The Case for Arm in Nextgen Consoles?

[TheChefO]

ARM Cortex A8 vs IBM PowerPC 476fp

Just ran the numbers on Arm:

Current die size 6.7mm2 for a "high performance" 2GHz a9/a15 @ 40nm

At 28nm, in a 100mm2 CPU budget, one could fit nearly 60 arm cores (2per a9). Of course this is not taking L2 cache into consideration, but it is an interesting alternative to Cell x2 or Xenos x2...

http://www.cs.virginia.edu/~skadron/cs8535_s11/ARM_Cortex.pdf

So what would such a CPU deliver to future games that a 16SPE 2PPE Cell or a 6core Xenos couldn't? What would be the drawbacks?

________________________________________________

Edit: after the info. presented below by liolo, I figure it makes better sense to compare the a8 vs the 476fp.

Both are very similar in size, power draw and overall performance, but the IBM chip brings some interesting potential to the table such as 4 issue, OOOE, IBM's VMX units, and of course backwards compatibility with xbox360/ps3.

Can anyone make the case for why Sony/MS would dump IBM and switch to ARM?

 

[liolio]

Just ran the numbers on Arm:

Current die size 6.7mm2 for a "high performance" 2GHz a9/a15 @ 40nm

At 28nm, in a 100mm2 CPU budget, one could fit nearly 60 arm cores (2per a9). Of course this is not taking L2 cache into consideration, but it is an interesting alternative to Cell x2 or Xenos x2...

Biggest concern with such an approach is: aren't there still plenty of operations which are inherently single threaded?

http://www.cs.virginia.edu/~skadron/cs8535_s11/ARM_Cortex.pdf

So what would such a CPU deliver to future games that a 16SPE 2PPE or a 6core Xenos couldn't? What would be the drawbacks?

I don't believe that Xenon or PPU would be significantly bigger if you take away the potent SIMD units on those CPU the L@ the glue logic, etc. IBM has tiny power efficient PPC, according to their benchmark they beat ARM offering (not taking in account A15 if memory serves right).
Clearly going with ARM has its advantages mostly on the software side but that mean BC will be tougher to achieve /not doable. It also has down side the huge software environment would create quiet some intensive for some hackers to break the system.
In regard to the proposal of gluing plenty of cores together I don't know I'm not an engineer but I would not forget about the glue it-self you would need to invest in a lot of glue and wires to make the thing works. Overall I don't believe that many cores (at least CPU cores) is an option for any almost any kind of systems now.

Sometime ago in a thread about Sony business model I said that they are missing the point and that they should have a clear goal like "android everywhere" and aim at be leader in a few years. I overshoot by a lot, Lenovo and others are already getting there with TV that ship in China with Ice Cream Sandwich running. This will soon open casual gaming to many household, with casual games you can play on your phone (OS agnostic on top of it), tablet PC and soon (if not now) on your tv. I think that Sony is already lagging too much in this regard behind competition / I wonder if they still have the 'how to" to beat Samsung, Lenovo and the likes on that front. Clearly Sony has no choice but to be a bigger fish in a tiner pond but it's not at all as good as a situation as it sounds.

To go back to ARM I would say the advantage ARM vs other RISK ISA in the kind of transistor budget you consider would be marginal in performance, the BIG added value is imho the software environment.

 

[TheChefO]

I don't believe that Xenon or PPU would be significantly bigger if you take away the potent SIMD units on those CPU the L@ the glue logic, etc. IBM has tiny power efficient PPC, according to their benchmark they beat ARM offering (not taking in account A15 if memory serves right).
Clearly going with ARM has its advantages mostly on the software side but that mean BC will be tougher to achieve /not doable. It also has down side the huge software environment would create quiet some intensive for some hackers to break the system.
In regard oto the proposal of gluing plenty of cores together I don't know I'm not an engineer but I would not forget about the glue it-self you would need to invest in a lot of glue and wire to make the thing works. Overall I don't believe that many cores (at least CPU cores) is an option for any almost any kind of systems now.

