AMD continues to restrict its latest upscaling technology to RDNA 4 graphics cards, despite evidence that an INT8 version could extend support to older Radeon GPUs. The company maintains there are "no updates to share at this time" regarding official support for RDNA 3 and earlier architectures.
The technology debuted alongside the Radeon RX 9000 series about a year ago, marking AMD's first AI-powered upscaling solution. It relies on FP8 instructions that require specific hardware capabilities only available in RDNA 4 GPUs. This hardware dependency initially justified the exclusivity, similar to Nvidia's Tensor core requirements for DLSS.
That justification weakened in August 2025 when AMD accidentally published source code for an INT8-based version. The leak occurred as part of the FidelityFX SDK 2.0 release and was quickly removed, but not before users downloaded the files. Hardware-accelerated INT8 support exists in RDNA 2, RDNA 3, and RDNA 4 architectures, theoretically enabling backward compatibility.
Testing confirmed the INT8 version functions on older hardware. In Cyberpunk 2077 at 4K Ultra settings, a Radeon RX 7900 XTX delivered 11 percent better performance than native rendering using the leaked INT8 build. However, it ran 16 percent slower than FSR 3.1, placing it between native image quality and previous-generation performance.
"a very difficult technical challenge for us to solve"
AMD executives have publicly addressed the compatibility question. David McAfee, Vice President and General Manager of Ryzen CPUs and Radeon Graphics, called expanding support "a very difficult technical challenge for us to solve" during CES 2026.
The situation frustrates RDNA 3 owners who purchased relatively recent hardware, including users of AMD-powered gaming handhelds running RDNA 3.5 graphics. The leaked build remains accessible through unofficial channels, where users can implement it on RDNA 3 GPUs via simple DLL swaps, with additional tweaks enabling RDNA 2 support. This workaround demonstrates the technical feasibility while highlighting AMD's reluctance to provide official backing.
Nvidia's contrasting strategy adds pressure. At CES 2026, Nvidia announced DLSS 4.5 with support extending back to RTX 20 series GPUs from 2018. While newer hardware delivers optimal performance, older GPU owners receive the option to upgrade. AMD's current stance risks reinforcing perceptions that Nvidia offers superior long-term software support.
Market conditions compound the issue. GPU prices have increased since the RDNA 4 launch, with the Radeon RX 9060 XT 16GB climbing from $350 MSRP to around $450. The 9070 XT sits at roughly $730, approximately $130 over MSRP. In this environment, improving existing hardware value through software updates could strengthen customer loyalty.
AMD's December 2025 Redstone update added features like ML-powered frame generation and ray regeneration while maintaining the RDNA 4 exclusivity. Compatibility tables published with Redstone show FSR 3.1 remains the only option for RDNA 2 and RDNA 3, despite six months of evidence that the INT8 version works effectively on those architectures.
AMD reported strong financial results earlier this month alongside Kintex UltraScale+ Gen 2 FPGA announcements, reinforcing enterprise momentum. The company's gaming division faces different challenges, with FSR 4 compatibility representing both a technical decision and a customer relations test as GPU market competition intensifies.
