AMD FSR 5 Scarlet Cortex: First Look

AMD’s next-generation FSR 5 Scarlet Cortex technology represents a bold leap beyond traditional upscaling, moving into the realm of real-time AI-driven visual enhancement. This system analyzes fully rendered frames to intelligently improve lighting, materials, and atmospheric effects without altering the underlying geometry or art direction. Based on exclusive early access to a pre-release driver, the technology showcases AMD’s ambitious response to competitive offerings, pushing the boundaries of what’s possible with on-the-fly neural rendering.

The core innovation lies in its adaptive online learning capability. Unlike systems that rely on static, pre-trained models, Scarlet Cortex progressively refines its understanding of a game’s unique visual identity as you play. It learns the specific color palette, material properties, and lighting language in real time, requiring no game-specific profiles or downloads. This approach is designed to respect and enhance the developer’s artistic intent rather than imposing an external standard of photorealism.

A critical technical foundation is its deep driver-level analysis pipeline. Before the first frame is drawn, the system intercepts textures, geometry, and shader programs at load time to pre-classify materials, surfaces, and rendering intent. This deep contextual understanding allows the subsequent neural renderer to operate with greater precision. The model itself leverages INT8 quantized inference and training, executed on the dedicated AI accelerators within RDNA 4 GPUs. This dedicated hardware ensures the neural workload does not compete with the GPU’s shader resources for game rendering.

Integration is designed for broad accessibility through a two-tier model. Universal driver-level support will work across all compatible games, while an optional SDK allows developers to provide additional rendering data. This can lead to significantly better results or, conversely, let developers limit FSR 5’s influence to preserve a specific visual style or maintain fairness in competitive multiplayer environments.

Exclusive to the Radeon RX 9000-series with RDNA 4 architecture, the technology’s hardware requirements are non-negotiable. The system’s ability to perform both inference and continuous weight updates simultaneously at high frame rates demands the high-throughput INT8 matrix operations that only these new AI accelerators can deliver. There is no software-based fallback for older hardware.

At launch, FSR 5 will operate via a whitelist, only activating in games AMD has validated. Support will expand through regular driver updates, with a planned eventual transition to a blacklist model where it is enabled by default for most titles. Users will control the feature through the Radeon Software overlay, choosing from performance presets and monitoring the adaptive learning system’s status. A public driver update enabling FSR 5 is anticipated later in the second quarter of 2026, initially for DirectX 12 titles with Vulkan and DX11 support under exploration.

This new technology is designed to work alongside the existing FSR 4 toolkit, including ML upscaling and frame generation. It functions as a post-process stage that enhances image quality after the game renders a frame but before final output, making it a complementary layer in the visual pipeline.