• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Pachter: PS5 to be a half step, release in 2019 with PS4 BC

Lady Gaia

Member
The age of Moore's law is basically ending in the next decade.

Maybe, but I've heard similar gloom and doom in the past regarding headroom for technical advancement. I clearly remember arguments about why 50MHz was an absolute ceiling for clock speeds, which seems cute and quaint in hindsight.

It is clear that there are significant hurdles that will prevent simply following the existing path to smaller silicon feature sizes. It's less clear that there aren't alternatives that will yield some pretty serious wins. There's enough economic upside once silicon hits a ceiling that it's worth finally looking at gallium arsenide or gallium nitride or other alternatives.

... but I completely agree with your point that the stall is likely to make a mid-generation bump an unlikely proposition. I think the timing of 4K consumer adoption made it more or less necessary this time around, and I don't see 8K hitting the mainstream in a similar way early in the next generation.
 

onQ123

Member
....

It has taken me a long time to figure out what you are saying here.

So, if I'm understanding you right, your arguments are:

'Sony added the id buffer to make solving a very specific problem more efficient, so that means they could add full ray tracing hardware to make algorithms for shadowing and reflections more efficient'

?

And also 'developers develop new techniques over time'

?

If this is the case, then the answers to both these things should be astoundingly obvious: 'lol' and 'duh'.


FWIW, in your quote, if you are saying that the id buffer isn't screen space, then I'm sorry but you're simply wrong and you fundamentally misunderstand its purpose.
The 'world space' part in that quote obviously refers to the depth buffer.

Yes to the 1st part but no to full ray tracing hardware part, I'm talking about things much like the ID buffer adding small things to the GPU that can help with things like lighting. because these SoC/APUs have shared address space extra things like the ID buffer hardware can be added to make it more of a MIMD processor so that the jobs can be broken down so it can handle things like ray tracing better than older GPUs.

I know the ID buffer is being used in screen space but the tracking data it's giving is from world space & things like that is what can be used for cheaper ray tracing tricks. they could add hardware for a Light-buffer or ZZ-buffer so they will have that information without using the ALUs of the GPU then use that for shadows & reflections.



Graphics processing enhancement by tracking object and/or primitive identifiers

59. A graphics processing unit (GPU) comprising: an ID module configured to: receive information corresponding to a plurality of objects in a scene in virtual space, each said object being defined by a set of vertices; receive one or more object identifiers in a command buffer for each said object in the plurality, and write the one or more object identifiers to an ID buffer; wherein the GPU is further configured to perform geometry shading in which one or more output primitives are generated from one or more input primitives, wherein performing geometry shading includes generating a primitive identifier for one or more primitives generated by the geometry shading by incrementing a primitive identifier for each input primitive by a maximum amount of geometric amplification of output primitives supported by an object with which the input primitives are associated; and wherein the GPU is further configured to render the two or more instanced objects by: manipulating parameter values of the vertices to generate output vertex parameter values; setting up a plurality of primitives from the vertices, each said primitive being defined by a set of one or more of the vertices and each said primitive belonging to one or more of the objects; rasterizing each of the primitives for a plurality of pixels; writing a depth at each said pixel to a Z buffer, the Z buffer having a first resolution; writing one or more of the object identifiers to at least one ID buffer, such that each pixel in the Z buffer is associated with an object identifier in the ID buffer; processing the pixels for each of the rasterized primitives to determine a set of color values in a color buffer, the color buffer having a second resolution that is less than the first resolution; and determining a set of display pixels for an output frame at the resolution of the Z buffer using the color values in the color buffer and the identifiers in the ID buffer.



The+Light+Buffer+Used+to+accelerate+shadow+calculations+using+the+directional+idea.+Assumes+point+light+sources..jpg



So how would 'not full' ray tracing work.


Hybrid Ray Traced Shadows



Hybrid rendering for real-time lighting: ray tracing vs rasterization
 
Maybe, but I've heard similar gloom and doom in the past regarding headroom for technical advancement. I clearly remember arguments about why 50MHz was an absolute ceiling for clock speeds, which seems cute and quaint in hindsight.

It is clear that there are significant hurdles that will prevent simply following the existing path to smaller silicon feature sizes. It's less clear that there aren't alternatives that will yield some pretty serious wins. There's enough economic upside once silicon hits a ceiling that it's worth finally looking at gallium arsenide or gallium nitride or other alternatives.

... but I completely agree with your point that the stall is likely to make a mid-generation bump an unlikely proposition. I think the timing of 4K consumer adoption made it more or less necessary this time around, and I don't see 8K hitting the mainstream in a similar way early in the next generation.

