Predict: The Next Generation Console Tech

[pjbliverpool]

#1- Why would you want to run Windows on a console? To get viruses?
#2, they already have that one locked down.
#3-ditto
#4 ? It would seem to mean lesser performance.

#1 - I agree. Consoles are unsuited to running Windows or a similar general purpose OS and I wouldn't want to see it happen. I was just referring to the previous "benefit" that was mentioned about the ability to act as a closed PC.
#2 - Not completely. the 360 and PS3 still share the same PPC heart. A switch to X86 will solidify Sony's position as the outsider.
#3 - Again, not completely. Current X86 chips are both easier to programme for and more familar to developers in general than PPC - or at least any new PPC that would go into a console.
#4 - I don't get this? What PPC option that could go into consoles do you see as having higher performance than a modern X86 quad?

Even somebody on another forum used this logic "well PS3 uses a piece of crap lowly 7800GT, and it seems to do just fine" as justification for using a low end part in next gen. Dont they get that what they're asking though would be the equivalent of Ps3 using an IGP from 2006?

I see you're point but it's not completely equivilent. Fusion chips offer far higher performance compared to todays mainstream GPU's than IGP's did back then
.

 

[archangelmorph]

Sorry I don't buy any of this justification of potential gains of an x86 based ISA...

It's bollocks...

[& Given that I work directly with a in-house cross-platform game engine codebase I think I'm fairly qualified enough to say this...]

CPU instruction set actually makes no tangible/material difference towards the ease of development on a closed platform like a games console. ISA visibility alone is esoteric at best & in a production environment you'll rarely find yourself in a position where you need to touch or be concerned with it. It's generally hidden away by the compilers & with ever fewer instances where one needs to hand tune CPU code, even when they do occur you'll be vectorizing and parallizing waaay before ever having to re-write anything in inline assembly for example.

As long as the ISA provides sufficient SIMD functionality nothing else matters & in that vein there's very little difference between writing code for an x86 platform or a PPC one (99.9999999999999999% of your CPU code will be C/C++ & interpreted script such as Lua, with the biggest outliers being your optimised fastmath & vectormath libraries, which these days are already provided by hardware vendors in a suitably interface-abstracted form...)

A switch by MS from PPC to x86 would likely only be done to facilitate some kind of "write once, run anywhere" software solution for the platform however I question the legitimacy of this given the silicon sacrifices one would need to make to accommodate such an over-bloated & unnecessarily legacy-laden instruction set. You'd basically be throwing away something like 10%-20% of your potential CPU performance on that decision alone I presume which seriously calls into question the importance/legitimacy of such an idea...

There's just no potential benefits when all of the conceptual gains one might consider from such a decision can and already have been achieved in many ways (e.g. cross-platform, hardware-abstracted codebases; pretty much the gold standard these days for most developers, Apple's universal binaries etc...)

 

[Barbarian]

Sorry I don't buy any of this justification of potential gains of an x86 based ISA...

There doesn't have to be any benefit outside of price/performance. And your entire point is correct if you assume IBM/PPC can compete with AMD/Intel + embedded GPU.
I personally don't think they can.

 

[hoho]

And your entire point is correct if you assume IBM/PPC can compete with AMD/Intel + embedded GPU.
I personally don't think they can.

Compete at absolute maximum performance (per-thread or entire chip), performance per watt or performance per transistor? In the last metric I'm quite sure a custom PPC like in this gen consoles can easily compete and likely even beat current x86 solutions. Current highest end CPUs have nearly 10x the transistor count of Cell/Xenon

 

[archangelmorph]

There doesn't have to be any benefit outside of price/performance. And your entire point is correct if you assume IBM/PPC can compete with AMD/Intel + embedded GPU.
I personally don't think they can.

Why do you assume an embedded GPU would be in any way a pre-requisite?

Also what exactly is it you think an AMD/Intel chip offers [in terms of general performance relating to games] that an IBM/PPC can't?

 

[ERP]

I actually doubt any PPC processor could compete cost/performance wise with Intel or AMD's current/future x86 architectures.
For general code we measured dramatic performance differences between our dev PC's and the MIPs processors in box X360 and PS3 before they were released, let alone today.
IBM's focus is just in a different place.

The ISA is irrelevant, and the wasted gates for legacy really are negligible at this point, outside large streamed workloads (most of which seem to be moving to the GPU), it's really all about how often you hit the cache and not very much else.

