I want to discuss something that's really been bugging me lately... Why the hell has AMD not released FSR 4 yet for owners of older Radeon GPUs? The lack of updates and the lack of support for these older architectures is baffling. Releasing FSR 4 upscaling improvements beyond RDNA 4 should be the easiest marketing win imaginable, yet AMD isn't taking the win. What is going on here? Is the Radeon division slipping back into its old habits?
Let's get into it.
As many people know, FSR 4 upscaling is only officially supported on AMD's latest Radeon RX 9000 series GPUs, which use the RDNA 4 architecture. These GPUs launched about a year ago, and FSR 4 debuted alongside the hardware as a key software addition. Across the testing we've done so far, we've been very impressed with FSR 4 as it's an enormous step forward in image quality compared to FSR 3.
AMD Said Older GPUs Couldn't Run FSR 4 – Until the Code Leaked
When AMD released FSR 4, they offered a straightforward explanation for why the technology was limited to the newest GPUs. FSR 4 was built to use FP8 instructions, and RDNA 4 is currently the only Radeon architecture with native FP8 hardware support.
As a result, enabling FSR 4 on older architectures without FP8 support was said to be impossible. This is the same reasoning Nvidia has used for DLSS over the years: DLSS requires Tensor cores, so it doesn't work on architectures that lack them.
Around the time gamers started getting their hands on RDNA 4 GPUs, rumors began to surface that FSR 4 might be backported to older architectures. Some of AMD's official communications at launch even left the door open to supporting FSR 4, or a version of it, on previous-gen GPUs. Still, these were only rumors. With a legitimate hardware explanation for why FSR 4 wouldn't work on older cards, it wasn't clear whether AMD could backport it in a way that was functional nor effective.
…or so we thought.
That changed when AMD accidentally published source code for a version of FSR 4 that used INT8 instructions instead of FP8. This happened on August 20, 2025, as part of the FidelityFX SDK 2.0 release. The code was quickly pulled because AMD had no intention of open-sourcing FSR 4, but by the time that happened, Radeon owners had already downloaded it. Once something hits the internet, it doesn't really disappear.
Once something hits the internet, it doesn't really disappear.
This was a big deal. Hardware-accelerated INT8 support was first introduced with RDNA 2 and is also present in RDNA 3 and RDNA 4. That meant an INT8-based version of FSR 4 should theoretically work on older Radeon architectures, at least going back to the popular RX 6000 series.
Less than a month later, Reddit user AthleteDependent926 compiled an FSR 4 INT8 DLL from the leaked source code and shared it for others to test. Since FSR 3.1 introduced support for upgrading FSR via DLL swaps, enabling FSR 4 on RDNA 3 GPUs became as simple as dropping the INT8 DLL into a game's directory. With some additional driver tweaks, it was also found to work on RDNA 2 GPUs, though the latest drivers at the time caused compatibility issues.
Testing Confirms FSR 4 INT8 Works on RDNA 2 and RDNA 3
This is when we decided to test FSR 4 INT8 for ourselves. In early October 2025, we confirmed that FSR 4 INT8 not only works on RDNA 3 and RDNA 2 GPUs, but works well, see for yourself:
AMD Tried to Hide This: FSR 4 Upscaling Already Works on Older Radeon GPUs – Tested: FSR 4 INT8 on RDNA 3 and RDNA 2
Image quality is significantly improved, often looking very similar to the full FP8 implementation in many scenarios. Performance overhead is higher than FP8 on RDNA 4, but the overall experience is still a major upgrade over FSR 3.1.
If FSR 4 INT8 was officially available, RDNA 2 and RDNA 3 owners would absolutely choose it over FSR 3.1.
In the conclusion of our FSR 4 INT8 analysis, we wrote that based on what we had seen, AMD should release the INT8 version of FSR 4. For RDNA 2 and RDNA 3 owners, it would be the preferred upscaling option in modern titles, extend the lifespan of their hardware, and reward loyal Radeon buyers.
That was four months ago. By now, you'd expect AMD to have released this feature officially. The code has existed since August, nearly six months ago, and has been proven to work well in real-world games for over four months. AMD even released a major FSR update in December 2025 called Redstone, which added features like ML-powered frame generation and ray regeneration. FSR 4 INT8 was not included, and as of February 2026, it's still missing.
AMD's Silence, Redstone, and a Growing Support Problem
With Redstone, AMD went out of their way to publish updated compatibility tables showing which features work on which architectures. FSR 3.1, now branded as the "analytical" upscaler, remains the only option for RDNA 2 and RDNA 3, while FSR 4 is still locked to RDNA 4. This is despite months of clear evidence that FSR 4 INT8 works perfectly well on older GPUs. It's completely ridiculous to justify this omission.
