 |
| DLSS 5 On in Starfield. Todd Howard says this is how he always wanted the game to look. |
By Jon Scarr
NVIDIA announced DLSS 5 during Jensen Huang's keynote at GTC 2026 (GPU Technology Conference), and if you've been anywhere near gaming social media since, you already know it has caused an uproar. The reactions have ranged from genuinely blown away to outright furious, and honestly, both reactions make sense once you understand what DLSS 5 actually is and what it isn't.
So let me break it down.
Breaking Down the Technology
Let's start with what DLSS 5 actually is, because a lot of the reaction has been based on what people think it is rather than what it does.
At its core, DLSS 5 is a neural rendering model, which is NVIDIA's term for an AI system that steps in after the game has rendered a frame and reworks the lighting and materials in real time. The game still builds the scene the same way it always has. DLSS 5 then takes that rendered frame and uses colour information and motion data to understand what it's looking at. It works out whether something is skin, fabric, stone, water, or metal and applies the kind of light response each material would have in the real world. All of that happens within a single frame, every frame, without breaking the consistency of the image over time.
The results across the GTC demos were hard to ignore. In Starfield, ordinary environmental objects, countertops, paper towels, everyday clutter all stopped looking like assets and started looking like things that actually existed in a space. In Assassin's Creed Shadows, light moving through tree canopies and down through foliage picked up a translucency that's hard to achieve in real-time rendering. Hogwarts Legacy, Resident Evil Requiem, EA SPORTS FC, the Elder Scrolls IV: Oblivion Remastered, and the NVIDIA Zorah tech demo were all shown at GTC. The environmental stuff is where DLSS 5 is doing something genuinely new, and it's the part that's been drowned out by the face debate.
 |
| DLSS 5 off vs on in Assassin's Creed Shadows. The environmental difference is where the real story is. |
The Controversy Is Real but Needs Context
Here's where it gets complicated. A lot of the backlash has focused on faces, and that criticism isn't completely wrong. What DLSS 5 does instead of reading the original model is infer. And when you're inferring human faces in real time, you can end up with results that look different enough from the original character that people are asking whether this crosses a line.
I get it. I looked at some of the before and after comparisons and there are cases where a character genuinely doesn't look like themselves. That's a legitimate concern. If I'm playing a story-driven game and a character's face looks like someone else's, that pulls me right out of the experience.
And there's a deeper technical question worth sitting with. Because DLSS 5 is working from a flat 2D image rather than the actual 3D model, it doesn't have access to the original character's facial scans or geometry. It's making educated guesses based on a much broader dataset. That's not a conspiracy. It's just how the technology works right now. And for some people, knowing that an AI is filling in details that weren't originally there is going to be a hard thing to accept. Regardless of how good it looks.
But here's what the backlash is missing. The studios that own those games signed off on every single demo you saw. Bethesda, CAPCOM, EA, and Ubisoft all approved what was shown. DLSS 5 can't ship in a game without developer buy-in. And importantly, it's optional. Developers can choose not to implement it at all. Players can choose not to turn it on. This isn't something being forced on anyone.
Todd Howard said seeing DLSS 5 in Starfield was amazing and that it brought the game to life the way he always wanted it to look. CAPCOM's Jun Takeuchi said it helps his studio get closer to the cinematic experiences they've always been chasing. These aren't quotes from people who feel their vision has been hijacked.
What we're actually looking at is a work in progress. The current demo ran on two RTX 5090s and NVIDIA has confirmed it's still being optimised. The fall 2026 launch version is expected to run on a single RTX 50-series GPU, and minimum requirements haven't been confirmed yet. A lot is going to change between now and then.
My Take on DLSS 5
My honest take? The environmental and lighting improvements are the real story here, and they're being drowned out by the face debate. The faces are the most obvious thing to react to because that's where the human eye is most sensitive to change. But go look at a countertop scene in Starfield. Go look at how light moves through foliage in Assassin's Creed Shadows. That's where DLSS 5 is doing something genuinely new.
Whether the face rendering improves enough between now and fall 2026 to silence those concerns is a real question. That part I'm not ready to call yet. Whether it all comes together is still an open question. But I'm watching this one closely.
What do you think of DLSS 5? Are the environmental improvements enough to win you over, or does the face debate have you holding off? Let me know in the comments below.
Comments
Post a Comment
Comments are moderated. Please stay respectful and on topic.