Predict: The Next Generation Console Tech

Status
Not open for further replies.
They can run an updated autonomous security kernel in a sealed SPU ! D^8
(That means they'd need a 7 core SPURS, even less likely haha)

One of the rumors mentioned that the PS4 audio DSP can decode about 200 concurrent MP3 streams. I can't find an equivalent measurement for the SPUs, except that Toshiba showed a full Cell decoding 48 MPEG2 1080p streams.

If Sony CTO's last interview holds any weight *at all*, I am very curious about the "configurable logic" he mentioned, and how they intend to tackle the security issues given that they were thinking of some form of PS4 Linux (again).
The "next-gen audio" article on Gamasutra said 400 MP3 streams in real-time using MultiStream.

A great example of the power of the Cell processor is that MultiStream can process 50 * 2 second convolution reverbs on one SPU in realtime. MultiStream can also decode approximately 400 MP3's on a single SPU in realtime." -- Jason Page, SCEE (Sony Computer Entertainment Europe)

http://www.gamasutra.com/view/feature/132208/nextgen_audio_squareoff_.php?print=1
 
For PS4's 200 concurrent MP3 streams rumor... assuming it's true, may be it's based on FPGA ?
http://www.bdti.com/MyBDTI/pubs/20061101_gspx06_fpgas.pdf

That slide argues that traditional DSP metrics can't be used to measure the performance of equivalent FPGA implementations effectively. As a result, Sony may choose the final outcome (e.g., # of concurrent streams) to communicate/advertise its performance ? It would avoid those dumb/smart FLOP slippery slope discussions. And they can reconfigure it after delivery to improve the implementation, and support new codecs ?

An FPGA implementation would essentially skip/save the FLOP count ? and explain why audio DSP is listed separately in the rumored feature list too.

Otherwise, I have no clue why the Sony CTO mentioned "reconfigurable logic" in his interview.

EDIT: Unfortunately it doesn't tell us anything about the SIMD mentioned in Eurogamer's article.

EDIT 2: For Cell, so far, I think the most compelling argument for PS4 SPU is the security framework. Security is one of those realms where cheaper, faster and better may not work for the management. Everyone in Sony up to Hirai and Stringer has been burnt before. The SPU security framework is more or less proven. They may be able to fix the holes and improve it further. None of the completely new implementation can claim this. Then again, it assumes that the Sony security team can't come up with another proven and better mechanism. Even if they follow a Cell-like approach, it will likely be a stripped down, custom implementation.

Honestly I would find the typical FPGA not much less uninspiring than the straight DSP scenario. What in the world applications are there where we are looking at future codec performance as something that is even in question in the context of the existant primary silicon? For thermal considerations in video/audio applications, certainly. And for the aforementioned audio assist in gaming and other applications. But iIf 200 streams is simply an impressive number to advertise performance, we need to have a sense as to what that horsepower can be leveraged towards, because codecs alone won't really benefit gaming except in a very limited capacity.
 
Talking about the cores for OS, AndyH (NeoGAF):

This information is the same as what I was told early this year. 8 cores @ 1.6ghz with 2 dedicated to the OS. You have to keep in mind that this is 8 cores @ 1.6ghz as opposed to 3 @ 1.2ghz.

Wouldn't want any heat issues and high failure rates this generation.

6 cores will be free for games.
 
Orbis has no such ambitions and may power past the new Xbox simply because it focuses its resources on out-and-out games power.

Or maybe there will are two SKUs, one of them "weaker" (focused on apps) and another one stronger, same base architecture, different power:

Microsoft is building an Xbox set-top box. Multiple sources familiar with Redmond's plans have confirmed to The Verge that the company plans to introduce a low-cost alternative to its Xbox console, designed to provide access to core entertainment services. The move will allow Microsoft to further increase its presence in the living room, providing consumers with a choice between a set-top box or a full next-generation Xbox console.

