XBox 360 scalability

[Carl B]

I would disagree with most of the sentiments expressed in this thread. While the trend in hardware will certainly be towards more parallelism, the architecture favored will most likely be symmetric cores like XeCPU. The reason is pretty simple: most general purpose CPUs (e.g. Intel, AMD, IBM Power line) are moving in that direction, and so most likely console CPUs will as well. For starters, there's going to be a lot more research in this area: how you build the microarchitecture, memory systems to keep them properly fed, efficient use of die space and transistors, etc. On top of that, most academic research around parallelizing algorithms target symmetric cores, and will continue to do so in the future.

The CELL is a very interesting idea, but it is an anamoly and it will not take off outside of specialized areas. It would probably not be getting much traction if not for the fact that Sony is pushing it in the PS3. The market can tolerate it now because it is in the middle of the transition away from instruction level parallelism towards thread level parallelism, and everyone is in the midst of adapting (and hence there's no entrenched mindset around that yet). But in the long run, symmetric cores will be the winner. Remember also that PC gaming, while shrinking in size, has massive mindshare among developers. Most console devs started there and still do, and thus symmetric cores will be the path of least resistance and smallest learning curve when transitioning or porting to a console.

I think Cell in PS4 - assuming again no paradigm shift occurs - makes a lot of sense if only because developers will already be familiar with the architecture, and it doesn't behoove Sony to move them away from that if they feel performance-wise it is still viable. Personally, I think there is still going to be a lot to be gained on the CPU side of things as non-graphics tasks get their importance upped this gen. The preceding comments also haven't been focused on whether or not an XeCPU-like architecture is viable in the long run (if you're talking symmetric cores), but rather if the XeCPU *itself* is viable in the long run; and I simply believe five years from now there will probably be a better option available to Microsoft than an XeCPU with nine cores at 6 GHz.

The 'larger picture' aside however, I think the future of Cell's vision involves an abstraction as to the number of cores/SPE's available for work, with the tool and programming models heavily favoring breaking code down to put on the SPE's. Anything that can go on them, will - even if they are not the best suited architecture to handle it. Five/six years down the line, regardless of what new architecture Sony could go with - I think it would truly have to be something special in order to warrant turning the development community on it's head, for exotic though the 1:8 model might be, it will be a model with which a great many game developers will be familiar with at that point.

This is also not taking into account that a cottage base of programmers might gravitate towards Cell for the defense and imaging fields in the same way that Itanium/Epic - though niche - has it's adherents as well.

 

[expletive]

I think Cell in PS4 - assuming again no paradigm shift occurs - makes a lot of sense if only because developers will already be familiar with the architecture, and it doesn't behoove Sony to move them away from that if they feel performance-wise it is still viable. Personally, I think there is still going to be a lot to be gained on the CPU side of things as non-graphics tasks get their importance upped this gen. The preceding comments also haven't been focused on whether or not an XeCPU-like architecture is viable in the long run (if you're talking symmetric cores), but rather if the XeCPU *itself* is viable in the long run; and I simply believe five years from now there will probably be a better option available to Microsoft than an XeCPU with nine cores at 6 GHz.

The 'larger picture' aside however, I think the future of Cell's vision involves an abstraction as to the number of cores/SPE's available for work, with the tool and programming models heavily favoring breaking code down to put on the SPE's. Anything that can go on them, will - even if they are not the best suited architecture to handle it. Five/six years down the line, regardless of what new architecture Sony could go with - I think it would truly have to be something special in order to warrant turning the development community on it's head, for exotic though the 1:8 model might be, it will be a model with which a great many game developers will be familiar with at that point.

This is also not taking into account that a cottage base of programmers might gravitate towards Cell for the defense and imaging fields in the same way that Itanium/Epic - though niche - has it's adherents as well.

I wonder if the upcoming XNA studio will favor one processor architecture over another. I'm not 100% sure exactly what XNA brings to the table, does it provide tools for optimizing game code and performance?

Seeing as the idea is for XNA studio to bring PC and console development closer together (allowing devs to realize synergies and reduced costs per 'installed unit' (i.e. consoles and PC gamers)), you would think that any tools would be optimized for the types of CPUs in PCs and their console. If its the case it seems likely the Xbox CPUs will continue to closely mimic the architecture and evolution of PC CPUs, that way the XNA development tools can optimize performance and multithreading more easily.

