At NVIDIA GTC 2021, the Unity game engine team announced native DLSS support, that will be available for all users since version 2021.2. DLSS is a smart image processing technique to get high resolution output from a low resolution source. The quality is slightly worse, that what native resolution can produce, however, the frame rate rises significantly.
How does DLSS work? Do you need it?
To achieve a stable upscale NVIDIA relies on machine learning: a special neural network is trained on frames with a real resolution of the 8K (image resolution standard). It over and over tries to pull the suitable quality out of more compact versions of the image until she finds the correct algorithm.
The best scenario (for now, with DLSS 2.x) is to upscale 1440p into 4K. And, of course, the technology is extremely useful in ray-traced games: in a special video, Unity showed their example of such a game, made in partnership with LEGO. The frame rate was raised from barely playable 18 to much more suitable 57 FPS.
The technology is useful only on NVIDIA 3000 and 2000 series of GPUs. Because it requires special Tensor Cores to process AI features without interrupting the main CUDA cores that are already busy enough, rendering your 3D game. This is why equal performance is impossible even on the newest (so far) Radeon 6000 series – RDNA 2 architecture doesn't have dedicated AI acceleration.
Earlier, we talked about how to enable the Resizable Bar:
- Log in to post comments
- 798 views