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

Status
Not open for further replies.
Since this is supposed to be the hardware prediction thread, I will make a very bold hardware prediction.

PS5 will have ID buffer, and fp16 RPM.

Anyone disagree?

I'll add to it...

Bespoke CBR logic will be in both next-generation systems.
 
If you think native 2160p will be the target for the next generation consoles along with ray tracing and higher framerates, you'll be really disappointed.

I agree. Not on the 2020 base consoles anyway. Such a target might be within reach of a PS5 Pro or Scarlett X, the next-gen midgen upgrades, assuming they happen in 2023/2024.

Lets say they have somewhere between 2.3x and 4.5x the flops of the 2020 base consoles (like PS4 Pro and Xbox One X respectively). Coupled with Zen 4 IPC gains and some good clock headroom/TDP.

All on 5nm+ with good use of EUV, or 4nm or whatever
still FinFET, and assuming TSMC 3nm GAFFET nanowire (or whatever they're doing after 3/4nm, is still a year or two away by that point)

Anyway, 2160p - 60fps - raytracing might happen then.

Alternatively
 
I assume the commentary around ray tracing and GI is around our modern implementations as well as the new hybrid ones. In isolation I wouldn't say that this piece of news is a slam dunk on dedicated hardware for PS5. But with respect to everything else we know, this is a safer assumption.
 
I assume the commentary around ray tracing and GI is around our modern implementations as well as the new hybrid ones. In isolation I wouldn't say that this piece of news is a slam dunk on dedicated hardware for PS5. But with respect to everything else we know, this is a safer assumption.

How are we deciding what is dedicated RT hardware?

Are we simply looking at cores/logic strictly dedicated to partial or full RT algorithms?

If so... then AMD's RT solution of an improved CU design (over the current Navi based 5700/XT cards) towards Compute related RT acceleration is exactly what then?

Or are you simply stating there is no official info on PS5 having dedicated RT hardware... or at the very least AMD's compute based RT acceleration solution?
 
How are we deciding what is dedicated RT hardware?

Are we simply looking at cores/logic strictly dedicated to partial or full RT algorithms?

If so... then AMD's RT solution of an improved CU design (over the current Navi based 5700/XT cards) towards Compute related RT acceleration is exactly what then?

Or are you simply stating there is no official info on PS5 having dedicated RT hardware... or at the very least AMD's compute based RT acceleration solution?
Correct no official info on PS5 having dedicated hardware. But this lines up with enough unofficial information that it exists.

Calculating the intersections using compute is the only method if there is no dedicated hardware to calculate the intersection. Whether it’s a core or not a core: if a dedicated piece of silicon isn’t doing the intersection or rather cannot do the intersection many magnitudes faster than generic compute shaders; I would say that it’s not accelerated but dedicated. As in support for Ray Tracing with specialized hardware, but you’re not breaking any speed barriers here. This would apply to both consoles.

Nothing is still quite confirmed yet.
 
How are we deciding what is dedicated RT hardware?
Currently we make assumptions mostly based on AMDs TMU patent, which is a FF solution for both traversal and triangle intersection (notice the former is the main problem because of random memory access and often 10 times more bounding box intersection tests than final triangle tests in practice.)
I would call this compute assisted, not compute based. Compute may be necessary to control the outer loop, e.g. moving from one level of BVH to the next (NVs RT Cores do this independent from shader cores, as far as we know).
But they mention this to be optional in the patent, and could also add another FF unit to control this outer loop (likely negligible area costs).
Releated questions would be: Would additional flexibility of compute based, fine grained control be worth it? Would additional FF control still limit compute perf on other tasks too much, because memory systems are shared?
And ofc. having a patent does not mean they will use this method it at all. Could end up completely different. But i don't see a chance for a pure compute based solution, also not by combining CPU and GPU.
 
Correct no official info on PS5 having dedicated hardware. But this lines up with enough unofficial information that it exists.

Calculating the intersections using compute is the only method if there is no dedicated hardware to calculate the intersection. Whether it’s a core or not a core: if a dedicated piece of silicon isn’t doing the intersection or rather cannot do the intersection many magnitudes faster than generic compute shaders; I would say that it’s not accelerated but dedicated. As in support for Ray Tracing with specialized hardware, but you’re not breaking any speed barriers here. This would apply to both consoles.

Nothing is still quite confirmed yet.
Come on we have official info: The Cerny interview. A GPU with Ray Tracing support can only be hardware based. Cerny will have never talked about software Ray Tracing for PS5 when even PS1 can already do that.
 
Come on we have official info: The Cerny interview. A GPU with Ray Tracing support can only be hardware based. Cerny will have never talked about software Ray Tracing for PS5 when even PS1 can already do that.

