Deep Learning Super Sampling, DLSS, is an Nvidia method for improving the performance in games. By leveraging RT cores on certain GPUs, the driver of these cards can optimize the delivery of assets and post-processing effects to the user. The result is an asset pipeline that offers a lot smoother performance for higher framerates and resolutions. It’s a great idea, and AMD has been caught on the backfoot in response. Almost immediately, team red teased their own response, but hit some issues.
You can check out the new upscaling tech in action in the video down below. Keep reading after the jump to learn more about how this all works, and what it could mean for AMD.
AMD Reveals FSR 2.0
AMD’s current version of FSR, which is now being referred to as FSR 1.0, was released last summer by the company. The shading tech relies on using upscaling technology inserted into the pipeline to deliver more refined image quality for higher resolutions. By relying on a more simple spatial upscaler, the ability to scale a wide array of images is pretty easy to obtain. This should allow more games to leverage the feature. And the more games that add FSR support, the larget the dataset that AMD has to tune their algorithms against. That new data is the goal of FSR 2.0.
FSR 1.0 is limited though, compared to DLSS. The simple upscaler is easier to work with, but it has a major limitation. These kinds of techniques work frame by frame, treating each frame as a separate image and then adding more pixels. A simple spatial upscaler has to be rendered against each frame being drawn. This incorporates a lot of issues for AMD, as the overhead can be pretty bad on lower-end GPUs. AMD is developing and will be releasing a new generation of FSR that incorporates both spatial and temporal data, hoping to bypass the frame-based issues. But the best outcome is that using temporal data allows for higher quality images.
Another big difference here is that AMD is not using any kind of neural network to handle the optimization of upscaling. This is in direct contrast with Nvidia, which is using such technology to handle processing data reports from users to further refine future versions of DLSS. It will be interesting to see if the pace of development for improvements and adoption between FSR 2.0 and DLSS starts to diverge, and by how much.
Because FSR 2.0 doesn’t rely on RT cores, it can support both AMD and Nvidia GPUs. AMD themselves tout support for AMD GPUs going all the way back to the RX 400-series with the RX 480, as well as the current generation of 5000-series and 6000-series GPUs. On team green’s side, the FSR support could in theory support all the way back to the 1060 and 1600 cards. The biggest hurdle to the wide adoption of FSR 2.0 is that it’s a software-driven solution. This requires the developer to implement support for the feature in their games. There is another option though, Radeon Super Resolution (RSR).
What is RSR?
Radeon Super Resolution is the more widespread option between the two AMD options. Radeon Super Resolution (RSR) takes this issue on directly. Radeon Super Resolution is AMD’s answer to this problem of limited support, baking the idea behind super sampling optimization into the driver for pretty much any GPU, provided it has the required driver.
The rendering for RSR does get hit by a similar limitation of FSR 1.0. Whereas FSR can be inserted into the rendering pass before any post-processing noise and filters; RSR can only come into play at the very end, after all other effects have been drawn for a frame. This could have a more direct impact on performance if compared to DLSS or FSR 2.0. The GPU scaling feature this tech relies on is much less efficient, and likely won’t produce as sharp an image as FSR 2.0, but it should offer some imrprovement.
Who knows, maybe FSR 2.0 and DirectStorage APIs can be the key to a new generation of PC gaming power.