How does this impact developers? I cant help but feel that if MS realizes its 'dream' with XNA, developers will flock to the development platform becauase of how widespread it would be.

 

[Carl B]

I wonder if the upcoming XNA studio will favor one processor architecture over another. I'm not 100% sure exactly what XNA brings to the table, does it provide tools for optimizing game code and wither performance tools?

Seeing as the idea is for XNA studio to bring PC and console development closer together (allowing devs to realize synergies and reduced costs per 'installed unit' (i.e. consoles and PC gamers)), you would think that any tools would be optimized for the types of CPUs in PCs and their console. If its the case it seems likely the Xbox CPUs will continue to closely mimic the architecture of PC CPUs, that way the XNA development tools can optimize performance and multithreading more easily.

How does this impact developers? I cant help but feel that if MS realizes its 'dream' with XNA, developers will flock to the development platform becauase of how widespread it would be.

I really don't see XeCPU as mimicing PC architecture - I mean IOE alone changes things up - but I agree XNA probably plays a strategic role down the line. I mean I've always thought that in an ideal world for Microsoft, PC and 360/720/1080 development would become near-seemless, but I guess I just don't see how that bridge is crossed between x86 and IOE Power - to say nothing of not knowing what a next-gen XBox would be based on. But still, perhaps it's not beyond the pale to envision some Microsoft tool down the line that would near perfectly recompile game code for one architecture to code for the other.

 

[expletive]

I really don't see XeCPU as mimicing PC architecture - I mean IOE alone changes things up - but I agree XNA probably plays a strategic role down the line. I mean I've always thought that in an ideal world for Microsoft, PC and 360/720/1080 development would become near-seemless, but I guess I just don't see how that bridge is crossed between x86 and IOE Power - to say nothing of not knowing what a next-gen XBox would be based on. But still, perhaps it's not beyond the pale to envision some Microsoft tool down the line that would near perfectly recompile game code for one architecture to code for the other.

In terms of development what do you think is more 'different' IO vs OO or symmetrical vs non-symmetrical?

I wonder which of those two differences is more easily made portable with development tools and compilers?

 

[Carl B]

In terms of development what do you think is more 'different' IO vs OO or symmetrical vs non-symmetrical?

I'd say IO vs OO is more 'different.' But of course I guess it depends partially on how bizzarre a non-symetrical setup would end up being. Cell I don't see as being so outlandish, but I could certainly envision some strange CPU's in my mind.

I wonder which of those two differences is more easily made portable with development tools and compilers?

That's a good question.

 

[Sethamin]

Sethamin, I would have to disagree that Cell is an anomaly, firstly its an old idea from way back when ramping up the clock speeds wasn't an option (bit like now ), secondly ideas like it ferature heavily in Intels 2015 plan.

Sure, it is an old idea - they're called DSPs. Granted, they're on-chip, full speed DSPs with their own local storage and huge pipe to memory - DSPs on crack, if you will - but the essential idea behind it is a DSP. But the DSP model is not the future IMO.

Do you have any links to the Intel stuff? I'm not sure exactly what you're referring to here.

Cell or a cell like system offers what people ideally need, you have a core set of "general" purpose processors, essentially your main code, stuff that cannot be threaded safely. Then you have your SPE type addons that give massive but limited functionality power. Cell processors running in groups (4, 8., 16...) would work well, the gains in SPEs are useful if you can fill them (say rendering?) however the PPEs also add more general power at the same time. It seems like a nicely adaptable solution to a problem, you need n cores then fit n 1-8 cells, you want more general power? fit n 2-4 cells, The problem becomes the interconnection and communication however cell style allows for expansion of the general power while providing more specialised hardware as well.

When scaling a system you can often find more small tasks that would suit SPEs or at least utilise more of them, yes there probably is a finite limit however as long as you keep expanding the tasks that are required of a computer the number of tasks will keep scaling.

What exactly makes this a system "what people ideally need"? Or at least more so than the general symmetric core system? Let's think about this in the abstract - the idea behind the CELL is to create a 2-tier like system where there are a few more general cores (PPEs) and a larger number of more specialized cores (SPEs). The fact that there are 2 different types of cores is what makes it asymmetric. But it's also what makes it harder to program. To write effective algortihms on the SPEs, you have to figure out how many cores you need, how to coordinate between them and main memory (where most of your program state lives), and also how they can coordinate between each other. All of this is a result of not having unified memory access between the cores. Any developer will tell you this is a pain.