Sometime ago in a thread about Sony business model I said that they are missing the point and that they should have a clear goal like "android everywhere" and aim at be leader in a few year. I overshoot by a lot, Lenovo and others are already getting there with TV that ship in China with Ice Cream Sandwich running. This will soon open casual gaming to many household, with casual games you can play on your phone (OS agnostic on top of it), tablet PC and soon (if not now) on your tv. I think that Sony is already lagging too much in this regard behind competition / I wonder if they still have the 'how to" to beat Samsung, Lenovo and the likes on that front. Clearly Sony has no choice but to be a bigger fish in a tiner pond but it's not at all as good as a situation as it sounds.

To go back to ARM I would say the advantage ARM vs other RISK ISA in the kind of transistor budget you consider would be marginal in performance, the BIG added value is imho the software environment.

There are interconnect issues to worry about along with L2 cache, but the concept of a many Arm core (50-60 A9/A15 cores) vs Cell x2 (or Xenos x2) is the bigger question in my mind. Assuming a ringbus will suffice and a suitable cache can be fit in the die budget, what advantage/disadvantage would such a CPU present vs Cell x2 (or Xenos x2)?

Access to software libraries that are ARM native is nice, but this can be achieved by simply adding one Arm core for very little cost to the Cell or Xenos.

I'm more interested in the argument of many smaller Arm cores vs Xenos/Cell x2.

 

[TheChefO]

IBM has tiny power efficient PPC, according to their benchmark they beat ARM offering

Can you post a link?

 

[liolio]

Can you post a link?

Honestly no

Spoiler

somebody linked a pdf here a while ago presenting IBM line of embedded but I failed repeatedly to found either by researching the forum of using google :| Research skillz failure...

May be somebody here will have save the link or prove to have have less sucky google skill

 

[TheChefO]

I found the following link:

https://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/D393643EC6B662E78525763200547AED/$file/476fp_wp_04_07_2011.pdf

Is this what you were talking about?

 

[TheChefO]

According to the links below, the PowerPC 476fp is actually very impressive.

2.71 DMIPS/MHz (IBM)
vs
2.5 DMIPS/MHz (ARM)

3.7mm2 per core @ 45nm (IBM)
vs
3.35mm2 per core @ 40nm (ARM)

1.2W per core @ 45nm (IBM)
vs
.85w per core @ 40nm (ARM)

The 476fp is also an OoOE with a 4 threads per core. And the side benefit of being compatible with the existing code base.



https://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/D393643EC6B662E78525763200547AED/$file/476fp_wp_04_07_2011.pdf

http://www.arm.com/products/processors/cortex-a/cortex-a9.php

Can anyone make the case for ARM?

 

[Prophecy2k]

In a comparison made against a possible evolution of the CELL i would ask one question:

Is any single core of either the IBM PowerPC 476fp or the ARM Cortex A9 more performant than a single SPU in the majority of game-related workloads?

If so then there is a case for the consideration of such a CPU design for a next-gen console.

If not then i don't believe that is a case for either at all.

Not all game workloads are parrallelisable across many cores, and many others would require substantial programming effort from the developers that could have been mitgated greatly by choosing a CPU chip with a moderate number of cores and good solid single threaded performance.

I also question the notion that there is even a software advantage in going with something like ARM. Is it simply because lots of touchscreen phones and tablets have oodles of software already written for them that could be easy portable to your console if you so choose to go with an ARM-based CPU? That same software that is designed predominantly for a touchscreen interface of phones and tablets...

And what software on these phones and tablets in particular would benefit a games console? Media apps? - of which the majority of the most important ones have been written for the PS3 and Xbox 360 already. So why not simply design a next-gen box that is fully BC and thus can retain alot of that app functionality right off the bat?

Or is it phone and tablet games that are seen as a benefit? The same games that are built, developed and designed for causal play on the go using a touchscreen interface? The same that when put up against current XBLA and PSN titles would simply go ignored by the prevailing majority of the userbase as they would struggle to match up to the quality level of those titles?

I fail to see what meaningful software benefit choosing an ARM CPU for your console would net you, when you could simply create a development platform for your console that is powerful, accessible and would allow those developers of software on phones and tablets to easily port dedicated-console versions of their games/apps to your console instead of simply slapping together quick hack-jobs that aren't even designed for it. I really don't think that it's a benefit that many would make out on these forums.

