AMD just didn't have the resources to do a high-end GPU whilst building CDNA, UDNA, APU's and worrying about RDNA4, their GPU teams whilst capable are significantly smaller than nVidia.

In-saying that, the 9060XT 16GB, 9070 and 9070XT are all absolutely brilliant in terms of price/performance/features.
If next-gen is based on RDNA4, I would still be pretty content considering how capable that hardware is.

AMD does need to pivot back to it's small-core design philosophy that they had with Terascale 1 and Terascale 2, which pushed them ahead of nVidia in terms of price/power/performance. That was peak AMD Graphics.

For those curious how big the differnce is between FSR 4 and DLSS 4.5 here it is:

https://www.youtube.com/watch?v=4JNmb54hR3A



FSR doesn't handle performance mode well.......  DLSS 4.5 can use lower base resolution and beat/match it.
For FSR to have a chance, it needs to run Quality mode.

Where did you hear about that? Not saying you're wrong, just that this is the first time I've heard this and yeah RDNA 4 is for sure one of AMD's better lineups in the past decade. They won't age quite as well as the RTX 50 series but the gap will be closer than RDNA 3 and the 40 series and way closer than RDNA 1 and the 20 series so hopefully the gap gets even smaller next time.

The nice thing about scaling is that eventually you can use transfer learning, distillation, etc to make smaller models better too. I suspect we'll see a lot of that going forward, and Nintendo will reap the benefits of Nvidia's advancements. DLSS lite and the regular CNN model aren't going to stay at their current quality-per-compute forever. 

There are many CNN-transformer hybrid architectures that are probably going to replace pure-CNNs for small-param models. 

There is plenty of articles around with interviews with AMD where AMD forked Graphics Core Next into CDNA and RDNA in order to build micro/macro-optimizations for those different and specific workloads and market segments, but the engineering resources to maintain that has been an irksome area of contention.

https://overclock3d.net/news/gpu-displays/amd-plans-major-changes-for-radeon-with-united-udna-gpu-architecture-for-data-centers-and-gamers/

It's the same issue AMD had prior to 2015, where the GPU development team was part of the CPU development team to drive their Fusion (APU) initiative, sadly it meant a contention of limited resources which impacted all of AMD's engineering teams.

https://fortune.com/2015/09/09/amd-graphics-chip-split/

Nice jump in image quality! Definitely seems worth the extra overhead, especially to get rid of those artefacts. Looking forward to my next visit to Night City with it.

I've been enjoying this on my 5080. The biggest difference I noticed is less ghosting with frame gen, unless that's just my imagination, and the new version also looks a bit sharper.
I'm interested in trying the 6x frame gen and dynamic multi-frame gen when that becomes available too.

I do wonder if DLSS improvements may filter down to Switch 2; obviously with its limited power as a portable device I don't expect it to run the latest greatest version, but it would be cool if they're able to improve the system's DLSS over time.