I think Cell in PS4 - assuming again no paradigm shift occurs - makes a lot of sense if only because developers will already be familiar with the architecture, and it doesn't behoove Sony to move them away from that if they feel performance-wise it is still viable. Personally, I think there is still going to be a lot to be gained on the CPU side of things as non-graphics tasks get their importance upped this gen. The preceding comments also haven't been focused on whether or not an XeCPU-like architecture is viable in the long run (if you're talking symmetric cores), but rather if the XeCPU *itself* is viable in the long run; and I simply believe five years from now there will probably be a better option available to Microsoft than an XeCPU with nine cores at 6 GHz.

I agree with you here. I would find it highly unlikely for Sony to abandon the CELL architecture, assuming they can scale all of it up to an acceptable level 5-6 years from now, since they've invested so heavily in it up to the this point. I also think the XeCPU is not likely to be used in the 360's successor, but a XeCPU like architecture will be (multiple symmetric cores). The point I was trying to make above is that Sony is throwing their weight behind the wrong type of CPU design (asymmetric cores). There is no other big player out there pushing this. They're going to be the only ones going this route, and as such, they are going to have to throw massive amounts of R&D $$$ at it to get the design to scale in the long run.

Symmetric core architectures, OTOH, are going to remain the focus of the industry and the target of innovative techniques for years to come. All the multi-core work the Intel and AMD are doing now means that 5-6 years from now it will be very well understood how to build cost effective multi-core chip that make very good use of die space with lots of power. In other words, a symmetric core approach is going to ride the wave of industry, while Sony's approach is going to go it alone. I would expect nothing less from them, of course, but I think it's a losing battle in the long run.

This is also not taking into account that a cottage base of programmers might gravitate towards Cell for the defense and imaging fields in the same way that Itanium/Epic - though niche - has it's adherents as well.

I find this highly highly unlikely. As I mentioned before, software development tools are more important than hardware. Unless the gains to be made were absolutely spectacular and completely eclipsed anything using commercial off the shelf parts, there is no way the defense industry is ever going to standardize on this. They're far better off string together farms of PC to do this work and using standard development tools. As with most software projects, the costs of employing the programmers far outweighs the cost of the hardware.

 

[Titanio]

Just a side note..

Tim Sweeney has suggested for a while now, and suggests again in his most recent presentation, that in next-gen consoles, the CPU and GPU will be unified (that seems to make sense, with GPUs becoming ever more programmable, and the emergence of CPUs like Cell). Cell seems to be a little bit of a nod in that direction, and in fact, we could have had Cell-based CPUs and GPUs in PS3 even. The idea may have been premature for now, but it could pay off for them later..

 

[expletive]

Just a side note..

Tim Sweeney has suggested for a while now, and suggests again in his most recent presentation, that in next-gen consoles, the CPU and GPU will be unified (that seems to make sense, with GPUs becoming ever more programmable, and the emergence of CPUs like Cell). Cell seems to be a little bit of a nod in that direction, and in fact, we could have had Cell-based CPUs and GPUs in PS3 even. The idea may have been premature for now, but it could pay off for them later..

Do we think a unified CPU/GPU would look more like Cell than Xenos?

 

[Carl B]

I find this highly highly unlikely. As I mentioned before, software development tools are more important than hardware. Unless the gains to be made were absolutely spectacular and completely eclipsed anything using commercial off the shelf parts, there is no way the defense industry is ever going to standardize on this. They're far better off string together farms of PC to do this work and using standard development tools. As with most software projects, the costs of employing the programmers far outweighs the cost of the hardware.

Cell isn't being left alone on the software tools side, so I think for the moment there's something to be said for the ability of the architecture to hang on in a sense in areas in which it is strong. Certainly refer to my use of Itanium as the analogy chip to keep in scope what levels I'm talking about here. As for the defense industry, well you have Mercury Systems already marketing a system to DoD and Raytheon apparently cosnidering a complete line of sensor equipment based around the chip. I'd say for it's relative age thus far, that is indicative of *some* level of promise in terms of breaching defense. The Raytheon announcement is particularly intruiging, IMO.

 

[Megadrive1988]

I agree that Sony will continue to go into the asymmetric path with next-generation Cell for PS4. perhaps they will add a third type of core, in addition to a main PPE core and SPEs, thus deepening their commitment to the asymmetric path. I don't see just PPEs and SPEs in the next-gen CELL, there will probably be something new, while still keeping the overall Cell architecture and concept.