We're told that the set-top box is part of a two-SKU strategy for Microsoft's next-generation of Xbox hardware that will be unveiled in 2013, with a release date ahead of the holiday shopping season. The device will run on the core components of Windows 8 and support casual gaming titles rather than full Xbox games typically found on a dedicated console. Although hardware specifications aren't fully locked down, we understand Microsoft will use a chipset to enable an "always on" device that boots quickly and resumes to provide near-instant access to TV and entertainment services.


Microsoft's Xbox set-top box work is said to be part of a broader effort to ensure its core architecture for the next-generation Xbox is scalable enough to be put together to run on a number of devices. We understand that the company could opt to combine its core system for the next Xbox with a phone stack to deliver a phone capable of running a full version of Microsoft's Xbox Live services. It has also investigated providing this functionality to TV OEMs, who could include the core services as part of a licensed Xbox television set.
 
The "next-gen audio" article on Gamasutra said 400 MP3 streams in real-time using MultiStream.



http://www.gamasutra.com/view/feature/132208/nextgen_audio_squareoff_.php?print=1

Nice !

Honestly I would find the typical FPGA not much less uninspiring than the straight DSP scenario. What in the world applications are there where we are looking at future codec performance as something that is even in question in the context of the existant primary silicon? For thermal considerations in video/audio applications, certainly. And for the aforementioned audio assist in gaming and other applications. But iIf 200 streams is simply an impressive number to advertise performance, we need to have a sense as to what that horsepower can be leveraged towards, because codecs alone won't really benefit gaming except in a very limited capacity.

The slide that caught my attention is #20 on page 10. Seems to offer very high performance per cost ratio compared to DSP. Not sure how an SPU would measure up.

I'm still interested to find out whether these h/w units are in, and if there're really extra compute units in Orbis.
 
Gotta be at least one, for security or something.

It is likely to use an ARM Cortex-A5 (or possibly A7) core for hypervisor security, as AMD plans to integrate an A5 core into its Jaguar and Steamroller-based consumer APUs for that purpose. It would be wasteful to dedicate one of the main cores for that task.
 
Last edited by a moderator:
I wish we knew when MS was coming out with their specs for the "Real" Xbox 720. They are hiding things pretty well. Not even Eurogamer has any idea of the specs of what MS has on their console.

It's possible some people have old information that is outdated. Some people might have old API's for another xbox version. Some people might have alpha dev kit specs.

That is why I think 1.25 Single Floating point TF is very basic for 2013/2014. There has to be more to this.
 
Last question , why would MS insist on using Win 8 for the Xbox ?

Why wouldn't they? They're not going to bring Windows CE back or license an OS or use FreeBSD or something.
Even Windows Phone is now a Windows 8 variant. They can use any Windows 8 components needed, see Windows XP Embedded and Server Core ; use another UI on top, and even then keep common APIs that build the interface (typesetting, compositing etc.).

It won't be Windows 8 in the sense that it doesn't ship with notepad, a print spooler and explorer.exe, but it surely will be Windows 8 otherwise.

Orbis will reserve some resources for OS too, PS3 reserves 1 SPU for the OS.

but PS3 doesn't reserve a CPU core for the OS (it only has one core). I'm not too convinced : if the PS3 has a quad core Bulldozer, reserving one full core seems too much. It's a scheduler thing anyway, you could have it set up so games can have up to 366% CPU usage, leaving 33% guaranteed for the OS (with affinity to one particular core if so be it)
 
but PS3 doesn't reserve a CPU core for the OS (it only has one core). I'm not too convinced : if the PS3 has a quad core Bulldozer, reserving one full core seems too much. It's a scheduler thing anyway, you could have it set up so games can have up to 366% CPU usage, leaving 33% guaranteed for the OS (with affinity to one particular core if so be it)

PS3 reserved an SPU to run sensitive security services. I remember the PPU also run some OS calls where needed.

Check out ARM's TrustZone in my link above. AMD licensed it. Don't know if Sony will use the same thing.
 
