Gamers Choose "Fake" Pixels Over Real Ones — And the Data Is Overwhelming

"Are You Afraid to Like It?"

This is the part of the WAN Show episode that stuck with me. Luke asked Linus straight up: "Are you afraid to like it?" Linus had a 4080, runs DLSS-capable hardware, hadn't actually tried DLSS 4 yet on his gaming machine. His answer: "A little bit." Not because of what other people would think. Because of what it would mean if he liked it. He's been building PCs and caring about native resolution for decades. If AI-upscaled "fake pixels" are actually better — if he prefers them — what does that do to everything he thought mattered about PC gaming? [1] That's an honest reaction. And it's the real story underneath this study. This isn't about compression artifacts or VRAM budgets. It's about a cognitive shift that's happening whether enthusiasts are ready for it or not.

The sheer volume of voting and the overwhelming preference for DLSS is kind of mind-blowing right now.

— Linus Sebastian, WAN Show

Nvidia Saw This Coming 8 Years Ago

Linus walked through the timeline: Nvidia started developing DLSS technology — deep learning supersampling, which uses trained neural networks to reconstruct high-resolution images from lower-resolution renders — when they started working on the 20-series GPUs. That was roughly 2017, 2018. The first DLSS was rough. DLSS 2 was genuinely impressive. DLSS 3 added frame generation. DLSS 4 uses a new transformer model that's widely considered a step-change in quality [1]. Eight or nine years to reach this level of adoption. Meanwhile, AMD and Intel are now both fully committed to their own upscaling tech (FSR and XeSS respectively). The entire industry followed Nvidia's lead — and the study shows AMD still hasn't caught up. Native still beats FSR, 24% to 15% [1]. For a developer, this changes the optimization equation entirely. Why spend GPU budget on native 4K rendering if 48% of your players would actually prefer the AI-upscaled result? For a GPU marketer, "4K native performance" becomes a selling point for enthusiasts who want bragging rights — not a mainstream requirement. The battleground moved.

What Actually Changes

• GPU marketing: "native 4K performance" becomes a niche spec, not the main pitch
• Game development: optimization budgets shift away from raw rasterization fidelity
• AMD's competitive position: FSR finishing behind native is a real problem, not a rounding error
• Hardware pricing: if upscaling is preferred, mid-range GPUs running DLSS quality mode may be "enough" for most players
• The "4K monitor" market: still valuable, but the use case is AI-upscaled content, not native rendering

The Developer's Take

Here's the part that matters if you build software or think about graphics pipelines: DLSS is a machine learning inference pass. The model was trained on high-quality images and learned to reconstruct detail from compressed input. It's not sampling what's actually there — it's predicting what should be there based on motion vectors, temporal data, and a lot of GPU math [1]. That means the output can look better than native in some scenarios. Not in every scenario. Not always. But often enough that 48% of real gamers, watching real game footage, picked it over the real thing. I've been writing about tech long enough to remember when DLSS 1 was a blurry mess and enthusiasts rightly dunked on it. DLSS 4 is not that. The gap between what Nvidia shipped in 2019 and what they're shipping in 2026 is the whole story of why you have to take seriously the idea that machine-learned upscaling has won the visual quality argument — not just the performance argument.

The WAN Show episode that sparked this was a two-hour window into how even the most technically literate gaming enthusiasts are processing this shift in real time [1]. Linus is scared to try it because he knows what it might mean. That's probably the most human reaction to a GPU benchmark result I've heard in years. Credit to Linus Tech Tips and the WAN Show for breaking this down with actual context instead of just posting a chart. If you want to watch the full breakdown — including the DLSS vs. FSR segment and the Two Clicks Philip video discussion — the episode is linked in the references.