Advancements in Artificial Intelligence (AI) has led to many achievements. Over the years, NVIDIA has been working on implementing AI in graphics processing, and thus has developed the Neural Graphics Framework called NVIDIA NGX. NVIDIA NGX is part of the NVIDIA RTX Turing architecture, whose main purpose is to make use of AI to help enhance the graphics quality and thus improving the gaming experience. With AI, the possibilities are endless!
NVIDIA NGX integrates neural network into graphics processing implementation by creating a “ground truth“, which is then used to train the AI model. The trained model will then be tested and optimized using the ground truth image (which is usually a very high resolution image). This allows the AI model to be able to infer a higher quality graphics based on the current given graphics.
NVIDIA NGX is also opened up to developers as an API. By integrating the NGX API into their game engine, the game can also make use of the trained AI model, and allowing the game to produce a higher quality image. With the trained AI model, there are some additional features that are available as well.
AI Super Rez
AI Super Rez makes use of the trained AI model to upscale an image from its original resolution to a higher resolution, while still retaining clarity and detail. On the other side of the spectrum, traditional methods such as nearest neighbor and bicubic are unable to do so. Deep learning is able to produce a much sharper image at a higher resolution.
Traditionally, super slow-mo videos are taken using specialized equipment. However with deep learning, super slow-mo video can be achieved by inferring the frames needed to produce a smoother slow-mo video, without the need of any specialized equipment.
Similar to Adobe Photoshop’s content aware fill, the trained AI model allows objects to be removed from an area, and be replaced with a background. This is achieved by using the model to infer the background context to be replaced with.
Deep Learning Super Sampling (DLSS)
With neural network integrated, NVIDIA announces Deep Learning Super Sampling (DLSS) technology. DLSS enables higher rendered quality graphics with better performance in a game. To make use of DLSS, game developers are required to send a part of their games to NVIDIA. NVIDIA will then create a model based on what was sent. This model will act as the “ground truth” for the game.
The ground truth image is a super high-quality image with 64 jittered sample rendering. Based on the ground truth, the game will be able to produce a higher quality graphics comparable to TAA, while still maintaining high game performance. NVIDIA also state that DLSS can be implemented to any games that has TAA.
Here are some comparison of 64x Super Sampled and DLSS 2x as well as TAA and DLSS 2x.
As compared to TAA with 1080Ti, DLSS with 2080Ti has a much higher performance, while maintaining similar or better graphics quality.
As of the release of this article, there are more than 15 games supporting DLSS.
Conclusion & Pricing
The integration of neural network has enabled games to produce better graphics without sacrificing too much performance. This will surely encourage more game developer to hop on to the NGX API, allowing gamers to enjoy their game with higher quality graphics and game performance.
- NVIDIA GeForce RTX 2070 (Available in October 2018)
- MSRP – $499
- Founders Edition – $599
- NVIDIA GeForce RTX 2080
- MSRP – $699
- Founders Edition – $799
- NVIDIA GeForce RTX 2080 Ti
- MSRP – $999
- Founders Edition – $1,199