Support NeoGAF

[winjer]

What exactly did they say that was in error?

They are still comparing performance with the old dll of FSR 4.0.2, which had performance issues on RDNA2 and a bug that cause ghosting.
The fixed dlls are already out for several months, but somehow DF is still unaware of what is going on.

 

[Zathalus]

They are still comparing performance with the old dll of FSR 4.0.2, which had performance issues on RDNA2 and a bug that cause ghosting.
The fixed dlls are already out for several months, but somehow DF is still unaware of what is going on.

FSR4, even with those fixes, runs worse on RDNA3 compared to RDNA4, and way, way worse on RDNA2. Perfectly usable on RDNA3 like Oliver said, and even likely usable on RDNA2. But the performance uplift is worse on every preceding generation.

Nothing they said in that segment was factually wrong.

 

[winjer]

FSR4, even with those fixes, runs worse on RDNA3 compared to RDNA4, and way, way worse on RDNA2. Perfectly usable on RDNA3 like Oliver said, and even likely usable on RDNA2. But the performance uplift is worse on every preceding generation.

Nothing they said in that segment was factually wrong.

The expected performance they stated was wrong. FSR 4.0.2C is now closer to XeSS in DP4A, meaning it's more usable than with the first dll.
And then they talk about TOPs, as if an upscaler uses all 300 in the Pro at once. Which is complete nonsense.

 

[Zathalus]

The expected performance they stated was wrong. FSR 4.0.2C is now closer to XeSS in DP4A, meaning it's more usable than with the first dll.
And then they talk about TOPs, as if an upscaler uses all 300 in the Pro at once. Which is complete nonsense.

I guess we will see. I'm near certain that unless they reduce the quality FSR4.1 will run a bit worse on RDNA3 and substantially worse on RDNA2. Which is exactly what Oliver was claiming.

 

[PaintTinJr]

FSR4, even with those fixes, runs worse on RDNA3 compared to RDNA4, and way, way worse on RDNA2. Perfectly usable on RDNA3 like Oliver said, and even likely usable on RDNA2. But the performance uplift is worse on every preceding generation.

Nothing they said in that segment was factually wrong.

We'll agree to disagree, but I'm happy for them to be trashing the PS5 capabilities at this point because Sony have totally dialled it in this gen and seem unbothered by weak SW2/SD hardware leapfrogging large spec deficits using ML AI to close in on the PS5 experience, so this negative shade might actually get Sony to pull a finger out and deliver better results on the OG PS5 and Pro.

 

[winjer]

I guess we will see. I'm near certain that unless they reduce the quality FSR4.1 will run a bit worse on RDNA3 and substantially worse on RDNA2. Which is exactly what Oliver was claiming.

No one here is saying that the performance rating will change. It will still be RDNA4, then RDNA3, then RDNA2.
The problem is that their numbers for RDNA2 are completely wrong.
Not only that, but they also used the injection method for Optiscaler, instead Optipatcher where available.
FSR4 on RDN2 being injected by the driver will have the lowest overhead, improving performance a bit further.

• OptiPatcher helps avoid performance overhead or crashes/artifacts caused by Dxgi spoofing!