Next Generation Hardware Speculation with a Technical Spin [post E3 2019, pre GDC 2020] [XBSX, PS5]

[PSman1700]

Why yes.

Why would it be?
C'mon AMD will have VG out in >1 design before this thing launches.

Most claimed rdna1.5 for both consoles, turns out its 2 for xsx, most likely ps5 also.

 

[PSman1700]

the XSX sits easily between 2080 and 2080ti.

For that we have to learn more about the gpus. A 2080 by then is two plus years old, it would match a 3060/3070 by then, so yeah, possible.

 

[Bondrewd]

Most claimed rdna1.5 for both consoles, turns out its 2 for xsx, most likely ps5 also.

No idea about PS5, since Sony is *weird*.
XSX was always what they've just announced.

 

[Kaotik]

What would be the difference between "custom" RDNA2 and RDNA2?

It's not "custom RDNA2", it's "custom RDNA2 + Zen 2 processor". Which is just another way of saying semi-custom RDNA2+Zen2 APU.

 

[ultragpu]

So verified Era insider hmqgg was wrong...hmm

In any case, I am 100% Sony is using RDNA2 as well, and Ariel was likely a chip based on Navi 10 made as BC testbed while Oberon, a chip featuring VRS and RT, was getting ready.

If that's the case then it would mean 7nm+ EUV and definitely get much more than just 36 or 40 CUs shouldn't it?

 

[Deleted member 13524]

Kleegamefan is one of the very few who claimed 12 TFLOPs RDNA2 for the SeriesX, IIRC.

Leather jacket Jensen Huang probably as well since the XSX sits easily between 2080 and 2080ti.

That presentation about the laptop with a 2080 being better than next-gen consoles was so much bullshit..

The laptop 2080 has abysmal clocks. Its sustained performance is barely above a desktop GTX 1080, which in turn is around hardly faster than a desktop 2060.

 

[AlNom]

Leather jacket Jensen Huang probably as well since the XSX sits easily between 2080 and 2080ti.

View attachment 3596

Must have been a topic at the Thanksgiving dinner.

 

[PSman1700]

No idea about PS5, since Sony is *weird*.
XSX was always what they've just announced.

Kinda would align with the testlab leak where xsx had rt but not ps5, rdna1.5, and ps5 being more mature (ready desingn before xsx)?

 

[turkey]

If that's the case then it would mean 7nm+ EUV and definitely get much more than just 36 or 40 CUs shouldn't it?

There was a good post here that claimed navi 5700 is already on N7P so it would seem that's the baseline design.

Ps5 will use that, but as Navi is already there you cannot double dip on the perf / heat savings when trying to scale up.

 

[Scott_Arm]

.

RDNA 2 the only semi new info.

Well, it's official now, so it's one less thing people have to argue about.

 

[ultragpu]

Kleegamefan is one of the very few who claimed 12 TFLOPs RDNA2 for the SeriesX, IIRC.



That presentation about the laptop with a 2080 being better than next-gen consoles was so much bullshit..

The laptop 2080 has abysmal clocks. Its sustained performance is barely above a desktop GTX 1080, which in turn is around hardly faster than a desktop 2060.

By the beard of Kratos hope he's onto something:sly:, if PS5 turns out to have more flops like he says...

 

[Globalisateur]

I am a bit surprised about that RNDA2 claim to be honest (still waiting for the receipts on that one). The rest is exactly what most of us were expecting like patented custom VRS API and Direct X Ray tracing.

 

[ultragpu]

I guess if Sony don't make a quick counter or statement on power in the next few days it would be safe to say they have the weaker system. Things should pick up the pace now, interesting times ahead.

 

[AlNom]

I am a bit surprised about that RNDA2 claim to be honest (still waiting for the receipts on that one). The rest is exactly what most of us were expecting like patented custom VRS API and Direct X Ray tracing.

Eh... it's just a label. Could be as straightforward as adding RT & VRS aside from any fab-related maturation.

GCN over the years had various labels.

GCN1.0 = Tahiti/Pitcairn/Cape Verde
"GCN1.1" = GCN2 = Bonaire/Hawaii, TrueAudio DSP, scaling up block units (SEs/ACEs)
"GCN1.2" = GCN3 = Tonga, delta compression, ISA changes
GCN4 = Polaris = 14nm, various updates
GCN5 = Vega = 14/7nm, various updates, HBCC, packed math

Nevermind the internal gfxip numbering.

 

[Xbat]

I am a bit surprised about that RNDA2 claim to be honest (still waiting for the receipts on that one)

I don't see how clearer it could be with Microsoft themselves saying RDNA 2.

 

[JoeJ]

With MS using 2, it's almost ensured Sony uses 2 too? This reduces the chances Sony has some custom RT solution as well i think. Same feature set for both, except VRS which could be just a software thing.

BTW, no talk about mesh shaders or anything about geometry.

 

[iroboto]

Pretty much nothing.

security from hacking and running copied wares.
Virtual Machine customizations to run faster in that type of environment.
etc etc.
its certainly heavily customized. Not necessarily for improving graphical performance though.

 

[Deleted member 13524]

I am a bit surprised about that RNDA2 claim to be honest (still waiting for the receipts on that one).

Why would you, or anyone? AMD is launching more than one discrete GPU later this year with RDNA2. They're not known for pulling a G80-3-days-before-PS3, on the contrary.

And we don't really know what RDNA2 is. Architecturally it could be RDNA1 + RT + VRS, with more enhancements on the electrical level to allow lower power consumption and heat dissipation at a given clock speed.

With MS using 2, it's almost ensured Sony uses 2 too?

I don't think so.
The PS4 Pro using Vega-like CUs with RPM didn't make them come to the XBoneX.

Sony could be using the older instruction set + custom raytracing, for example.

BTW, no talk about mesh shaders or anything about geometry.

Nor memory bandwidth and amount.

 

[JoeJ]

Side question - I did not really look into VRS yet, and i wonder...
It's clear how forward shading can take benfit.
But deferred? A gbuffer is just flat memory with texture tile order? Or does VRS allow virtual tiling with different resolutions somehow? I don't see how this could work, especially when thinking of compute shaders working on gbuffers. I can only think of multiple threads ending up processing the same data which is no real win?

Edit: But pixel shader could work. So i conclude only pixel shaders give the full advantage with VRS + deferred.

 

[Kaotik]

Sony could be using the older instruction set + custom raytracing, for example.

Do you seriously consider it even a remote chance that Sony would develop their own RT acceleration units or buy them from someone else to be integrated into semi-custom AMD APU, when AMD is offering RT acceleration too?