Having said that it doesn't really matter the nature of consoles is you deal with what you have. You won't be throwing away tons of CPU time in the display driver to start with.

 

[homerdog]

Hey ERP, could you give us an idea of the performance gap between modern CPUs and Xenos? I recall hearing that Xenos is about on the level of a 2GHz Athlon X2, but this could be completely off for all I know.

It's easier to compare GPU power between consoles and modern PC GPUs, but since the CPU designs are so dissimilar I haven't really been able to draw a good comparison.

 

[Tahir2]

Imagine* a console from Sony with dual outputs, with a robust OS that has some basic productivity applications, a media player and a browser.

Hook the console up to your big TV and a monitor - kids play on console, you do your tax return.

I bet that would hurt Microsoft.

* Completely far fetched parallel universe type of imagining unfortunately.

 

[ban25]

Sorry I don't buy any of this justification of potential gains of an x86 based ISA...

It's bollocks...

[& Given that I work directly with a in-house cross-platform game engine codebase I think I'm fairly qualified enough to say this...]

I'm sorry, but IBM's boutique micro architectures are simply no comparison for the modern, out-of-order devices produced by AMD and Intel. In fact, IBM hasn't really been competitive since the mid-'90s. Why do you think Apple dumped PPC? If you've actually run code on the PPU, you know that thing is slow, and the Xenon is only a little better, relatively speaking.

Next time around, I'm fairly certain that IBM will only be making it into the Wii U.

 

[Xenus]

I kind of doubt it. IBM has 2 major advantages when they design custom solutions they end up being pretty damn competitive for their era and unlike AMD and Intel they simply give better deals because AMD and especially Intel simply don't need the consoles to sell all the processors they can make.

 

[archangelmorph]

I actually doubt any PPC processor could compete cost/performance wise with Intel or AMD's current/future x86 architectures.
For general code we measured dramatic performance differences between our dev PC's and the MIPs processors in box X360 and PS3 before they were released, let alone today.
IBM's focus is just in a different place.

The ISA is irrelevant, and the wasted gates for legacy really are negligible at this point, outside large streamed workloads (most of which seem to be moving to the GPU), it's really all about how often you hit the cache and not very much else.

Having said that it doesn't really matter the nature of consoles is you deal with what you have. You won't be throwing away tons of CPU time in the display driver to start with.

I think the most important reasons IBM's PPC-based offerings will continue to provide the edge in terms of serving as a more ideal console CPU offering are:-

- An architectural focus on multi-threaded versus single-threaded performance (since this seems to be the way in which high performance computing is going & especially in the games space) as in practice, you'll pretty much always find greater gains through parallelisation vs trying to do more in a [faster] single thread's execution pipeline
- A focus on smaller, cooler & more streamlined chips; better for cost & cost-reduction as you're only going to get so many die shrinks over the lifetime of your hardware platform & so the less silicon you have to start with, the easier it'll be to enable things like die-sharing of CPU+GPU later on down the line for example

Fundamentally I just don't think the size-to-performance [multi-threaded] ratio of a core i7 for example vs the CELL is really going to be enough to justify the cost difference, both for the IP+manufacturing as well as additional costs relating to possibly extra cooling, loss of future revenue due to reduced cost-reduction strategies etc...
Then there's the fact that the CELL's computational throughput would put it either not too far away from & in some cases even ahead of the intel-based offering & it doesn't look too good for an intel/AMD based solution IMO...

 

[AmFreak]

Some people don't seem to get that Xenon and Cell are the way they are because of Microsoft and Sony and not because they use PPC ISA. It's not that the ISA restricts OOE or anything. Look up PPC970 or Power7 for example. They both could have gotten out of order designs but they chose not to. And if u look at what was in the pc market when the 360 came out i would say their choices weren't bad. Intel's highest end processor at the end of 2005 was the Pentium D and i doubt anyone wanted that abomination in a console. AMD had the Athlon X2 @ 2,4 Ghz (with a TDP of 110 W). The Athlon might have a higher performance, than the Xenon, in some fields but has nowhere the same SIMD capabilities. I think i don't need to compare these cpu's with the Cell cause everybody reading here, by now, should know that the Cell is helping the RSX in ways an Athlon X2 or a Pentium D could never have done.

 

[ERP]

People overrate the importance of SIMD.
It's a big number that applies to a very small percentage of the actual code a game runs.
I've always said that the 360/PS3 CPU designs had more to do with big numbers and marketing than what game developers actually do/need.

