My Thoughts on Notch's DLSS Post

Notch’s recent tweet that DLSS frame generation “fundamentally makes no sense” sparked a wave of debate across gaming forums. While many dismissed it as a subjective gripe, the comment forces a closer look at what DLSS actually does: it leverages deep learning super sampling to upscale both spatial resolution and temporal frame count using a neural network. The controversy isn’t just about personal preference; it raises an objective question about how much of a graphics card’s silicon budget should be devoted to pure rasterization versus AI‑driven upscaling.

Modern NVIDIA GPUs, especially the Ada Lovelace‑based RTX 4090 and the upcoming 5090, embed dedicated tensor cores that run the DLSS neural net. These cores occupy a distinct portion of the die, separate from the rasterization pipelines that render the original frames. Because the hardware is purpose‑built, the AI workload doesn’t simply steal cycles from game rendering—it runs in parallel on its own silicon. This means the perceived “free” AI processing is actually a deliberate design trade‑off, where a fraction of the chip’s die area is allocated to upscaling tasks, delivering higher frame rates and 4K quality without a proportional performance hit.

Understanding this hardware split matters for consumers deciding whether to invest in a high‑end GPU. If a buyer values raw raster performance above AI‑enhanced visuals, they might prioritize cards with more traditional shader units. Conversely, gamers who embrace DLSS can treat the tensor cores as an added value rather than an unwanted cost. Die‑shot analyses of the RTX 4090 illustrate the physical separation of these blocks, reinforcing that the AI accelerator isn’t a hidden surcharge but a purposeful feature of modern GPU architecture. Recognizing this nuance helps buyers make informed choices and demystifies the technical arguments behind Notch’s critique.