NVIDIA
Nvidia’s latest GPU drivers introduce new AI-based upscaling techniques to make low-res videos streamed offline look great on high-res displays. Nvidia’s RTX Video Super Resolution (VSR), now available via GeForce driver 531.18 released on Tuesday, is the fastest of the 480p and 1080p videos I’ve seen in Chrome using a system with a 3080 Ti laptop GPU. I’ve successfully cleaned up some of the edges and blockiness, but a word of caution.
Nvidia measures that 90% of videos streamed from Netflix, YouTube, Hulu, Twitch, etc. are at 1080p or lower resolution. For many users, especially those with systems with his Nvidia GPUs, the browser will stretch this content when moving to 1440p and 4K screens. This can lead to image artifacts such as soft edges.
Not to be confused with AMD VSR (Virtual Super Resolution, aimed at low-resolution displays) (somehow), Nvidia VSR uses AI and RTX Tensor Cores from Nvidia’s 30- and 40-series desktop and mobile GPUs to increase sharpness and, according to Tuesday’s blog post by Nvidia’s Shield TV product manager Brian Choi, eliminate “blocky compression artifacts” when upscaling content to 4K resolution.
In addition to blockiness, Nvidia’s Choi said VSR targets “ringing artifacts around edges, washout of high-frequency detail, and banding in flat areas while reducing lost texture.”
This feature is currently built into Chromium, so it only works with the latest versions of Google Chrome and Microsoft Edge. Nvidia said it has not tested the feature in other browsers that use Chromium.
Additionally, support for other GPUs, including the 20-series, is not yet available, although Tom’s hardware tests describe it as a “relatively small” compute workload that older cards are likely to be able to handle. On its support page, Nvidia states that his VSR support for 20-series GPUs comes after refactoring “many” of the feature’s algorithms.
sharper image
I used the VSR to watch some lower resolutions on my 4K monitor (no doubt my system’s fans started spinning louder). A keen eye may be able to catch up more, but in extensive testing by Tom’s Hardware, Nvidia’s upscaling tech works “pretty well”, but the kind Nvidia uses for demonstrations. We concluded that it works best for videos featuring slow-moving content.
The Nvidia VSR has 4 levels, and when I used the maximum level with 1080p video, details seemed sharper. For example, in one YouTube video, the clouds hovering by the peaks were more defined and some areas of large bodies of water didn’t look blocky.
When viewing a Netflix 1080p show in the Chrome browser, I noticed a slight improvement in image clarity.
But don’t expect this feature to make your 480p content look like crisp 4K. This feature slightly reduced the blockiness in his YouTube video of a baseball player swinging his bat slowly, but it’s not a dramatic improvement to the naked eye.
Of course, your experience will vary depending on factors such as your system and display specs (which support a more powerful GPU than the one I used) and the video in question.
To dig deeper, Tom’s Hardware has an interesting gallery of lossless images. This publication points out a notable difference with Chrome’s upscaling. Sports videos, for example, “didn’t see much improvement, but there is certainly some sharpening and deblocking, at least in our subjective view.”
Performance also depends on how big the jump is. For example, upscaling from 1080p to 4K is less taxing than upscaling from a lower resolution. Upon upscaling his 480p video of the football match, Tom’s Hardware reported seeing “a large amount of detail” being lost, much like the soccer net stands.
For laptop users, even if you’re not using VSR, using VSR can reduce battery life. This is because his GPU automatically starts consuming power when the browser starts. Therefore, Nvidia recommends using a secondary browser to watch videos on VSR.
Naturally, VSR also consumes GPU resources (in addition to impacting battery life for laptop users). Nvidia acknowledges that “playing games or using GPU-intensive creative apps in parallel will result in slightly slower performance.”
For more examples, you can check the Nvidia demo below, but it was provided by Nvidia and contains very little detail, so think carefully.
Introducing RTX Video Super Resolution – 4K AI upscaling for Chrome and Edge videos.
