AMD Ends FSR4 Branding, Unveils FSR Redstone With Machine Learning Graphics Features

AMD Ends FSR4 Branding as FSR Redstone Introduces Machine Learning Graphics Features

AMD has officially ended the FSR4 branding, folding it into a broader graphics platform called FSR Redstone. Rather than existing as a standalone technology, what was previously known as FSR4 is now labeled FSR Upscaling (ML) — clearly identifying it as AMD’s machine learning–based upscaler.

With Redstone, AMD is consolidating its FidelityFX Super Resolution technologies while introducing new ML-powered graphics features, including frame generation, ray regeneration, and radiance caching.

🔄 What Changed With FSR Redstone?

Under the new structure:

• FSR4 no longer exists as a brand
• ML-based features are grouped under FSR Redstone
• Older, non-ML technologies remain available but are renamed for clarity

This move aligns AMD’s branding more closely with how modern GPU features are evolving — particularly in response to AI-driven rendering techniques.

🧠 FSR Upscaling: ML vs Analytical Versions

FSR Upscaling (Machine Learning)

• Previously known as FSR4
• Uses neural networks for improved image reconstruction
• Exclusive to RDNA4 GPUs
• Same technology, new name — no functional change

FSR Upscaling (Analytical)

Includes:

• FSR1
• FSR2
• FSR3
• FSR3.1

These versions:

• Do not use machine learning
• Continue to support RDNA1, RDNA2, RDNA3, and RDNA3.5
• Remain unchanged in image quality and performance

The renaming is purely to distinguish ML-based vs non-ML upscaling.

🎞️ Frame Generation Under FSR Redstone

AMD has now split frame generation into two distinct paths:

FSR Frame Generation (Analytical)

• Works on supported RDNA GPUs
• Same behavior as before
• Broad game compatibility

FSR Frame Generation (ML – Redstone)

• Uses machine learning
• Exclusive to RDNA4
• Requires games with FSR 3.1.4 or newer
• Enabled via driver-level override, not AFMF

Because of these requirements, game support is currently limited. AMD expects over 30 games to support ML-based frame generation by the end of the year.

🔦 Ray Regeneration & Radiance Caching Status

FSR Ray Regeneration

• ML-based denoiser for ray/path tracing
• Reconstructs clean images using fewer rays
• Currently available only in:
  • Call of Duty: Black Ops 7
  • Multiplayer & Zombies modes
• Produces sharper reflections than analytical denoisers
• Still behind Nvidia’s ray reconstruction in complex scenes

FSR Radiance Caching

• Neural network–based real-time radiance cache
• Aims to reduce ray tracing cost significantly
• Not available in any released games yet
• Accessible only via the FSR Redstone SDK
• First supported titles expected in 2026

🎮 Game Support & Driver Overrides

Upscaling remains more widely adopted because it boosts performance without adding latency, unlike frame generation.

🖼️ Image Quality: ML vs Analytical Frame Generation

Testing shows:

• Similar results in slow-motion scenes
• Noticeably better stability in fast-moving scenes with ML
• Reduced:
  • Ghosting
  • Trailing
  • Edge instability

In stress tests jumping from 30 FPS to 60 FPS, ML frame generation delivers cleaner results with fewer artifacts.

⚙️ Performance Comparison

Performance impact is minimal:

Image quality improves without meaningful performance loss, making ML frame generation a clear refinement rather than a raw speed boost.

❌ Hardware Exclusivity Drawback

One of the biggest concerns is strict RDNA4 exclusivity:

• No support for RDNA1, RDNA2, RDNA3, or RDNA3.5
• Includes desktop GPUs and APUs
• Earlier leaks suggested broader compatibility, but AMD has made no commitments

This decision reflects RDNA4’s stronger machine learning acceleration but limits adoption.

🧾 Final Thoughts

FSR Redstone marks AMD’s serious push toward machine learning–driven graphics.
Key takeaways:

• Cleaner branding
• Improved image stability
• Competitive ML features
• Limited by hardware and game support

While Redstone brings AMD closer to feature parity with competing solutions, its long-term impact will depend on broader adoption across games and GPUs.