I recall Cerny mentioning Ray Tracing lighting support in the Wired interview, but never linking it to the GPU or CPU. He gave a more detailed outlining of PS5 having a dedicated audio chip towards RT sound refractions and other real-time audio algorithms. But nothing solid on which processor (i.e., GPU, CPU or possible dedicated silicon) RT lighting, RT shadowing, RT reflections and so-on are being performed on.
 
On last week I read article about AMD roadmap on Finnish site and it stated that next gen consoles will use RDNA2. And we know that RDNA2 will have some kind of raytracing support.

So it will be hardware based on some level at least.

Did I miss something or did others miss this?

source:
http://ir.amd.com/static-files/240c38bf-fe60-4fa1-b20c-a36bee0c12ca slide 27 "next gen consoles with next gen rdna"

Consoles = more than one and we know that it wont be Nintendo, we also know that next gen xbox uses raytracing = must be PS5, right? And there is also pic of xbox + playstation lols
 
Last edited:
Come on we have official info: The Cerny interview. A GPU with Ray Tracing support can only be hardware based. Cerny will have never talked about software Ray Tracing for PS5 when even PS1 can already do that.
We have information on hardware support. I don’t know what that means in terms of acceleration. I guess that is what is in question for both consoles. I’m debating how much investment they both have into RT; as opposed to debating whether the future consoles will be able to support it.
 
I recall Cerny mentioning Ray Tracing lighting support in the Wired interview, but never linking it to the GPU or CPU. He gave a more detailed outlining of PS5 having a dedicated audio chip towards RT sound refractions and other real-time audio algorithms. But nothing solid on which processor (i.e., GPU, CPU or possible dedicated silicon) RT lighting, RT shadowing, RT reflections and so-on are being performed on.

Cerny never mentioned what kind of hardware implementation there will be for real-time ray tracing on the PS5. At this point it could be:

- The same GPU implementation that AMD will implement in their PC Radeon products (e.g. TMU based like we saw in AMD's patents);
- A custom CPU implementation similar to what we could see in other implementations, (e.g. one o the rumors for Scarlett);
- A custom and exclusive implementation that will only be present on Sony and nowhere else, like we saw with e.g. ID Buffer and some other custom hardware dedicated to checkerboard (e.g. separate chiplet).

What should be clear of any doubt is that the PS5 is coming with hardware dedicated to real-time ray tracing (anyone claiming otherwise is just wishful thinking IMO).

I know there are some who assume the obvious situation is for Sony and Microsoft to use the exact same implementation of real-time raytracing. However, the recently available roadmaps leave some space for discussion here.
We have AMD saying next-gen RDNA GPUs will have ray tracing hardware. In separate slides they claim RDNA 2 (still in design at the moment) at 7nm+ is coming until 2021, meaning it could come in 2021 only 2 years from now.
Is next-gen RDNA the same as RDNA 2, with the ray tracing PC GPUs only coming in 2021? Or could we see in 2020 something similar to Zen 1.5 (14->12nm transition) where ray-tracing in RDNA (next-gen RDNA) is actually coming with a RDNA 1.5 iteration?

If it's the first option, we could be seeing drastically different implementations between consoles' ray tracing and PC GPUs from AMD because they're being developed within very different time periods.
 
If we ignore the options and just look at final output performance; if we are getting RT support that is less performant than 2060RTX, that’s sort of where I’m unsure of how dedicated they are to RT. I would call that hardware support though.

Imo; For RT to be a future for next gen consoles (a generation length); the RT performance should beat gen 1 RTX RT cores and there should be lots of them for RT processing
 
Last edited:
I recall Cerny mentioning Ray Tracing lighting support in the Wired interview, but never linking it to the GPU or CPU. He gave a more detailed outlining of PS5 having a dedicated audio chip towards RT sound refractions and other real-time audio algorithms. But nothing solid on which processor (i.e., GPU, CPU or possible dedicated silicon) RT lighting, RT shadowing, RT reflections and so-on are being performed on.
Actually, you are confused here. They actually first talked about a GPU with RT support, comparing to Nvidia RTX GPUs in order to do graphic Ray Tracing "Ray tracing’s immediate benefits are largely visual." In a later tweet the author precised their whole RT discussion was mainly discussing graphics benefits (he probably didn't transcribe the whole thing in his interview). Then Cerny talked about applying Ray tracing to sound, then he talked about another thing, some custom 3D audio unit (probably created first for PSVR).

So he talked about 3 different things here. He is a man about details, some may get confused. He did a similar thing with the GPGPU 14 + 4 thing on PS4 confusing many people, even tech savvy ones, for many years.
 
Last edited:
Yep. If you're going in, go all in. Little to be gained from half-arsed implementations. You'd get ineffective RT and a loss of normal compute power with a weak compromise.
It would be absurd! Like if a company uses die space for ineffective esram just to have 8GB of ram, and when processes and technology allow it, sells a half gen console with a better implementation.
 
Status
Not open for further replies.
Back
Top