Whilst gallium arsenide and gallium nitride have some strong benefits over silicon for use as mediums to build integrated circuits, there are significant factors which make these materials in many ways inferior to silicon.

Given the discussion about Moore's law and the end of the silicon roadmap, it's worth mentioning that the significantly higher impurity density of GA/GN means they cannot be manufactured to geometries as low as silicon can. So in-spite of the higher clock-speed ceiling they afford, they'll most likely still be unlikely to outperform the top end silicon chip, while being significantly more expensive—even when you ignore silicon's economies of scale, GaArs and GaNit are more expensive materials.

Fundamentally, the limits that define the end of the silicon road-map are real physical limits... not just a matter of economics. So I think i take a much more pessimistic view than you on the replacement technologies that will fill the gap. We're almost at 5nm now, surely the "next big thing" in semiconductor manufacturing, that will step in take over from silicon wholesale should be universally known and understood by now... the fact that it isn't is frankly a worrying sign, imho.
 
This is what I expect as well, combined with a decent CPU upgrade allowing for native 4K, improved graphics and allowing AAA games to reach 60fps minimum in support of VR.

lol that depends entirely on the developer. If they want to prioritize graphical fidelity and/or resolution, it's not happening. You always have to balance the three out(fidelity, resolution, and framerate). Can't have the best looking and performing game without sacrifices and vice versa
 

Shin

Banned
I think you read too much into this two patent applications for BC by the PS4's (and probably also PS5's) lead system architect.

How so, what else would they be using it for, it could be filling for the sake of filling/protection.
But that does read like whatever CPU > older CPU's, perhaps their way of BC moving forward?
Regardless it's solid proof that they are working on BC in some form or another, which is great and probably takes the fear away for some.
What the end result is or will be is anyone's guess, at least it's not gossip.
 
How so, what else would they be using it for, it could be filling for the sake of filling/protection.
But that does read like whatever CPU > older CPU's, perhaps their way of BC moving forward?
Regardless it's solid proof that they are working on BC in some form or another, which is great and probably takes the fear away for some.
What the end result is or will be is anyone's guess, at least it's not gossip.

I think this is the important bit for many; especially here on Gaf.

To allay fears of some that PS5 won't have some form of BC (I mean, it might still not, but at least this lends further confidence in it actually happening).
 

c0de

Member
How so, what else would they be using it for, it could be filling for the sake of filling/protection.
But that does read like whatever CPU > older CPU's, perhaps their way of BC moving forward?
Regardless it's solid proof that they are working on BC in some form or another, which is great and probably takes the fear away for some.
What the end result is or will be is anyone's guess, at least it's not gossip.

The things described in there even partially fit for the Pro. And there is also nothing you wouldn't expect from a BC point of view, I'm surprised they filed a patent for that.
 
Good that Sony are already looking at BC. It's something they need to nail as Microsoft almost certainly will. And the rise of digital ownership means it needs to be a mandatory inclusion. "Keep your old console" isn't good enough anymore, people want just one box under the TV.
 

Shin

Banned
The things described in there even partially fit for the Pro. And there is also nothing you wouldn't expect from a BC point of view, I'm surprised they filed a patent for that.

Except this was filed recently so that rules the Pro out, there's more than one road that leads to Rome.
Perhaps it's their way to get there without patent infringement.
 
Those patents are extensions of patents first filed January 2016, actually, so yeah, they could relate to tech inside the Pro.

EDIT: He's got a few extra patents on BC, by the way, all of them filed in 2015:

- BACKWARD COMPATIBILITY THROUGH USE OF SPOOF CLOCK AND FINE GRAIN FREQUENCY CONTROL
- BACKWARD COMPATIBILITY BY ALGORITHM MATCHING, DISABLING FEATURES, OR THROTTLING PERFORMANCE
- BACKWARD COMPATIBILITY BY RESTRICTION OF HARDWARE RESOURCES
- BACKWARD COMPATIBILITY TESTING OF SOFTWARE IN A MODE THAT DISRUPTS TIMING

They're all Pro-related, IMO.
 

Taelus

Member
Wouldn't it be easier for Sony to keep same basic architecture as the PS4 but give it more powerful components? That way it could be fully backwards compatible the same way PCs are, without needing extra tech. It doesn't make sense to reinvent the wheel when consoles are basically just PCs at this point.
 

c0de

Member
Wouldn't it be easier for Sony to keep same basic architecture as the PS4 but give it more powerful components? That way it could be fully backwards compatible the same way PCs are, without needing extra tech. It doesn't make sense to reinvent the wheel when consoles are basically just PCs at this point.