I'm glad these APU + GPU rumours have (hopefully) been put to rest all together. I guess the less tech savvy didn't understand that they had to just use what was available in the early dev kits, and that the only reason we don't see larger APU's on PC is that the current supporting socket, memory system and demand just aren't there yet. As the advantages of HSA become apparent and stacked memory a reality, I think we will see APU's similar to those that will feature in these next consoles on PC. It will still be a while before discreet GPU's become less necessary, but that is the future I see PC moving towards.

I don't know enough about these things, but I would hope that having PC games designed around APU's could eliminate this draw call bottleneck? Anyhow, it's nice to see people accepting that Sony don't need to use a completely off the shelf APU + GPU as the only possible option any more. The whole "A10 + 7660" thing getting spouted all over the web was getting annoying. The MS followers seemed to come to terms with this quite a while ago.

APU + GPU is what AMD recommends explicitly for core gaming. We will see APU + GPU setups in future desktop PCs, but I doubt we will see very big APUs in desktop PCs.

Whether you use a single APU or a combination of APU and GPU depends on a couple of things: Target computing power, target wattage, target die size, etc. The single APU will have the better performance bandwith- and latency-wise, but since it's limited by die size and yields it will have inferior performance than an APU + GPU combo when it comes to raw computing power. The latter is what most Gaming-PCs are and will be aiming at due to the thick abstraction layer that is DirectX. Consoles use very thin abstraction layers, if at all, and therefor it is a smart move to target bandwith and latency instead of raw (paper)FLOPS.

At the end of the day Leadbetter is describing a pretty potent next gen console, that is basically nothing else than APU + GPU, but integrated in a single die (which is undoubtably superior to a 2.5D stacked package of APU + GPU with the same computing power). As the APU part we have 8 Jaguar Cores in combination with the "special sauce", which are most likely customized GCN elements that will focus solely on GPGPU algorithms.This combination could be quite powerful compared to conventional homgeneous desktop processors. On the GPU side we have the 18 GCN compute units with 800Mhz frequency, which is in my eyes the maximum of graphics power that is reasonable for a next gen system.

But we also have the 4GiB of GDDR5 and I'm a little disappointed with that and to be honest I still see a little chance that GDDR5 will be used for the latest devkits only. GDDR5 is expensive, hot and yaws for power. WideIO-RAM is superior in almost every case and they used it for the Vita already. Additionally, WideIO-RAM is probably going to be cheaper than GDDR5 in the long run, bearing the long life cycles of gaming consoles in mind. Call me a fool but I still think we will see the PlayStation SoC stacked on an interposer with WideIO-RAM. This technology is just perfect for gaming consoles in almost every respect. It's expensive in design but economic in production. RAM-wise I think there's more to come. The opera ain't over until the fat lady sings.

But all in all I'm pretty excited for the new System. In my eyes the new PlayStation has a massive architectural advantage over a modular constructed PC with homogeneous processors, much more than PS360 had over PCs a couple of years ago. This will compensate for the missing raw computing power. I'm wondering when desktop PCs will be able to catch up with the degree of heterogeneity that is achieved within this SoC...
 
Has somebody speak with Bkillian of late? I feel weird looking at his attitude throughout the years that it would get out of his way and confirm anything (even in an elusive manner) as in the quote in that Eurogamer article (I hope they ask for his agreement...), I feel like he could be trolling the web and laughting at the bar with his (previous) workmates... :LOL:

EDIT 1 WRT to BKillian statement migh t be my English but the wording definitely excites my sarcasm detector.
Yeah, Leadbetter did not ask before quoting my post. Does not make me happy, especially since it was a pretty vague post.
 
Eurogamer Article on Orbis

8 Core Jaguar
18 CU's @ 800 MHz
+ Some special Compute Unit
4 GB GDDR5

http://www.eurogamer.net/articles/df-hardware-orbis-unmasked-what-to-expect-from-next-gen-console

I do hope the SPE theory I floated a while ago comes to pass.

