AMD FSR 4 Stays Locked to RDNA 4 as INT8 Leak Teases Older GPU Support

AMD recently reported strong financial results and unveiled Kintex UltraScale+ Gen 2 FPGAs, reinforcing momentum across enterprise and embedded markets. New details now clarify where AMD draws the line for gamers.

FSR 4 support status and hardware limits

According to TechPowerUp, FidelityFX Super Resolution 4, now branded simply as FSR 4, supports many games but does not officially cover all RDNA GPU generations. AMD told Hardware Unboxed that it has no update on official FSR 4 support for the Radeon RX 7000 series or older GPUs.

FSR 4 depends on RDNA 4’s 8-bit floating point (FP8) instructions. RDNA 4 GPUs include hardware support for Wave Matrix Multiply Accumulate operations in FP8 format. RDNA 3 and RDNA 2 lack this capability. Older Radeon GPUs fully support INT8 data formats, but they cannot process FP8 in the required way.

INT8 leak, performance results, and segmentation

AMD accidentally leaked an INT8-based FSR 4 library on GPUOpen, which hints at potential compatibility with older hardware. Testing this leaked build produced mixed results. In Cyberpunk 2077 at 4K Ultra, a Radeon RX 7900 XTX delivered 11 percent better performance than native rendering, but ran 16 percent slower than FSR 3.1.

These results place INT8 FSR 4 between native image quality and FSR 3.1 performance. Performance trade-offs likely explain AMD’s reluctance to release it officially. Product segmentation also plays a role. AMD already restricts FSR “Redstone” features, including Ray Regeneration and Radiance Caching, to RDNA 4 GPUs in the Radeon RX 9000 series. RDNA 3, RDNA 2, and RDNA 1 cards continue to rely on FSR 3.1 for upscaling and frame generation.

Despite the lack of official support, the leaked INT8 library suggests FSR 4 could still reach older GPUs later, with lower performance than RDNA 4 implementations. Not all news favors AMD, however. The company also faces server CPU supply constraints in China, adding pressure on the enterprise side.