Markus “Notch” Persson, the renowned creator of Minecraft, recently ignited a social media discussion by publicly sharing his critical views on NVIDIA’s DLSS (Deep Learning Super Sampling) technology.
Notch firmly stated his belief that DLSS is inherently pointless:
“DLSS is essentially meaningless. You take a graphics card that’s too slow to run a game at an acceptable frame rate, then use THE SAME hardware to run a neural network that generates frames between existing ones.”
This provocative statement quickly drew responses from various developers and tech enthusiasts who attempted to challenge his perspective or simply criticized his apparent lack of understanding.
“Do you even understand how DLSS works? It uses Tensor Cores for DLSS, and it looks far from bad — latency is either absent or almost imperceptible. [Frame generation], yes, that’s a real headache, but not DLSS. DLSS is the future and the best technology ever. Neural rendering is the future.”
— hell_bl00dy
To this, Notch offered a concise dismissal: “Wow, that’s literally the worst opinion ever.”
“It’s crazy when a grown man publicly admits to being stumped by something so simple. But then again, this is the same grown man who wears Mickey Mouse ears.”
— mitch_kawasaki
Notch retorted, “I understand how computations work. You haven’t provided a single argument to refute my statement.”
“It utilizes a separate part of the graphics card specifically for frame generation (Tensor Cores), so it can all work synchronously, and even if it didn’t, creating a ‘fake’ frame is still faster than a real one. It’s unclear why Notch pretends not to understand this.”
— BOENSAW
Notch’s response was sharp: “Are you trying to say there are two types of computations? I’ll laugh at you now. You are an idiot.”
“You’re talking about frame generation, don’t call it ‘DLSS.’ DLSS is a suite of technologies, like Super Resolution, which is very commonly used. I don’t use Frame Generation myself – I think it’s not quite good enough yet for the games I play. However, Reflex 2 with Frame Warp might change that. Nevertheless, it’s perfectly reasonable to use neural rendering to fill the gaps when pure rasterization isn’t evolving fast enough.”
— TruthByRockwell
Notch has not yet responded to this specific argument.