 

[Sethamin]

Cell isn't being left alone on the software tools side, so I think for the moment there's something to be said for the ability of the architecture to hang on in a sense in areas in which it is strong. Certainly refer to my use of Itanium as the analogy chip to keep in scope what levels I'm talking about here. As for the defense industry, well you have Mercury Systems already marketing a system to DoD and Raytheon apparently cosnidering a complete line of sensor equipment based around the chip. I'd say for it's relative age thus far, that is indicative of *some* level of promise in terms of breaching defense. The Raytheon announcement is particularly intruiging, IMO.

What software development tools are you referring to? Outside of some very game specific middleware tools for the PS3, I am not aware of any tools made for CELL whatsoever.

Also, your Itanium example is not a very heartening one. The Itanium is, by all accounts, a massive flop. Given the amount of R&D $$$ it took to get there and the amount it pulls in, there is almost no way Intel (or HP) will ever even make the investment back, never mind turn a profit on it. But it does seem to be a prudent example; the reasons it failed were that it required new development tools (particularly compilers) that still haven't matured, and that its cost to performance metrics simply couldn't beat the x86 ISA chips. These are roughly the same reasons that the CELL will not take off outside of the PS3 - the development tools aren't there, and it won't be able to beat mainstream CPUs on price/performance in the long run.

 

[Guden Oden]

These are roughly the same reasons that the CELL will not take off outside of the PS3

One could argue it's already in the process of taking off. Who can say where it'll lead, or what niches it will carve its way into. If you persist in being such a staunch sceptic, I'll just have to remind you of that 19th century patent office director who so famously claimed everything inventable had already been invented. You seem to follow a similar line of reasoning here.

the development tools aren't there, and it won't be able to beat mainstream CPUs on price/performance in the long run.

Well, so you say. Forgive me if I do not take your personal opinions as undisputable fact. How privvy exactly are you as to what may or may not be available/under development as far as cell software tools anyway?

Anyway, this is console tech forum so cell use outside PS3 (or possibly 4, and so on) is actually beyond the scope of what we're "supposed" to discuss here. In any case, the argument you're making that because the PC world is going symmetric means that's the way things are going to go period feels fundamentally flawed. The PC world is a different universe compared to consoles, and the needs and requirements are different. PCs live and die by their x86 backwards compatibility and the massive amount of apps already created which have to run swiftly and efficiently, sometimes with little genuine regard to costs. Consoles don't have to play by those rules, and in fact can't play by them. That's why next-gen consoles are multicore instead of one large and complex OoOE processor like in today's PCs. You simply can't reach high enough performance levels using PC methods of reasoning for the price consoles are sold at.

Thus, if the main processor is asymmetric or not doesn't really matter, because consoles aren't PCs. They're a market unto themselves.

 

[Sethamin]

One could argue it's already in the process of taking off. Who can say where it'll lead, or what niches it will carve its way into. If you persist in being such a staunch sceptic, I'll just have to remind you of that 19th century patent office director who so famously claimed everything inventable had already been invented. You seem to follow a similar line of reasoning here.

One could argue that, but I don't think it'd be a very convincing argument or else I wouldn't have taken the position that I have. By all means try, though - that is the whole point of discussion.

Well, so you say. Forgive me if I do not take your personal opinions as undisputable fact. How privvy exactly are you as to what may or may not be available/under development as far as cell software tools anyway?

Don't get so defensive. It goes without saying that this are my personal opinions and not fact. If you want to dispute them, I welcome the debate. That is how intelligent discussion happens: someone takes a position and backs it up with evidence, someone disputes it and presents other evidence, etc. If we were only able to post hard facts on this board, I imagine it'd be pretty empty.

Anyway, this is console tech forum so cell use outside PS3 (or possibly 4, and so on) is actually beyond the scope of what we're "supposed" to discuss here.

Sure, but it's my opinion that the general direction of CPU design is extremely interrelated with how the CPUs of future consoles will look. I don't think that's a radical leap of logic. Besides, I thought we were having a good discussion here without being too off topic.

