By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Pemalite said:
Kyuu said:

In AMD terms, Kepler said it would be around 9070XT level (which is comparable to 5070 Ti in raster), but be more advanced than Blackwell (which is superior to RDNA4). All in all, the implication is that it would be around or slightly weaker than a 5080, even with the more advanced architecture/feature-set in mind. Alex Battaglia thinks having a 9070XT+ of raster performance is too high, unnecessary, and supposedly contradicts Cerny's vision.

There wouldn't be anything wrong with a 9070XT level of raster performance.. That's still a doubling in capability over the PS5.
The RTX 5080 is a 368mm2 chip with 45.6~ billion transistors @360W TDP not including the CPU.
For comparison sake the PS5 chip is 270mm2 with 10.6~ billion transistors with a TDP of 225W for the entire SoC.

An RTX 5080 sized chip in a console is just not going to be feasible for a very long time to come, plus the RTX 5080 represents nVidias "Go big or go home" approach, where their high-end GPU's are made to be as large as possible without any regard for efficiency or cost.

I have to agree with Alex Battaglia that it's pretty unnecessary to have stupidly high levels of rasterization performance going into next-gen, the focus is on Ray Tracing and A.I.
Ray Tracing is the future in video game graphics rendering... And costs are stupidly important right now, so something needs to be prioritized in order to get the best bang-for-buck.

Plus A.I will reduce memory demands, will reduce rasterization demands (Better upscaling), will reduce RT demands, so the chip can do more work with less.

I also think the RAM improvements will be less of a leap... I would expect around 24GB... Which might seem garbage, but the usable memory space will be a lot larger thanks to the new compression being deployed in AMD's next-gen GPU's.

Kyuu said:

What I'm saying is that PS6 matching the upper end of midrange AMD GPU's from 2027 or later requires a greater offering than what's being implied by those rumours. But I realize that they're just rumours, and that a lot of the leaked plans and targeted specs are subject to change. Rumours claimed a 2027 release, when Cerny's wording kinda makes it clear the console(s) isn't coming before 2028. At this point I'm not sure if the handheld thing is even happening.

I don't disagree. But I think the Playstation 5 highlights that in the current climate, costs are an important factor.

AMD and Cerny likely have other plans and technology for the next-gen chip that they haven't highlighted yet either, I am sure those details will be showcased at some point.

What will be interesting is if Microsoft beats Sony to the market with the Xbox 5 in 2026 and what kind of implications that might have... Another Xbox 360 moment sans-platform exclusives? Time will tell.
Sony made the right moves this generation as the PS5 chip is smaller, cheaper and "good enough" compared to the Series X, but if Microsoft has the only next-gen console for a couple of years, that might shake things up.

Sometimes just making a chip larger and more power hungry just doesn't provide any real tangible benefits as Microsoft learned the hard way.

But 2-3 times the raster performance for an 8 year later successor isn't really a huge gap at all per generational standards, so I don't see how it's unrealistic or contradictory to Cerny's vision. Typical raster jump is closer to 8-10 times if not higher every 6-7 years. A 2-3x jump would be modest in comparison, and nothing extraordinary even if the priority goes to RT, ML, efficiency, etc.

Digital Foundry has imo been underestimating console performance since at least the PS5, and I think this is going to continue with PS6 where Alex seems to think the rumoured specs are optimistic (the same specs that I consider underwhelming lol). 5070ti-5080 of total performance would be too weak for 2028, maybe unless the targeted price is $500 or lower. Upper midrange 2027/2028 GPU's should be a lot more capable than a 9070XT even in raster performance. I don't think raw rasterization will abruptly stop improving just because the priority is elsewhere.

I don't know... Perhaps the lower specs are needed to make scalability smoother between it and the rumoured handheld.