AMD Responds To Accidental FSR 4 Source Code Leak That Revealed A Surprising Radeon Detail
Many people praised the move initially, before just hours later AMD attempted to pull some of the code from Github. You see, AMD only intended to post the source required for implementing FSR 4, not the code for the upscaler itself. The former allows developers to easily integrate FSR 4 into their games. The latter lets anyone implement and modify FSR 4 for their own purposes, especially considering the permissive MIT license that the code was posted under.
Due to the way Github works, AMD hasn't really been successful in removing the wrongly-released source from the website. Besides that most of the changes can still be seen in the commit history, there are also many users who duplicated and forked the repository, creating duplicates. The cat's out of the bag, so to speak, and it's not clear what happens from here.
AMD stated that the code was published in error, but given that it was published publicly under an explicitly-irrevocable MIT license, it's highly arguable that folks who grabbed it before it was taken down may still have a right to use, copy, modify, merge, publish, distribute, sub-license, and even sell their own versions of FSR 4.
AMD confirmed to Chips & Cheese that the release was accidental.
Even if AMD takes someone to court over it, and the court decides that the lack of intention behind AMD's gaffe means that it's actually still proprietary, the fact remains that the code is out there. Anyone who wants to can take it and use it. After all, laws and legal proceedings cannot acutally make people stop anything; all they can do is assign penalties to the doing. Good luck enforcing such a judgment in China, Russia, or even Korea.
One interesting detail about the leaked source is that it included FSR 4 AI models in INT8 format. Normally, FSR 4 requires the FP8 format for which support was only added to the Radeon lineup with the RDNA 4 architecture, so far found only in the Radeon RX 9000 series GPUs. The existence of INT8 versions of the FSR 4 models does not necessarily prove that AMD plans to launch FSR 4 for older-generation RDNA 3 GPUs, but it does mean that the company was at least experimenting with same.
We sort-of already knew that; AMD did say that it was looking into bringing FSR 4 to RDNA 3. An INT8 version of the model would certainly offer better performance than the FP16-based hack some enthusiasts used to get FSR 4 working on Radeon RX 7900 series GPUs. While that trick was functional, it offered very little benefit given the intense cost of running the FSR4 models on the Navi 31 GPU.
The road ahead is interesting. Despite a promise from Intel (that has gone wholly unfulfilled), AMD is the only company to date that has open-sourced a modern upscaler. The company never promised to open-source FSR 4 the way it did previous version of FSR, but now it has done so, inadvertent as that release may have been. Anyone who grabbed the files while they were up is legally licensed to use them. We'll see what AMD's response is the first time someone does so.
