Baseless Next Generation Rumors with no Technical Merits [post E3 2019, pre GDC 2020] [XBSX, PS5]
- Thread starter BRiT
- Start date
- Status
So my theory is that PS5 is based on RDNA1 arch due to the need for native GCN/PS4 backward compatibility. It will still be fabbed with 7nm EUV to achieve higher clocks. Raytracing was custom-designed by Sony (thus we'll never see it on Desktop) and will mostly be used for Sound/Physics acceleration. I doubt it will be fast enough for any meaningful graphical purposes.
Xbox Scarlett will use the more advanced RDNA2 arch using AMD's own RT solution (which we will also get on Desktop cards next year). It will achieve backward compatibility via DX12, just like a PC.
Xbox Scarlett will use the more advanced RDNA2 arch using AMD's own RT solution (which we will also get on Desktop cards next year). It will achieve backward compatibility via DX12, just like a PC.
HBRU
Regular
I think Sony is targeting RDNA1 on 7nm TMSC and then on 6 nm that is an effortless evolution already announced... They go wide and they go bit above 10 TF... I say 10.32 TF
As I said they may evaluate to keep compatibility with Stadia's net & server infrastructure just in case MS agreement fails... That means be close, maybe a bit weaker than Stadia 10.7 TF....
As I said they may evaluate to keep compatibility with Stadia's net & server infrastructure just in case MS agreement fails... That means be close, maybe a bit weaker than Stadia 10.7 TF....
Last edited:
That's more like it, 12.9 Vega Tflops equivalent is where things should be at. We could then at least run Heretic level graphics at 1440p or 4k CBR at 30fps in a console space while the much improved Zen 2 cores, SSD and GDDR6/HBM2 take care of the rest of game mechanics. Imagine next gen Uncharted, God of War looking this good!
IBM started using 90nm soi in 2004, PS3 launched in 2006, don't remember anymore if the delay was more on cell or more on BD.
Anyway, Kutaragi was involved in this choice, so we are allowed to not consider it an example of rationality :O
PS3 was delayed because of BRD, but when it was designed, I'm reasonably confident the expectation was 65 nm would be ready. Unable to find the quote though.
AbsoluteBeginner
Regular
This is rational behind my 9TF projection.
9.75TF NAVI outperforms 10.5TF Vega56 by ~30%.
This would mean that something like 9TF Navi in console would perform better then 12TF Vega card. There is absolutely no need to speculate for 12-13TF Navi cards when we know die sizes/TDP and comparison to Vega cards.
9.75TF NAVI outperforms 10.5TF Vega56 by ~30%.
This would mean that something like 9TF Navi in console would perform better then 12TF Vega card. There is absolutely no need to speculate for 12-13TF Navi cards when we know die sizes/TDP and comparison to Vega cards.
Again, and its shocking how many times this needs be repeated, people need to stop fixating on the teraflop number! The next-gen consoles need next-gen visuals and games. If that can be achieved through 9 TFs of next-gen GPU, or 0.1 TF of quantum computing, that's a win!
Love_In_Rio
Veteran
Cant be RDNA1 because it lacks things that we know PS5 will include and are inside RDNA2 (Variable rate shading for PSVR2, Ray tracing SFUs...). IMHO RDNA1 is the way AMD took advantage this year to get something in the market to fight Nvidia from its real full next gen design (now called RDNA2) that is designed for 7nm+ process, that probably is more suitable for getting higher clocks.
Next Iphone will already use this year 7nm+ with the A13 SOC.
It helps now that we have and will have when the Navi cards hit retail what kind of performance we could get. It's still one of the easiest ways to rate the power of a Graphics card. It's obviously not the whole story but it gives an idea.
I'm happy if we get 12TFlop Vega performance from next gen consoles.
7nm+ is even better if you want to make the APU smaller (20% higher transistor density, about 10% reduced power consumption). That is my final prediction:
- 72 CUs RDNA2
- 7nm+
- Conservative clocks (this was PS4 winning strategy): about 1400mhz = 12.9 Tflops
This means the cooling inside a small box (but no smaller than Pro) should be possible with a vapor chamber.
AbsoluteBeginner
Regular
Of course, but alot of people believe that 40-36CUs at 1800MHz is too little for next gen consoles, but this is h
So 500mm2+ SOC and 280W TDP? Doubt it.
The notion that AMD has their own RT acceleration solution targeting 2020 (probably targeting an earlier release a couple years ago, before navi's delays) but both MS and Sony would develop their own in-house RT from scratch is ludicrous to me. AMD's roadmap is planned for and worked on about as long in advance as consoles are. It's not because we only heard of AMD's ray tracing now that it only became a thing now. Sony amd MS were certainly informed of AMD planned future tech, and I can't see anyone of those two chosing to build their own disruptive proprietary tech when they know the industry will go a different direction. They've both shown a certain commitment to avoid straying from standardised tech, and they see HW similarity with their competitors as a strenght, not a wealkness.
borntosoul
Regular
72 cu is definitely not out of the question, seeing as stadia has 56cu, why I’m quoting stadia a lot is because that’s the closest thing we have to next generation Xbox and ps.
Look at how much die sizes have increased over the years. Anything is possible as long as they can cool it and pay for it.
AbsoluteBeginner
Regular
VGleaks had stonking record for leaks last gen, and its funny to get back to their reported PS4 (Orbis) revisions back in early 2013.
So first dev kit for PS4 was released around July 2012 and was PC box with Bulldozer and discrete GPU inside.
First SOC based hardware came only in January 2013 and had pretty much retail hardware with 8GB of RAM (this is why I dont believe current dev kits have retail SOC or Navi containing one for that matter).
http://vgleaks.com/orbis-devkits-roadmaptypes/
So first dev kit for PS4 was released around July 2012 and was PC box with Bulldozer and discrete GPU inside.
First SOC based hardware came only in January 2013 and had pretty much retail hardware with 8GB of RAM (this is why I dont believe current dev kits have retail SOC or Navi containing one for that matter).
http://vgleaks.com/orbis-devkits-roadmaptypes/
AbsoluteBeginner
Regular
You are comparing completely different architectures. We know Navi with 40CUs is 251mm2, which, coupled with Zen2 will bring your SOC to ~320mm2 without RT hardware that is said to be included.
Basically, 40CU + Zen2 = PS4Pro die size on more expensive node. 40CU + Zen2 + RT hardware = >350mm2 on more expensive node.
There is absolutely no way these new consoles will feature anything north of 56CUs, let alone 72.
borntosoul
Regular
Your probably right, we don’t know and I’m being optimistic, but l do know that die sizes have been increasing a lot over the last few years so comparing dies sizes to last gen is old hat.
Otherwise we would have little performance increases if we stayed with the same die sizes for the last 5 years.
AbsoluteBeginner
Regular
Look at my post here and AMDs official performance difference between 9.75TF Navi and 10.5TF Vega.
Navi is on avg 30% faster, so Navi with 36/40CUs at 1800MHz would be faster thrn Stadias GPU.
Given RT isn't appearing in AMD's GPUs alongside next-gen hardware, I wonder if there are elements that aren't suitable for a PC solution meaning the tech isn't ready, but which would work in a console? Or at least provide a useful improvement in a console where it wouldn't be particularly useful for a PC? Though I've no idea what form that could take.
Alternatively, given MS also have photon mapping tech patents, maybe Sony and MS are getting a joint photon-map+KDTree solution unique to consoles?!
- Status
Similar threads
- Locked
