AMD Software: Adrenaline Edition is the driver and software platform for AMD Radeon GPUs. With the arrival of RDNA and the Radeon RX 9070 and RX 9070 XT combo, it's getting a massive AI overhaul. One of the brand-new opt-in features is AMD Image Inspector, a driver-based AI convolutional neural network (CNN) model that will run live on the Radeon RX 9070 and Radeon RX 9070 XT and watch you game.

"Shortly after launching your game," AMD explains. "AMD Image Inspector starts sampling on-screen content at a fixed interval." Translation, it will take screenshots. The reason? It's designed to look for artefacts and issues by assigning the screenshot a score to see if it requires further investigation.
You can review all captures before sending them off to AMD, as "nothing is sent until you give the go-ahead." This will then be passed on to AMD's team as part of the AMD User Experience Program to identify and rectify issues internally. Yes, it's an AI tool that runs and takes screenshots of the games you're playing to streamline driver and Adrenaline Edition software improvements.

AMD will also use the results to work with game developers and its various hardware partners to rectify any issues.
AMD notes that the AI model will only run, capture, and analyze screenshots when GPU utilization is low. The feature has been designed with privacy in mind and is entirely optional and opt-in. It will be available for Radeon RX 9000 GPUs via the Custom installation option for AMD Software: Adrenaline Edition, and you'll have to tick the AMD Image Inspector box to enable it.
"AMD Image Inspector's neural network is trained on more than 100 games, so its ability to pinpoint graphical glitches will vary depending on what you're playing," AMD writes. "As the CNN evolves and comprises more training data, its precision will improve. Regardless of whether a game is in the model's dataset or not, though, most of the images captured by AMD Image Inspector won't contain any corruption. With this in mind, we normalize the scores assigned by the model over time. This gives us an expected percentage of frames that are perfectly fine compared to those that need to be looked at more closely. If we suddenly see an abnormal spike in positive values, then we can more confidently investigate the cause."
AMD confirms that it's using similar tools in its labs and quality assurance, and is actively leveraging AI models to improve Radeon drivers and game support.