Except the way they are used is not like PCs. Especially with Sony's SDK, developers code of the very specifics to the target hardware (or better: can do so). This includes CPU features *and* their specifics in regards to timing and behavior. This includes also specifics to RAM and buses, like the patents mention.
A PC is almost always BC because games are also build to target a wide range of different systems and APIs take care of the specifics.
 

Elandyll

Banned
Filing Date Jan 22, 2016
But if you look at the content of the patents, it doesn't seem to match the PS4/ PS4 Pro situation.

It specifically calls for a different processor, when the PS4 Pro has the same Jaguar CPU, just running at 2.1Ghz instead of 1.6.
That does not require "backwards compatibility".

It could be fluff... Or it could be a core design element of the PS5 system which they certainly were already working on in 2016 (specially if they indeed release in late 2019).
 

onQ123

Member
The provisionals of these were filed January 2016, so these almost certainly have to do with vanilla PS4 games in Pro. Some people seem to be reading too much into these.

I thought the same thing with the bus patent but the CPUID patent seem to be about a CPU that's different from the one that's in the console it's emulating.


17. The computing device of claim 16, wherein the information regarding the different processor identifies certain features of the processor on the computing device as being different from ones that are actually supported by the computing device.
 

BDGAME

Member
In 2019 the gap between pc and consoles will be so large that a new generation will be necessary?

And one thing is a middle gen refresh for the new resolution (from 1080 to 4k),other thing is a true next generation machine.

A Playstation 5 needs to have a perceptive difference between it and the Xbox one X.

If necessary, they can delay it to 2020 or longer. But create a true "next generation" machine.
 

Shin

Banned
If necessary, they can delay it to 2020 or longer. But create a true "next generation" machine.

”I've said this is very very early days and a very new medium and it's going to grow slowly and organically over time. This is VR 1.0. It's going to be a long road."
https://www.usatoday.com/story/tech...erience-coming-sony-playstation-vr/102433756/

PSVR 2.0 might be a thing by then, PS5 needs to be strong enough in order to offer a big jump for VR 2.0 also.
 

Hexa

Member
That's a provisional for the current application. Provisionals have application numbers stating with 62. Provisional applications are applications that have the same subject matter as the actual application, but it's usually in a messy state without fulfilling the full requirements of a proper application. They are filed so as to obtain earlier priority dates. So the subject matter in that application was invented prior to Jan 2016.
 

c0de

Member
But if you look at the content of the patents, it doesn't seem to match the PS4/ PS4 Pro situation.

It specifically calls for a different processor, when the PS4 Pro has the same Jaguar CPU, just running at 2.1Ghz instead of 1.6.
That does not require "backwards compatibility".

It could be fluff... Or it could be a core design element of the PS5 system which they certainly were already working on in 2016 (specially if they indeed release in late 2019).

The file itself is not specific at all and, at least to me, describes nothing unusual if you want to build a system with bc capability. And the patent around the bus speed and behavior is quite that what the Pro offers, not to mention that the CPU of the Pro could actually have a different CPUID (which is, in my opinion, something that is easily controllable from AMD anyways).
 
It specifically calls for a different processor, when the PS4 Pro has the same Jaguar CPU, just running at 2.1Ghz instead of 1.6.

I thought the same thing with the bus patent but the CPUID patent seem to be about a CPU that's different from the one that's in the console it's emulating.

We don't actually know if the CPU within the Pro is exactly the same as the base PS4. It's the same arch, but it might have included some changes, like an extended instruction set, for instance. It's possible it included Puma or Puma+ features.
 
The file itself is not specific at all and, at least to me, describes nothing unusual if you want to build a system with bc capability. And the patent around the bus speed and behavior is quite that what the Pro offers, not to mention that the CPU of the Pro could actually have a different CPUID (which is, in my opinion, something that is easily controllable from AMD anyways).
Maybe PS4 Pro has a Puma ("Tiger") CPU, but what about the following?

http://appft.uspto.gov/netacgi/nph-..."&OS="cerny;+mark+evan"&RS="cerny;+mark+evan"

In another example, a more powerful APU may contain a L3 cache for the CPU, compared to a less powerful APU that did not have such a cache. In such a case, the memory latency characteristics differ as the time needed to access data that misses all caches increases for the more powerful APU, but average latency will decrease for the more powerful APU.

Pretty sure they're describing a Ryzen APU for the PS5. Jaguar/Puma have no L3 cache.