In any case, the argument you're making that because the PC world is going symmetric means that's the way things are going to go period feels fundamentally flawed. The PC world is a different universe compared to consoles, and the needs and requirements are different. PCs live and die by their x86 backwards compatibility and the massive amount of apps already created which have to run swiftly and efficiently, sometimes with little genuine regard to costs. Consoles don't have to play by those rules, and in fact can't play by them. That's why next-gen consoles are multicore instead of one large and complex OoOE processor like in today's PCs.

I disagree wholeheartedly on several points. What significantly distinguishes PC CPUs from console CPUs is definitely not backwards compatibility. The portion of a modern x86 CPU devoted to backwards compatibility is actually pretty neglible; they've been able to do amazing things with that unwiedly ISA by converting everything to simple RISC ops behind the scenes. It's not multi-core either - all PC CPUs are moving that direction as well. Just look at the newest offerings from AMD and Intel, and you'll see dual core CPUs with quad cores on the horizon. They are hitting the exact same barriers to speed increases in monolithic CPUs that consoles are. The biggest difference is OoO execution versus in-order, which does add quite a bit of real estate to PC CPUs. But I think that's mostly irrelevant to my main argument.

Anyway, the point is that it's not just PC CPUs were talking about here. It's CPUs as a whole. Everyone is moving to symmetric multi-core designs, from IBM's Power line and Sun's Ultrasparc IV to the Itanium down to AMD and Intel's offering. And my main argument is that the CELL is the odd man out, and will not benefit from the massive research that will undoubtedly go into this area in the future. That's all.

You simply can't reach high enough performance levels using PC methods of reasoning for the price consoles are sold at.

Also, as an aside, this is not true. Xbox1 had a PC CPU, and was absolutely fine as far as CPU price and performance went (the HDD, OTOH,...). Modern PC CPUs are arguably the absolute kings in terms of price/general performance. There's a reason that Google's server farms (and others as well) are built exclusively out of these.

 

[scooby_dooby]

well Intel is actually going teh same approach, from a very high level, general purpose cores with sourrounded by 'auxilary' cores to help multimedia applciations. I think CELL is like 10 years to early though, multiple symetric cores have lots of power left for now.

"The latest news on this front: at Intel, they are no longer talking about the "Multicore" () processors the way it is done relative to 2х, 4х, 8x, 16х or even 32х core solutions, but about the Many-Core () processor architecture which implies an entirely new architectural macrostructure of the chip comparable (but not similar) to the architecture of the Cell processor.

The structure of such a Many-Core chip implies operation with the same instruction set but using a powerful central core or a number of powerful CPUs "surrounded" by multiple auxiliary cores, which will help effectively processing complex multimedia applications in the multithreaded mode. Besides the "general purpose" cores, Intel processors will offer specialized cores for executing tasks of various classes – graphics, speech recognition algorithms, processing communication protocols."

http://www.digital-daily.com/cpu/intel-roadmap/index02.htm

 

[Carl B]

What software development tools are you referring to? Outside of some very game specific middleware tools for the PS3, I am not aware of any tools made for CELL whatsoever.

There are such tools in progress, and some of them are available through the IBM website. Research at IBM is also ongoing and Cell for it's part enjoys at least some aspects of an 'opened' architecture to help encourage grass-roots development.

Also, your Itanium example is not a very heartening one. The Itanium is, by all accounts, a massive flop. Given the amount of R&D $$$ it took to get there and the amount it pulls in, there is almost no way Intel (or HP) will ever even make the investment back, never mind turn a profit on it. But it does seem to be a prudent example; the reasons it failed were that it required new development tools (particularly compilers) that still haven't matured, and that its cost to performance metrics simply couldn't beat the x86 ISA chips. These are roughly the same reasons that the CELL will not take off outside of the PS3 - the development tools aren't there, and it won't be able to beat mainstream CPUs on price/performance in the long run.

I used the Itanium example specifically because Itanium is such a failure, I doubted anyone would take issue with my claiming Cell could reach those levels of penetration. But Cell differs from Itanium in several important respects; it will always be a high volume - relatively low-cost chip (looking at the larger picture), in contrast to Itanium, and it's billions in investment will almost assuredly be recouped - especially since in actual R&D, the investment was *only* ~$500 million, the billions for Cell going to fab capacity with utility beyond Cell should it fail.

I think any sustainable traction outside of the console space - and not saying Cell's there yet by any stretch - but any sustainable traction in the realm of compute intensive tasks will be qualifiable as a win for Cell.

Afterall, if Itanium was profitable for Intel, I'd consider it a win for them as well. It's just that it's not.