 

[tunafish]

According to the links below, the PowerPC 476fp is actually very impressive.

It is. However, a minor nitpick: 476FP is the hard core implementation, with a scalar FPU optimized for doubles. If it is used in a console, it will be a version of the 470S, which is the modular IP core, with the FPU swapped out for a VMX unit.

The 476fp is also an OoOE with a 4 threads per core.

I believe this is a terminology mixup. I think IBM calls it 4-way SMT because the design is meant to be synthesized in clusters of 4 cores that share a L2. Everything I have seen about the core design says that it's a single-threaded one.

Which makes sense, because looking at the design, I really don't think more threads would help it at all. The biggest bottleneck of the design is the single load/store unit, and with 4-wide issue and a nice large scheduler, that load/store unit is probably fully employed, even on a single thread.

Can anyone make the case for ARM?

IP Licensing costs per chip would likely be much lower with the ARM chips. However, considering the amount of IP they both have locked in the PPC ecosystem, I honestly think that ARM could give them the designs for free and they'd still go IBM.

Is any single core of -- the IBM PowerPC 476fp -- more performant than a single SPU in the majority of game-related workloads?

I'd say *hell yes*, but the issue is a bit complicated. The 470 series is pretty much the exact opposite of the Cell SPU. The latter is designed for maximum throughput in the best case, mostly ignoring latency, efficiency, and the worst case stalls. The 470 is designed for low-latency, high-efficiency operation, and making the most of the resources available.

On loads where the SPU rocks, it's better than a 470. On loads where the SPU doesn't rock, and let's be honest, there are a lot more of these, the 470 wins, often with a huge margin.

 

[TheChefO]

It is. However, a minor nitpick: 476FP is the hard core implementation, with a scalar FPU optimized for doubles. If it is used in a console, it will be a version of the 470S, which is the modular IP core, with the FPU swapped out for a VMX unit.

I saw that and figured they'd change it for single precision in a console implementation. Didn't realize that already existed in the 470S! Thanks for the info!

I believe this is a terminology mixup. I think IBM calls it 4-way SMT because the design is meant to be synthesized in clusters of 4 cores that share a L2. Everything I have seen about the core design says that it's a single-threaded one.

Ah. Yeah I wasn't sure if that was right.

IP Licensing costs per chip would likely be much lower with the ARM chips. However, considering the amount of IP they both have locked in the PPC ecosystem, I honestly think that ARM could give them the designs for free and they'd still go IBM.

Indeed. The fact that Both Sony and MS both own their CPU IP helps, but I think more importantly, they'd have to analyze the alternatives to see if they provide a performance or cost benefit over their existing PPC architecture.

I'd say *hell yes*, but the issue is a bit complicated. The 470 series is pretty much the exact opposite of the Cell SPU. The latter is designed for maximum throughput in the best case, mostly ignoring latency, efficiency, and the worst case stalls. The 470 is designed for low-latency, high-efficiency operation, and making the most of the resources available.

On loads where the SPU rocks, it's better than a 470. On loads where the SPU doesn't rock, and let's be honest, there are a lot more of these, the 470 wins, often with a huge margin.

Interesting.

What's your take on ARM Cortex a9 vs the above?

 

[TheChefO]

I fail to see what meaningful software benefit choosing an ARM CPU for your console would net you, when you could simply create a development platform for your console that is powerful, accessible and would allow those developers of software on phones and tablets to easily port dedicated-console versions of their games/apps to your console instead of simply slapping together quick hack-jobs that aren't even designed for it. I really don't think that it's a benefit that many would make out on these forums.

My thoughts for potential Arm advantages would be:

1) Extremely low standby power consumption for things like Wake-on-Kinect and background/idle downloading/updating.

2) Simultaneous/seamless OS interaction with having the OS running on an arm not interfering with system resources of the main app (game).

3) Cross compatible with win8 apps.


But aside from these functional perks, I'm looking for someone to make a performance/game case for ARM.

 

[Duck Dodgers]

But aside from these functional perks, I'm looking for someone to make a performance/game case for ARM.

