshot and instantly know what it is: a GBA rally game going for a PS1-ish vibe, tuned for speed and clarity on a tiny display. It knows exactly what it is trying to do.

DLSS Used to Be the Smart Kind of Magic

That clarity of purpose is exactly why DLSS used to excite me. When NVIDIA first rolled out DLSS, and especially when DLSS 2 arrived, it felt like a miracle with an honest bargain attached.

You render a game at a lower internal resolution, feed it into a trained AI model on the GPU, and it spits out an image that looks close to native resolution. The benefit is obvious: suddenly mid-range hardware can push higher frame rates, or turn on demanding options like ray tracing, without everything turning into a blurry mess. AMD’s FSR and Sony’s PSSR chased similar goals. Fewer raw pixels, smarter reconstruction. A neat, almost elegant way to cheat physics and budgets a little.

There were teething issues. Early DLSS could look smeary. Some games implemented it badly. But in principle it respected what developers put on screen. It did not vandalise the underlying art direction. It sharpened the image, lifted performance and then got out of the way.

That is why I have spent the last few years happily switching DLSS on in a ton of PC games. I get my smooth frame times; the artists keep their look. Everyone wins.

DLSS 5 and the Rise of the “Yassify Filter”

DLSS 5 feels like it comes from a different planet. From what NVIDIA has shown and what outlets and creators have dissected, this is not just smarter upscaling or even just frame generation. It is a step toward neural rendering that tries to reinterpret the scene in a more “cinematic” and photoreal way, sometimes effectively replacing large chunks of the frame with AI hallucinations trained to look like real-world footage.

That sounds impressive on a slide. In motion, the first wave of demos landed with a thud. Close-up comparisons from games like Resident Evil and Hogwarts Legacy made the rounds, and the pattern was hard to ignore. Faces were smoothed out and brightened like someone had stuck a beauty filter over the entire cast. Lighting shifted so that characters looked permanently ring-lit. Skin took on that uncanny plastic sheen social media loves. Even grungy, stylised worlds suddenly looked one step closer to generic prestige TV drama.

People called it the “Yassify” filter for a reason. Instead of respecting a game’s look, DLSS 5’s model seemed determined to drag everything toward a narrow idea of photorealism, informed more by shiny marketing footage than by the mood the original artists were going for. And because it operates in real time, every frame is competing with the original intent.

On top of that, the hardware needed to make these early tech demos run at decent frame rates sits in the “only streamers and benchmark fetishists own this” bracket. So the trade-off starts to look absurd. V-Rally 3 squeezes a whole rally sim onto a 2002 handheld and still hits about 60fps. DLSS 5 eats monstrous GPUs so that a horror game can look like its characters all spent an hour in a mobile photo-editing app.

Underneath the memes there is something more serious going on. DLSS 5 is not just another performance knob. It is an algorithm stepping into the director’s chair, imposing a house style that can flatten wildly different game aesthetics into something weirdly samey. And once that switch is in the options menu, the pressure to target and design around it can creep in whether the devs like it or not.

Why V-Rally’s “Fake” 3D Feels More Honest Than Neural Photorealism

On paper, comparing a GBA racing game to cutting-edge AI rendering sounds ridiculous. The point is not raw tech; it is philosophy.

V-Rally 3 starts from brutal constraints. A 240×160 display. Limited memory bandwidth. A CPU that would be laughed out of a modern smartwatch. Within those limits, Velez and Dubail ask what needs to be real and what can be faked. The track geometry is real enough to give your brain a 3D space to inhabit. The cars are sprites because no one actually cares how many polygons the hood has if the slide feels perfect. Weather and time-of-day are cheap but effective tricks layered on top to give each run flavour.

That is optimisation as a creative act. It is the same mindset that gave us gorgeous PS1-era racers that knew exactly how much detail they could push before sacrificing clarity, or N64 platformers that chose chunky shapes and flat shading over unstable, over-ambitious textures. These games are not chasing reality; they are chasing a particular sensation and using the tools they actually have to get there.

DLSS 5, in its current form, feels like the opposite. It starts from a modern game that already had a finished look. Artists have spent years tuning materials, lights and colour grading so their world has a specific identity: moody, comic-book, grimy, whatever. Then an AI model trained on a completely different data set leans over their shoulder and decides it knows better what looks “real”. It brightens eyes, polishes skin, boosts contrast, fiddles with reflections. It does not care why the shadows were soft in that shot or why that character had a slightly exaggerated nose. It optimises for a vibe that lives outside the game’s own language.

One approach admits its fakery and uses it with intent. The other pretends it is serving realism while quietly homogenising everything it touches. In a strange way, the low-res cockpit view in V-Rally 3 feels more honest than a DLSS 5 “enhanced” cutscene that no longer quite matches the raw render underneath.