6 SPEs would be pretty small. AMD's fusion designs are intended to be modular and allow for specialized execution units to be added. In addition to providing a path for backwards compatibility, they could have many uses for PS4 software.

Things I imagine a SPE array could be used for in Orbis:

real-time audio rendering
security
media decoding/encoding
MLAA
physics
image analysis/object recognition for Move

Sony wouldn't even need to ask developers to program them directly. They can have ICE create a number of OS services and APIs for the different SPE accelerated function so all this stuff is transparent to the programmers. You'd just have a SPE time budget and you can use as much or as little as you want.

Obviously things like audio could be done with fewer transistors in a specialized DSP, but I think there's a lot of upside to a flexible SPE design. You get value from backward compatibility, enhanced security, and a plethora additional services and function, plus the ability to think up more if you need them down the line.
 
Yeah, Leadbetter did not ask before quoting my post. Does not make me happy, especially since it was a pretty vague post.

That's not the first time they've pulled quotes from B3D. I suppose they assume it's ok since it's an anonymous forum, however sooner or later someone connects the dots and that leads to trouble. That's pretty much the main reason I don't post here anything of substance anymore.
 
Cause in the Grand scheme of things usable RAM is also detimined by bandwidth aka you could have a 1TB of ram but if only 1GB is usable per frame it's nothing more then a giant space for cache in which the transfer rates and seek times are much faster then a HDD. Bandwidth determins the amount of usable RAM at any given time within reason. aka 1TB/s of bandwidth for 10MB of ram will give you more usable RAM then any of these configs but can you generate enough but can you get the data to the RAM fast enough to use anywhere near that in unique data?

True but from Sebbi's posts, I mean you dont exactly NEED all your RAM every frame.

And most games will be 30 FPS next gen again, so you can access 2GB per frame with durango.

I can argue the Orbis is over engineered. It can access over 6GB per frame at 30 FPS, yet it's total RAM is only 3.5GB. Now what? Only thing you can do is go back and hit the hard disk or worse, blu ray disk and load more in. Good luck with that.

192 gb/s might be good on pc, where my game might need to go up to anywhere to 200 fps and my resolution up to 4k, but is it good for a console?

In other words over say 3 frames, Durango might be able to do more. It can use 2GB/2GB/1GB of unique data. Orbis will be like 3.5GB/uh-oh/uh-oh

RAM speed is not a substitute for quantity. Both are important.

It actually sounds to me like the PS4 might be really good at 60 FPS games. I mean literally the majority of games might be 60 on it cause it's got so damn much bandwidth. But, they probably wont look much better than 30 FPS Durango counterparts.

For the article, the main thing to come out of it is Orbis using 8 Jag cores supposedly. It was generally thought it would switch to Jag cores by me, but that it's 8 of them and all..

It just removes one more point of differentiation to Durango. However, vs 4 Steamroller cores, it's likely a downgrade realistically.

Also interesting DF says its an SOC and even has a code name. In that case I guess it's true both should be doing SOC's for cost reasons. I was skeptical of SOC's in either, especially PS4, but it's looking more and more likely.

At the end of the day anyway, you may get most multiplatform games optimized for 1.2 teraflops and 3.5GB RAM. Lowest common denominator.
 
totally agree ...

again this is all speculation, and im only going off the several patents ive read from MS-AMD-IBM ...

Every one of them that discusses BTE/VTE and raytracing.. all require
1. coherent memory cache
2. low-latency high-bandwidth communications network..

I know these are very general goals but some of the rumors ive heard about the design of the 3 SOCs as well as things like blitters, large caches, and 3d vertically stacking ic for the oban, syncs well with the requirements..

I know all rumors... but interesting none the less

I've been lurking in this thread for a while mostly due to conversing about it on GAF and to lesser extent TXB. In my opinion the notion of a magical chip that will be doing raytracing at 720p/1080p is just crazy talk. I think some people need to definitely reel in their expectations a bit. But that's just me.
 
Status
Not open for further replies.
Back
Top