We reached out to AMD for comment and received a single-line response: there are "no updates to share at this time." That's where things stand – no official explanation for why a known, working feature remains unsupported.
What makes this even more frustrating is that instead of releasing FSR 4 for older GPUs, AMD is spending time adding an optional AI bundle to a recent Radeon gaming driver. A gaming driver, for gaming GPUs, marketed to gamers. Talk about not reading the room whatsoever. It's a baffling decision, especially when there's a highly requested, functional feature sitting on the sidelines.
And it's not like we're talking about a half-baked experiment on beta here. Based on our testing and user reports, FSR 4 INT8 appears stable, visually impressive, and appropriately performant given the complexity of the algorithm. It shows no obvious graphical issues and seems ready for public release with minimal additional effort. Adding it to the driver stack should be far easier than developing and integrating unrelated AI features.
Why Holding Back FSR 4 Is the Wrong Move Right Now
The huge mistake AMD is making here is that over time it's becoming increasingly clear that Nvidia offers better long-term driver and feature support for GeForce GPUs than AMD does for Radeon. At CES 2026, Nvidia announced and released DLSS 4.5, which introduces a new second-generation transformer model. That update is available on GPUs going back to the RTX 20 series from 2018, over seven years of continued feature expansion.
Yes, DLSS 4.5 runs best on newer hardware, and FP8 acceleration is limited to the RTX 40 series and newer. But Nvidia still gives older GPU owners the choice. Users can decide which DLSS version works best for them. Having the option matters, even if it isn't ideal in every case. This shows that there's nothing inherently wrong with releasing features that may not perform optimally on older GPUs. If the market leader is willing to do it, AMD can, too. Yet AMD seems stuck in a follower mindset, selectively copying Nvidia while ignoring one of the most important lessons: long-term support builds trust.
This FSR 4 situation also mirrors AMD's recent attempt to move RDNA 1 and RDNA 2 drivers into maintenance mode. That decision sparked heavy backlash and was eventually reversed, but it reinforced the perception that AMD is quicker to move on from older products than Nvidia is.
The reality is simple. Releasing FSR 4 INT8 for RDNA 2 and RDNA 3 is an easy win. It's good PR, good for community sentiment, and good for customer loyalty. It would reward buyers who supported Radeon during weaker market periods and help counter the narrative that Nvidia simply does software support better – because right now, that narrative is accurate.
These decisions affect purchasing behavior. One of the reasons Nvidia dominates the GPU market is that they are seen as the company at the forefront performance, software features and software support. Gamers have shown they are willing to pay more for that assurance. AMD falling further behind in this area is the last thing the company needs.
Because let's be honest, AMD's overall GPU sales are not good enough to justify stupid moves that hurt the community. Lots of Radeon owners are aware that they could have access to FSR 4, except AMD isn't willing to support them officially. That's bad for a company that is trying to recover from several poor quarters of gaming division revenue. Poor sales means needing to try harder to win back customers, not leaving customers behind.
In a normal GPU market offering FSR 4 INT8 officially would be an easy win, but the win would be even more substantial in the current market. GPU prices are rising, availability is tightening, and accessing FSR 4 through new hardware is becoming increasingly expensive. The Radeon RX 9060 XT 16GB has climbed from its $350 MSRP to around $450, and the 9070 XT is sitting at roughly $730, about $130 over MSRP. In this environment, the best move AMD could make is to improve the experience for existing Radeon owners.
The only plausible explanation for withholding FSR 4 INT8 is that AMD wants to push RDNA 2 and RDNA 3 owners toward upgrading to RDNA 4. If that's the strategy, it's a poor one. Most buyers who wanted to upgrade already have, and rising prices make further upgrades less attractive. Frustrating existing customers risks driving them away entirely.
Over the last few years, we've been highly critical of the Radeon division's poor decisions and repeated blunders – we've covered them before. We thought AMD was finally on the right track when it chose not to sabotage the RDNA 4 launch and largely got pricing right, but more recent moves have made us question that progress. The RDNA 1 and 2 driver support issue was a serious blunder, and now this situation with FSR 4 support.
Our position is straightforward. New features should be supported on as many GPUs as possible unless there is a genuine technical barrier. In this case, there isn't one: we know FSR 4 INT8 works. Sitting on it is not acceptable.
The silver lining is that FSR 4 INT8 is relatively easy to use unofficially. Thanks to AMD's own source code slip-up, gamers already have access to a functional version via simple DLL swaps on RDNA 3, with additional tweaks enabling RDNA 2 support. But that does not excuse the lack of official backing. AMD should be doing better, and it knows it.