![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
- Joined
- Dec 19, 2005
- Messages
- 17,697
“Conclusion
FSR 5 Scarlet Cortex is the most ambitious graphics technology AMD has ever shipped, and in many ways, it's the most interesting response to NVIDIA's DLSS 5 that AMD could have designed. Rather than trying to out-train NVIDIA on cinematic datasets—a battle AMD would likely lose given NVIDIA's compute advantage—AMD has taken a fundamentally different architectural approach that plays to its own strengths.The adaptive learning system is the star of the show. The idea of a neural renderer that learns from the game itself, in real time, without requiring per-game profiles or developer integration, is genuinely novel. In practice, it works remarkably well. After just a few minutes of gameplay, the enhancements feel like they belong to the game rather than being imposed on it. The fact that it respects artistic intent by design—because it learns from the actual art direction rather than imposing an external aesthetic—is a meaningful philosophical advantage over DLSS 5's approach, which has drawn legitimate criticism for making everything look like the same "cinematic" style.
The driver-level analysis pipeline is equally impressive from an engineering standpoint. Intercepting textures, mesh data, and shader bytecode to reconstruct material and rendering semantics without developer help is clever work that leverages AMD's position as a driver vendor in a way that hasn't been done before. The shader introspection in particular—analyzing compiled DXIL/SPIR-V to identify PBR parameters, lighting models, and render pass types—is the kind of deep systems engineering that makes this feel like a real technology rather than a marketing exercise.
Performance overhead is reasonable. Quality mode's ~9% hit is noticeable but acceptable for the visual improvement you get, and Balanced mode at ~4–5% is a sweet spot for most players. The fact that the overhead is consistent across games, thanks to the dedicated AI cores handling the workload independently, is a design win that avoids the "it depends" caveats we see with so many GPU features.
There are limitations. The whitelist approach means you can only use Scarlet Cortex in AMD-validated games—9 titles at launch. That's understandable for a new technology, but it means most players won't be able to use it in their favorite game on day one. The RX 9000-series exclusivity, while technically justified by the INT8 AI core requirements, means the feature has a very small addressable market right now. The convergence period—while short in Quality mode—also means you're not getting the full benefit during the first few minutes of each new game.
The SDK integration path is smart forward-planning. Driver mode is good enough to stand on its own, but giving developers the option to feed explicit G-buffer data, light source metadata, and material IDs will meaningfully improve quality in games that support it. The FidelityFX framework integration lowers the adoption barrier, and the SDK also provides the hooks needed for competitive multiplayer whitelisting with enhancement masks—a thoughtful solution to the anti-cheat problem.
Looking at the bigger picture, Scarlet Cortex represents a genuine inflection point in how we think about real-time rendering. Both AMD and NVIDIA are now pursuing AI-enhanced visuals as the next frontier, but their philosophies diverge sharply. NVIDIA's DLSS 5 relies on massive pre-trained models with per-game tuning. AMD's FSR 5 takes the opposite path: adaptive, game-agnostic learning that works everywhere without manual intervention. History suggests there's room for both approaches, but AMD's approach offers a more intriguing scaling narrative. Every new game is supported automatically, and the system improves the more you play.
What makes this potentially disruptive for the GPU market is that FSR 5 changes the terms of the conversation. For years, GPU purchasing decisions have revolved around frame rates and efficiency—how many FPS can you get at what resolution and power budget. FSR 5 introduces a new axis: visual quality that goes beyond what the game engine itself can produce. If FSR 5 delivers on its promise at scale, an RX 9070 XT owner wouldn't just be getting competitive frame rates—they'd be seeing a fundamentally better-looking image than what any GPU without this technology can display, regardless of raw performance. That's not something you can achieve by throwing more shader cores at the problem or by generating extra frames. It could shift the value proposition in AMD's favor in a way that traditional benchmark comparisons simply don't capture.
The transition from whitelist to blacklist, when it comes, will be the real test. That's when FSR 5 stops being a feature you opt into for specific games and becomes part of the baseline Radeon experience. AMD tells us that's the plan, and the technology seems ready for it. The question is whether AMD can validate games fast enough to make the whitelist feel comprehensive before the transition happens.
Once the driver update lands later this quarter, if you own an RX 9070 XT, RX 9070 or RX 9060-series GPU, and play any of the supported titles, FSR 5 is absolutely worth exploring. Balanced mode is the recommended starting point. If this is what AMD can deliver as a first generation product, the future of adaptive neural rendering looks very promising. No doubt, NVIDIA will adapt to this new situation, too, and reallocate resources to be able to fight AMD's new tech.”
Source: https://www.techpowerup.com/review/amd-fsr-5-scarlet-cortex/
THe best leather jacket in the cyberpunk screenshots and many other easter eggs in there 