Also, who says that PS4 Pro isn't "beta testing" for the inevitable PS5? What they're already doing in regards to BC may prove to be useful for the PS5 as well.
 

onQ123

Member
We don't actually know if the CPU within the Pro is exactly the same as the base PS4. It's the same arch, but it might have included some changes, like an extended instruction set, for instance. It's possible it included Puma or Puma+ features.

The documents said it was the same Jaguar cores from the PS4


450_1000.jpg
 

c0de

Member

What about it?
Pretty sure they're describing a Ryzen APU for the PS5. Jaguar/Puma have no L3 cache.

Or any other CPU. What I was trying to say is that they are not specific here and they don't have to be. This is not what a patent is for.

Also, who says that PS4 Pro isn't "beta testing" for the inevitable PS5? What they're already doing in regards to BC may prove to be useful for the PS5 as well.

Of course no one. What I argued since the beginning of the Pro is that, although the Pro is largely the same as the amateur, Sony hesitated to give customers the boost mode.
To enable something like the non-boost mode on very different hardware (at least from the CPU point of view and the memory subsystem with the garlic/onion buses), you can't "just" downclock - the CPU is different in many ways and not just the CPUID.
The patents are a general way of describing what is, from an abstract point of view, needed to accomplish bc on different hardware. How that will turn out in the end, we will see.
 
But the PS4 also had 1TB HDD
The original (2013) model had a 500GB HDD.

Maybe I'm reading too much into it, but the fact that they're talking about "Jaguar" (quotes) cores might indicate a Jaguar evolved uArch (aka Puma).

What about it?
I was talking about the L3 cache quote. Pretty sure this isn't a patent about the PS4 Pro specifically.

Or any other CPU. What I was trying to say is that they are not specific here and they don't have to be. This is not what a patent is for.
What other CPU? We're talking about AMD and more specifically their APU roadmap.

Only Ryzen has an L3 cache in that context, so it must be relevant to the PS5 (or a prototype form of it at least).

Of course no one. What I argued since the beginning of the Pro is that, although the Pro is largely the same as the amateur, Sony hesitated to give customers the boost mode.
To enable something like the non-boost mode on very different hardware (at least from the CPU point of view and the memory subsystem with the garlic/onion buses), you can't "just" downclock - the CPU is different in many ways and not just the CPUID.
The patents are a general way of describing what is, from an abstract point of view, needed to accomplish bc on different hardware. How that will turn out in the end, we will see.
I understand the technical hurdles, but Sony execs are going to crucify Cerny if he doesn't deliver PS4 BC for the PS5, especially when the competition makes it clear that BC is a given. He must find a way, no matter how hard it is, if he wants to keep his job at least. That's what all these patents are for, IMHO.
 

c0de

Member
Good joke, most patents are vague as hell and doesn't tell you anything that's just how it is.
The L3 cache tapantaola pointed out is rather interesting.

It is indeed interesting as currently both CPU package caches can't snoop each other, making data shared between the two extremely expensive.
As the patent state, they might even do it on a software level, managing program counters in software, as well as timing critical data transfers behave like it was before.
That the next PlayStation will offer something Ryzen equivalent is a given, though.
The only questions left is which incarnation and considering bc, if they are able to solve the problems via software, hardware or both - or not at all.
 
Maybe PS4 Pro has a Puma ("Tiger") CPU, but what about the following?

I'm pretty sure that's just simple core optimizations required to "port" the design onto a new manufacturing process. Puma is the same, in that Puma cores are in effect Jaguar cores optimized to be manufactured on a smaller process node.

In terms of the micro-architecture, running actual code etc, Puma/Jaguar/Tiger are more or less identical.

If we're talking about BC, then running 28nm Jaguar code on a 16/14nm Jaguar, isn't backwards compatibility. The code will run natively.

In which case, I don't think these patents relate (solely) to the PS4 Pro.
 
I'm pretty sure that's just simple core optimizations required to "port" the design onto a new manufacturing process. Puma is the same, in that Puma cores are in effect Jaguar cores optimized to be manufactured on a smaller process node.

In terms of the micro-architecture, running actual code etc, Puma/Jaguar/Tiger are more or less identical.

If we're talking about BC, then running 28nm Jaguar code on a 16/14nm Jaguar, isn't backwards compatibility. The code will run natively.

In which case, I don't think these patents relate (solely) to the PS4 Pro.
Yeah, the Bobcat -> Jaguar transition had some uarch changes. Puma(+) is basically a FinFET optimized Jaguar.
 
Top Bottom