Performance isn't going to be an issue as attention is drawn away from traditional consoles. For example, yearly refreshes, like this year, with quad-core tablets will make a Vita live by its Sony genes alone. This looks a little bleak to me in a expanding sea cheap and just as entertaining software.

http://www.industrygamers.com/news/mobile-gaming-no-reason-it-wont-replace-consoles-says-developer/

If I were a console maker I would be very wary of spending big on a next-gen project that doesn't utilise the Android/iOS mobile paradigm in a meaningful way. What's happening to the Vita and 3DS will happen to the living room boxes as well. It's entertainment convergence or bust and you won't need anywhere near 60 ARM cores to do it.

 

[TheChefO]

What's happening to the Vita and 3DS will happen to the living room boxes as well.

I strongly disagree.

Mobile gaming has always been a 2 part niche:

1) a cheap "toy" for the kids

2) an additional gaming medium to the home console for those on the go


I haven't seen a study that conclusively has shown where Portables have replaced consoles in any meaningful way. In fact, Console sales this gen have been stronger than last gen.

Now, portable sales may become even more niche with the advent of smart phones, but I don't see any sign that people are looking to replace their ps3/xb360 with an iPhone anytime soon.

Different markets different uses.


Having said that, I do agree that there will be better cross platform/device integration WITH the consoles, but make no mistake, the console will be the center of the action, not the other way around.

There are too many limitations with any portable device to truly substitute a console.

Kinect on an iphone/pad? Nope.

Wimote/move on an iphone/ipad? Nope.

Controller pad integrated in an iphone/ipad? Well that's pretty much a vita isn't it?

200w guts in an iphone/ipad? How many seconds runtime are we shooting for?

_______________________

If Sony/MS are jumping ship to ARM as the CENTER of the console design (not just a small arm core for doing background type tasks), then it would presumably have to show some advantage over similarly capable cores such as the 470S or ... the existing cores in ps3/xb360.

 

[Prophecy2k]

I'd say *hell yes*, but the issue is a bit complicated. The 470 series is pretty much the exact opposite of the Cell SPU. The latter is designed for maximum throughput in the best case, mostly ignoring latency, efficiency, and the worst case stalls. The 470 is designed for low-latency, high-efficiency operation, and making the most of the resources available.

On loads where the SPU rocks, it's better than a 470. On loads where the SPU doesn't rock, and let's be honest, there are a lot more of these, the 470 wins, often with a huge margin.

Very intruiging... although i believe it would beg the question, how many of these loads where SPUs don't rock and the 470 does, are applicable to gaming? Also bearing in mind that the SPUs don't exist on their own, but are also coprocessors to the PPE, which in a hypothetical CELL2 (since we are talking next gen) would be an improved PowerPC core(s). So for the whole range of differing workloads, would a bunch of 470 cores be able to outperform a CELL2 where the PPE cores would be handling those loads not suited to SPUs and vice versa?

On the converse (and with the possibility of Backward Compatibility in mind) where SPU rocks and the 470 lags, how wide is that margin? And would it inhibit any notion of BC of the PS3 in software? (I believe this question has been answered already in another thread).

I'm grateful for your input Tunafish, and would greatly appreciate more of your insight since i've been wondering for a while now about ARM core performance and how is compares to the current crop of console CPUs

 

[Prophecy2k]

My thoughts for potential Arm advantages would be:

1) Extremely low standby power consumption for things like Wake-on-Kinect and background/idle downloading/updating.

2) Simultaneous/seamless OS interaction with having the OS running on an arm not interfering with system resources of the main app (game).

3) Cross compatible with win8 apps.


But aside from these functional perks, I'm looking for someone to make a performance/game case for ARM.

My comments:

1) I'm not sure how important this would be for a home console that is connected to mains power, as although i see it as a valid consideration in the design of a console, i don't see it as a decisive factor in any platform holder determining which CPU to go with next-gen.

2) I don't understand this actaully. How is this any different to say the PS3 where XMB runs on a SPU with it's own preallocated RAM and thus doesn't interfere with the resources dedicated to the main game/app?

