That's just speculation though; it's not available on PS/Nintendo, it likely never will be. 

Gamepass has nowhere near the ubiquity of Netflix, and it's unlikely it ever will.

Going from RDNA 2 to RDNA 4 in the same "tier" of product segment, there is definitely a massive leap.

It's all in the Ray Tracing and A.I. upscaling which are providing some of the largest improvements I have seen in a very long time.
In RT alone I am seeing a 5x improvement.

...And with Neural materials, textures, compression, shaders and more all coming before next-gen... We could see some extremely large visual gains if AMD is able to adopt and implement them in a tasteful manner in time. (They have a couple years yet.)

There is a paradigm shift occurring in the industry (Hardware and software) in how we render games, which has historically provided the largest leaps in graphics in previous console generations.

Gamepass allows you to stream console quality games to mobile.

If game streaming takes off, that's when Gamepass will take off in massive numbers I think.

If it pans out that would be cool, I'm just sceptical for the moment, as the last two gens were also promised to be the "biggest leap ever" yet diminishing returns were still a thing compared to prior gens.

Well, if the leaks are to be believed (and apparently they come from a source that was pretty spot on in previous occasions), UDNA is 2x in RT/AI per CU than RDNA4. So compared to current gen consoles (which are RDNA2 with very poor RT, with no dedicated AI hardware), this will be humongous leap in those particular aspects (compare that to very modest 20% per CU increase in raster over RDNA4, which is already way ahead of current gen consoles).

Of course, it will be up to devs to use this how they see fit - personally, I'm most excited to see what this brings to gameplay, since even something as "ordinary" as full path traced lighting can be huge gameplay-wise for dungeon crawlers and stealth games - but I'm mostly excited to see how it goes with AI NPCs and similar stuff.

There will be a gen-on-gen improvement in RT.. That's not really up for debate as it's what nVidia and AMD are concentrating on in their GPU designs currently. (And A.I)

That's actually not true.

RDNA2 actually has a Ray accelerator in each CU.

RDNA3 actually keeps the single Ray accelerator in each CU, but tends to offer more CU's or clocks at every Tier... Plus a few efficiency improvements in the RT accelerators, but that wasn't the focus of the design... Which is why in RT the jump between the 6600XT and 7600XT wasn't massive.

https://gamersnexus.net/gpus/amd-radeon-rx-7600-xt-gpu-benchmarks-review-power-efficiency-gaming#7600-xt-ray-tracing-benchmarks


With RDNA4 - AMD implemented a crap-ton of improvements, but still kept the single Ray Accelerator per CU... Improvements like a doubled intersection engine in each Ray Accelerator (Which allows for more ray-box and ray-triangle intersections per cycle) and a wider BVH as well as BVH compression and more.

https://chipsandcheese.com/p/rdna-4s-raytracing-improvements

And despite the same number of CU's and Ray Accellerators... I still saw an upwards of a 4x-5x improvement in RT heavy games like Oblivion Remastered and Cyberpunk at 1440P ultra settings. (But limited VRAM likely played a role in that on the 6600XT)


https://www.techspot.com/review/2996-amd-radeon-9060-xt/


Considering the de-focus on raster and priority focus on RT and A.I... A doubling in RT over the RDNA4 parts would be entirely what I would expect for next-gen UDNA GPU hardware... Especially if AMD -finally- re-aligns how many ray accelerators they have in their parts.

I am normally extremely skeptical of outlandish claims in the technology space.
However... I cannot understate how massive the potential next couple of years is going to be in regards to rendering technology is. - I haven't been this excited since we started to see programmable pixel shaders.

AMD and nVidia are investing 10's of Billions in A.I and RT stuff.

Sorry mate, mostly taking issue with your statement that the RDNA2 architecture doesn't have dedicated RT hardware. I should have made that point clearer.

No problem mate, but I think you've misread...it was "very poor RT, with no dedicated AI hardware"

Ah. Yeah. Did misread that.

In saying that... The A.I Cores is just replicating certain formats from the CU's into separate cores.
RDNA2 supports all the important stuff like bfloat, INT4, INT8... Where RDNA2 falls behind is FP8.

The advantage of separate cores for A.I is that it doesn't tie up CU's used for rasterization/ray tracing, but otherwise they don't technically bring new functionality.

So on a technicality, RDNA2 does have "AI Hardware" just not as separate independent cores... But RDNA3 doesn't support every AI operating on it's A.I cores anyway, so some AI operations fall back onto the regular CU's.

RDNA4 stepped things up further again. - In some instances I am seeing an 8x gain over RDNA2 for AI stuff... A.I video upscaling saw a 12x fold performance improvement depending on algorithm.

That's just speculation though as you dont know if they wont..