At Computex 2025, AMD revealed the next iteration of its FSR tech, codenamed FSR Redstone. AMD FSR Redstone will be based on the foundations of FSR 4.0, and it will come out in 2H 2025. So, let’s see what it will bring to the table.
AMD FSR Redstone will bring support for Neural Radiance Caching, Machine Learning Ray Regeneration, and Machine Learning Frame Generation.
Neural Radiance Cache (NRC) is a smart shader that uses neural networks to make indirect lighting look better and run faster. With NRC, fully ray-traced lighting will look more realistic and will be more responsive. This should make everything look smoother and more natural.
On the other hand, Machine Learning Ray Regeneration will be similar to what DLSS 4 already has. By using it, FSR Redstone will be able to provide a crisper ray-traced image. Plus, there won’t be big delays between different lighting changes.
Finally, Machine Learning Frame Generation will aim to improve the quality of AMD FSR Frame Gen. To be honest, I was really happy with AMD FSR 3 Frame Gen. AMD FSR 3 Frame Gen performed better than DLSS 3 Frame Gen in a lot of games. It also did not suffer from any major visual artifacts. So, this is the least interesting feature of FSR Redstone. At least in my opinion.
Now I know some will be quick to say that AMD FSR Redstone will be on par with DLSS 4. However, that’s not the case. DLSS 4 still has some features that have not appeared in AMD FSR.
For instance, I have not heard anything about FSR Redstone supporting Spatial Hash Radiance Cache (SHaRC) or Shader Execution Reordering. There is also no support for Multi-Frame Gen. And believe me. AMD WILL follow NVIDIA’s example when it comes to Frame Gen. They WILL add MFG to FSR. It’s just a matter of time.
The reason I’m raising this is because there has been a controversy surrounding NVIDIA DLSS 4 MFG. For those unaware, NVIDIA demanded GamersNexus to use MFG in its coverage, otherwise they’d restrict access (or even blacklist them). And you know what? That was one of the dumbest moves I’ve seen. Seriously, what was NVIDIA thinking? Needless to say that we’re siding with GamersNexus on this. No company should determine how the media covers its products.
With this out of the way, MFG can be useful. Well, that is if you can hit at least 40-50FPS in a game without Frame Gen. I’ve extensively covered MFG in almost all the games that support it. MFG currently has some issues that need to be addressed. While the input latency is fine, there are a lot of visual artifacts and ghosting issues when using MFG X3 or X4. In some games, these artifacts are easily noticeable. This is what NVIDIA has to fix as soon as possible, otherwise, MFG X3 and X4 are a no-go for me.
Before closing, I should note that AMD FSR Redstone will be exclusive to the RDNA4 GPUs. After all, it’s based on FSR 4. So, don’t expect these new features on the RDNA2 or RDNA3 GPUs. Again, this shows how limited the free version of FSR was. It also proves that NVIDIA was right to invest, from the get-go, in Machine Learning for DLSS.
Stay tuned for more!
John is the founder and Editor in Chief at DSOGaming. He is a PC gaming fan and highly supports the modding and indie communities. Before creating DSOGaming, John worked on numerous gaming websites. While he is a die-hard PC gamer, his gaming roots can be found on consoles. John loved – and still does – the 16-bit consoles, and considers SNES to be one of the best consoles. Still, the PC platform won him over consoles. That was mainly due to 3DFX and its iconic dedicated 3D accelerator graphics card, Voodoo 2. John has also written a higher degree thesis on the “The Evolution of PC graphics cards.”
Contact: Email
Typo..
But the feature is dubbed as "Ray Regeneration" in AMD's case. Confusing name to say the least though.
As per AMD:
Ray Regeneration is a trained neural network is used to regenerate pixels that can not be accurately path-traced.
Ray Regeneration uses machine learning to predict and filter grainy noise in real time.
https://uploads.disquscdn.com/images/f0a5821c198587ea08981532c9effe26bdad94ee3eac7bd8d4095ca6a845ecb1.jpg
….
Nvidia is going to drop a motherload of RTX features with the 6000 series. I just hope that Microsoft has the DirectX Work Graphs API GPU autonomy feature ready in time.
Motherload of features like what?
I appreciate you bringing up the Nvidia Media Coverage controversy. It's absolutely mind boggling to me they would go to these lengths and obfuscations especially after seeing that the card seems to even be a substantial upgrade from even the 4060 without considering MFG.
I am also of the mind that while native performance is optimal. If the tech can produce an experience close enough to native that I would struggle to tell the difference from I would have no issue with using it. MFG is probably the only one now that I'm very interested to see come to the RDNA cards due to all these new 240Hz panels that the tech seems suited for.
I am excited for the Redstone Developments but kinda sad that it seems that FSR 4 and similar tech being back ported to RDNA3 seems dead. Maybe the MFG will be something that could be allowed to be backported to FSR 3.X. I guess ROCm to Windows will be something RDNA 3 will be able to look out for at the very least.
The main reason why FSR4 won't be backported as-is to RDNA3 is because it makes use of FP8, which is only supported by the RDNA4 architecture on the AMD side.
In fact, Valve's lead developer of Proton's DX12-to-Vulkan translation layer has managed to get FSR4 working on RDNA3 by making use of FP16, but as expected, the performance is abysmal.
If anyone is interested in the gory technical details, look up his article on the matter (and trust me when I tell you this guy knows what he's talking about):
https://uploads.disquscdn.com/images/7214b76c8feb1754f1550a01a80846bcd4b097caba491dd9c475088162d3b67f.png
Typo, or oversight ? ..
But the feature is dubbed as "Ray Regeneration" in AMD's case. Confusing name to say the least though.
Anyway, as per AMD:
Ray Regeneration is a trained neural network is used to regenerate pixels that can not be accurately path-traced.
Ray Regeneration uses machine learning to predict and filter grainy noise in real time.
https://uploads.disquscdn.com/images/f0a5821c198587ea08981532c9effe26bdad94ee3eac7bd8d4095ca6a845ecb1.jpg
….
I think he's technically correct? Wasn't Ray Reconstruction something introduced in DLSS 3.5 meaning it's also included in DLSS 4?
Regeneration and reconstruction is the same process. Nvidia may have a copyright on the term Ray Reconstruction forcing AMD to use a slightly different name. That's just how corporate bullsh*t works in the US.
Never forget, advertising and marketing is just a friendlier names for Propaganda
Yep, fixed.
Metal, why don't you write for DSOG anymore? Did John try to corrupt you to write about Mods only and you didn't want to put your professional reputation on the line lol?
BTW, AMD is also further enhancing the "Super Resolution" upscaling capabilities with FSR Redstone.
The new tech will use an enhanced ML model, and upscaled in real time.
https://uploads.disquscdn.com/images/d28274132885ccc0d931b79cecf4cb4cb83332d49a7a3df00e2f1dce866c1852.jpg
….
If only all of the GPUs that can actually use this tech weren't selling for 150% of MSRP because of a shortage of 9000 series GPUs. Probably the best chance in the last 20 years to take away a significant market share from Nvidia and they grossly underproduce their product
AMD – We steal defeat from the jaws of victory …….