3) Again i'm not sure how this is important when these win8 apps will be written for win8 tablets/PCs and not for a game console whose primary input device is a controller and not keyboard & mouse/touchscreen (lets not forget that the whole point of the ARM version of Win8 is for MS to make its long overdue forray into the tablet OS market). Why not have a dedicated software development platform that allows for easy porting of these apps from the win8/ARM environment to your console if you really care about that? Not all of these apps will even be useful on a games console, since they would have been written for a portable tablet device in mind for use on the go.

I personally think that the perfomance case would be the only one that matters for both Sony and MS. Plus, considering Nintendo, who out of the three needs to concern itself the least with stuff like BC etc, and is known for cheaping out, went with an IBM PPC CPU, I think it's an indication of the direction Sony and possibly MS will also choose to go with their nextboxes.

 

[TheChefO]

Plus, considering Nintendo ...went with an IBM PPC CPU, I think it's an indication of the direction Sony and possibly MS will also choose to go with their nextboxes.

Blasphemer!

Power architecture couldn't possibly win a new HW design as it is too inefficient. Nintendo obviously didn't want to start from scratch as they are lazy and just went with IBM cause they are friends.

/sarcasm

:smile:

The functionality case I was making WRT ARM was with the intent of having an ARM core onboard, but not necessarily the focus of the CPU with a huge multicore array of ARM CPUs. Just a small ARM chip to run those background functional tasks could do the trick without needing the entire architecture of the console being built around ARM.

But this thread I'm looking for someone, anyone, to take up the battle for ARM, and throw the gauntlet down on PowerPC yelling "THIS IS SPARTA!"... or not. :smile:

 

[Prophecy2k]

I think there'd be quite a bit of inherent difficulty and cost implication for a board containing multiple CPU cores with differing ISAs.

I'm not all that sure it's as simple as just chucking a little ARM core on the end, when as i previously stated, the benefit of it for a stationary home console is rather questionable.

I'm intrigued by the notion of a fat cluster of ARM cores as a main CPU though, and how perfomant such a part could be.

 

[TheChefO]

I'm not all that sure it's as simple as just chucking a little ARM core on the end, when as i previously stated, the benefit of it for a stationary home console is rather questionable.

If Nintendo can afford and manage to put one in Wii, I'm sure Sony/MS can figure it out too...

 

[liolio]

I strongly disagree.

Mobile gaming has always been a 2 part niche:

1) a cheap "toy" for the kids

2) an additional gaming medium to the home console for those on the go


I haven't seen a study that conclusively has shown where Portables have replaced consoles in any meaningful way. In fact, Console sales this gen have been stronger than last gen.

Now, portable sales may become even more niche with the advent of smart phones, but I don't see any sign that people are looking to replace their ps3/xb360 with an iPhone anytime soon.

Different markets different uses.


Having said that, I do agree that there will be better cross platform/device integration WITH the consoles, but make no mistake, the console will be the center of the action, not the other way around.

There are too many limitations with any portable device to truly substitute a console.

Kinect on an iphone/pad? Nope.

Wimote/move on an iphone/ipad? Nope.

Controller pad integrated in an iphone/ipad? Well that's pretty much a vita isn't it?

200w guts in an iphone/ipad? How many seconds runtime are we shooting for?

_______________________

If Sony/MS are jumping ship to ARM as the CENTER of the console design (not just a small arm core for doing background type tasks), then it would presumably have to show some advantage over similarly capable cores such as the 470S or ... the existing cores in ps3/xb360.

I somehow agree with you and Duck, for gaming consoles rules but their marginal value is going down as TVs start to ship with Ice Cream Sandwich. A lot of the services console manufacturers included or planned to include to their systems are made irrelevant by the market evolutions.

Spoiler

for any important business being stuck to a 7 years or more business plan is madness especially now when we are in many regard at a corner point

 

[east of eastside]

IP Licensing costs per chip would likely be much lower with the ARM chips. However, considering the amount of IP they both have locked in the PPC ecosystem, I honestly think that ARM could give them the designs for free and they'd still go IBM.

Reusability of the Microsoft 720 ARM chip in other hardware i.e. tablets, desktop, laptops and/or set top boxes would be another huge advantage extending from this.

Microsoft could turn around and generate licensing revenue from licensing out the chip to other device makers.

PPC has no life in the consumer space beyond consoles. ARM the total opposite.