Shocking math: What exactly is 1%? When you run a game at 8K resolution (7680×4320), your hardware has to process over 33 million pixels in each frame. In the Pragmat game test, a radical method was used: frame rendering was adjusted using the DLSS Tweaker program and only at 0.10 scaling. Although some interpret it as 10%, the math is clear because when the area is calculated (0.10 x 0.10), it is actually only 1% of the original (native) number of pixels in the frame.
This practically means that we perform an incredible x100 image scaling. The graphics card natively processes only 768×432 pixels, and the AI has to reconstruct the remaining 99% of the image to fill the 8K format. This is where the new DLSS 4.5 comes into play, which together with “Preset L” proved to be outstanding. This setting profile was developed by Nvidia specifically for extreme conditions and is recommended for 4K resolution in Ultra Performance mode, but here we pushed it to the limit.
What exactly is DLSS and how does it see what is not there? If we did this the old way, without modern technologies, you would only see an unrecognizable, blurry blob of color on the screen. DLSS is not ordinary upscaling that just stretches the image. It is an AI algorithm that uses neural networks trained on NVIDIA supercomputers. He does the following:
- It takes a very low resolution image.
- It uses data from previous frames and motion vectors to predict where the pixels should be.
- The AI literally draws in the missing details, filling in the gap created by this extreme x100 magnification.
Power paradox: Why is the GPU at 99% usage?
Interestingly, even though the native resolution is low, the graphics card (RTX 4070) still works at 99% capacity. This is because the processor of the card does not rest, but redirects all the power to the Tensor cores that perform complex AI reconstruction. The result is a stable 40-50 FPS, which makes the game fully playable in 8K, despite the card working at the limit of what is theoretically possible.
AI Rendering: The Vision Jensen Huang Talks About
This is a living proof of the vision of the director of Nvidia who claims that in the future we will not render every pixel by brute force of the hardware. We’ll render a small part and the AI will finish the rest.
With the presented test, the author wanted to point out the fact that the border between the real image and the AI reconstruction becomes practically invisible. The future of gaming is no longer in the number of transistors, but in the intelligence of the algorithm.
The test raises many questions, and one of the most important is whether advanced AI algorithms will make powerful and expensive hardware unnecessary. Based on what we saw in the example of the RTX 4070 and completely fluid animations in 8K resolution, you wonder if weaker GeForce models like the RTX 5050 can do something similar only in 4K resolution. In order not to theorize too much, the concrete answer is here in the form of a real demonstration of what a budget GeForce RTX 5050 with 8 GB can deliver together with DLSS 4.5L and MFG technology.
Are you surprised? What is your opinion, after everything shown about the future of gaming and the need to buy expensive hardware? Write and comment…
Author: Nenad Durmiši