 

[archangelmorph]

People overrate the importance of SIMD.
It's a big number that applies to a very small percentage of the actual code a game runs.
I've always said that the 360/PS3 CPU designs had more to do with big numbers and marketing than what game developers actually do/need.

Not so sure...

If your physics code is running on the CPU for example then thats a massive percentage right there...

& if you're writing SPU code for particle simulations, animation blending + transforms, visibility & occlusion tests etc I'd expect you to be making very good use of your SIMD units.

Otherwise "you're doing in wrong"...

 

[bkilian]

As far as I'm aware Windows doesn't run on PPC yet.

1. The ability to run Windows
2. Ease of cross platform development between PC and Xbox Next - edging out Sony
3. General ease of development thanks to the forgiving architecture itself and developers familiarity with it
4. Greater performance

So, um.. you do realize that the 360 is running Windows, right? Sure, it's a heavily modified, limited windows kernel, but it's still mostly compatible. Most of my tools will compile for either xbox or windows with a single makefile change.

Not so sure...

If your physics code is running on the CPU for example then thats a massive percentage right there...

& if you're writing SPU code for particle simulations, animation blending + transforms, visibility & occlusion tests etc I'd expect you to be making very good use of your SIMD units.

Otherwise "you're doing in wrong"...

Also any audio stuff. Especially if you don't have any purpose built audio hardware.

 

[Colourless]

A Win32 like API != Windows. XBox is missing a lot of the services and subsytems that make Windows. Even saying the XBox kernel is 'Windows' is dubious. The XBox devs have said its not based on the NT Kernel.

 

[Sonic]

Imagine* a console from Sony with dual outputs, with a robust OS that has some basic productivity applications, a media player and a browser.

Hook the console up to your big TV and a monitor - kids play on console, you do your tax return.

I bet that would hurt Microsoft.

* Completely far fetched parallel universe type of imagining unfortunately.

I doubt it. Especially considering Microsoft could just release Office for the 360 or next iteration of Xbox and that would negate any advantage Sony had up to then. With Office being the standard productivity software in the computing industry it would be right up MS's ally.

If Sony were able to make the software good enough vs Office 360 then for Sony console owners it would definitely be an option.

 

[Blazkowicz]

Some people don't seem to get that Xenon and Cell are the way they are because of Microsoft and Sony and not because they use PPC ISA. It's not that the ISA restricts OOE or anything. Look up PPC970 or Power7 for example. They both could have gotten out of order designs but they chose not to. And if u look at what was in the pc market when the 360 came out i would say their choices weren't bad. Intel's highest end processor at the end of 2005 was the Pentium D and i doubt anyone wanted that abomination in a console. AMD had the Athlon X2 @ 2,4 Ghz (with a TDP of 110 W). The Athlon might have a higher performance, than the Xenon, in some fields but has nowhere the same SIMD capabilities. I think i don't need to compare these cpu's with the Cell cause everybody reading here, by now, should know that the Cell is helping the RSX in ways an Athlon X2 or a Pentium D could never have done.

yes, I'm sure IBM can make a powerful CPU, just do a new Power PC 970 / G5 equivalent.

 

[Rangers]

Imagine* a console from Sony with dual outputs, with a robust OS that has some basic productivity applications, a media player and a browser.

Hook the console up to your big TV and a monitor - kids play on console, you do your tax return.

I bet that would hurt Microsoft.

* Completely far fetched parallel universe type of imagining unfortunately.

Not really, that type of thing never works out.

In Wal Mart they have a full fledged, modestly specced computer in a box, including a 18.5" widescreen LCD, for $368 (looking online this one is 299, but I think a similar thing is a bit more in store). They have netbooks for 249, sometimes on sale for 199.

I'm pretty sure almost everybody in USA own a computer or laptop at this point. With full fledged Windows (or OSX maybe). Why would they bother with a game console that's hooked to the TV, probably underspecced (512MB RAM in PS3), hot, possibly loud, does not have a operating system to rival Windows, etc etc etc...computing on the TV screen itself is not a common paradigm, so most people would prefer to hook it to a monitor for such duties, but, it's a game console that's supposed to go under TV...

Anyway wasn't that part of the point of linux on PS3? And that never took off, before they removed it.

 

[Shifty Geezer]

Not really, that type of thing never works out.

Actually, in the '80s and early '90s it was remarkably popular.