Tech That Liberates vs Tech That Overrides

This is where my feelings about DLSS in general get tangled. On one hand, the original promise of DLSS — and similar tech from AMD and Sony — is still one of the most exciting ideas in modern graphics. We are long past the era where brute-forcing higher resolutions is cheap. Tricks that let mid-range PCs and consoles punch above their weight are essential. Used well, they let players experience more ambitious worlds without needing a new GPU every year.

On the other hand, there has always been a fear that these tools become an excuse. Why sweat over clever level-of-detail systems or engine-side optimisation if you can just ship something bloated and hope an upscaler or frame generator papers over the cracks. Players are already seeing games that basically assume you will enable some form of AI-assisted scaling to get acceptable performance.

V-Rally 3 is a reminder of what the other path looks like. No AI. No fallback hardware magic. Just developers staring down a weak handheld and figuring out how to make it sing anyway. They did not have the luxury of pushing costs onto the player via expensive components. If they wanted a 3D racer on GBA, the work had to be done on their side.

I do not want to romanticise old hardware ceilings too much. Limits also meant compromises that sucked: tiny draw distances, frame drops, cramped controls. And AI-assisted rendering has enormous potential when used as a humble, subordinate tool. Imagine DLSS-class tech that invisibly denoises ray tracing, or helps a small team scale a PC game down to Switch without rewriting half the renderer, or fills in peripheral vision detail in VR where no one is counting pixels anyway. Those are liberating uses.

DLSS 5’s current push toward aggressive, style-altering photorealism feels closer to overriding than liberating. It is tech that wants the final say, running after a marketing fantasy of “cinematic” games that forgets cinema has cinematographers, not neural networks, making the call on every shot.

The GBA’s 25th Birthday and an Awkward Mirror

The timing of all this is what really stings. While NVIDIA is busy defending why players are recoiling from DLSS 5’s aesthetic, the Game Boy Advance is quietly turning 25 and reminding everyone how much style can come from strict limitation.

V-Rally 3 was not alone in pushing that little handheld beyond its supposed role as a portable SNES. Doom and Duke Nukem 3D ports used their own flavour of pseudo-3D tricks. Other Velez & Dubail projects, like Stuntman and Driver 3, stretched the same ideas into different genres. It was an era defined by engineers and artists pulling in the same direction under harsh constraints. The hardware said “no” and the software replied “watch this.”

Modern AAA development lives in a different reality. Consoles are much closer to PCs. Engines abstract a lot of the metal away. Tooling is vastly more powerful. That is good in endless ways, especially for accessibility and for small teams. But it also means the industry’s default answer to visual problems is increasingly “throw more tech at it.” Reflections too noisy. Shadows too grainy. Textures too flat. There is always another algorithm waiting in the wings, trained on a bigger data set, ready to reshape the frame.

The GBA era shows another mindset, where the frame itself is sacred. Artists and engineers accept the screen and the silicon as boundaries, then design for them. V-Rally 3 does not waste a pixel. That is why its tiny resolution still evokes forests, mud, dusk and snow in a way that feels specific rather than generic, despite being built out of big chunky blocks.

Watching DLSS 5 smooth over faces and pump up highlights while V-Rally 3’s blocky trees whip past on a twenty-five-year-old handheld is like looking into a funhouse mirror version of progress. Raw capability has exploded; discipline and intent have not always kept pace.

Where I Draw the Line, For Now

I am not convinced DLSS 5 is doomed forever. Early versions of DLSS looked shaky too, and they improved dramatically once NVIDIA refined the models and gave developers more control. The idea of AI-assisted rendering is not going away, and in the right hands it could still become one of the quiet workhorses of game visuals rather than a showy gimmick.

Right now, though, I am drawing a personal line. If a game offers a DLSS 5-style mode that obviously changes faces, lighting and overall atmosphere, I am turning it off. I would rather take a small hit to performance or fidelity than let a black-box algorithm rewrite the look the team actually shipped. I am happy for tech to help fill in missing pixels; I am not happy for it to decide a moody, earth-toned scene needs to look like a fashion commercial.

On the flip side, I will still go out of my way to celebrate work like V-Rally 3. Not because it is perfect, and not because nostalgia should override modern comfort, but because it represents the side of graphics tech that still excites me the most: people squeezing soulful experiences out of uncooperative hardware through clever design, not glossy filters slapped on top.

The awkward part is accepting that both futures are probably coming along for the ride. AI will find quieter, more respectful roles in rendering, and it will also keep trying to impose itself as an artistic co-author. Some studios will chase the Yassify look because it demos well. Others will double down on strong art direction that resists homogenisation. Somewhere in the middle, players will keep flipping options on and off, hunting for the point where tech enhancements stop helping and start hijacking.

On one desk I have a hulking PC waiting for the next driver that promises smarter AI frames. On the other, a beat-up GBA still happily boots into V-Rally 3’s tiny, roaring rallies. The tension between those two machines is not going away any time soon, and honestly, living in that contradiction might be the most “modern gaming” feeling of all.