I posted this theory on the GCNGB over at IGN back on 2/11/2005 and admittedly it had a little more spin on it there but I just wanted to see why the Beyond3D crowd thinks of it 
NOTE: I do reference or look at things on the macro scale throughout the formulation of this theory. I do this admittedly out of self preservation at times but also to try to keep it simple as best I can. I know...I'm not very good at it, but at least I'm trying right?
Anyway...
Most of what I say next is a theory based upon rumors floating here and there. I cannot back up everything I say with a direct link to where I got it from because some of it is my own conjecture and some of it is stuff that I took notice of months ago but did not make an effort to keep track of. I will say this. Most anything I say can be dug up over at Rage3D, Beyond3D, Anandtech, Tom's Hardware, Hard Ocp, and a myriad of other technical sites that I frequent. If you don't take my word for it you can dig it up for yourself. I'll provide what I can if asked, but I'm not going to spend the next two weeks digging for stuff so if I can't find it fast...well that's just how the cookie crumbles.
pieces or rather rumors:
-Xbox2 CPU to be dual or triple cored that is compliant with the PowerPC ISA (instruction set architecture)
-IBM is also making Nintendo's CPU
-cores will be similar to the PPE ('newly' designed from the ground up PowerPC core that has full compatibility with the PowerPC ISA...this thing is not as capable as the beefier Power5 core or derived from it as once thought)
-cores should then be at least dual issue (like the Cell's PPE) or (perhaps 4-issue SMT pure spec on my part) (simultaneous multi threaded---means the core can handle more than one thread at a time)
Cell facts:
-Cell's PPE core is dual issue SMT at this moment
-Cell SPE's (synergistic processing engines) are not nearly as capable of handling branching statements as the PPE core. This is bad for allot of common software and things such as AI and physics on the large scale (you can handle smaller chunks of such tasks if you're crafty). However each SPE is like unto a mini computer unto itself within a network of other SPE's and a PPE where each SPE has it's own personal set of memory etc.
-there is one PPE and 8 SPE's in the prototype Cell and although Cell is customizable and these numbers can be played with by design it's most likely this is what will end up in the PS3 so I'll use this setup. (The scalability of the Cell architecture is the largest single caveat to my whole theory just so you know and to be fair)
Here's where the rubber meets the road. The Cell chip at this point is capable of handling 10 threads simultaneous with 2 of them being on the PPE and 8 others on the SPE's. The thread on the PPE most often will be the one that's telling the SPE's what to do or the one that's handling branching and other things the SPE's aren't good at. Make note of this and I'll explain why this is important later. The SPE's threads will mostly be things that are burdened with intense computation and things that have predicable code. (as a branch mis predication is quite costly for an SPE and SPE's are not good predictors from what I've gathered). The threads or rather 'APULETS' will also need to be things that can be run independently of each other. That is to say that one process is not dependent upon the results of another process or a single task that can be split up into chunks. Stuff with allot of branches or immediate inter-dependcy is going to knock the SPE's for a loop due it's lack of intelligent branch prediction and limited ability to unroll loops. I need to state that inter-dependency problems can cripple ANY architecture be it single core or highly parallelized. I am simply asserting that SPE's due to various reasons will have a more adversely affected if the simply sit idle while dependencies are dealt with. With a single core even though the problem is still there the processor is not sitting idle. Coding in the first place will be harder for multi-core as you have even more reason to stave off inter-dependencies in the and you must find ways to leverage all the cores in when they cannot be avoided. I am in line with most common thinking that it won't be as simple a task and will require a change in thinking with respect to coding.
Things like AI may be hard to implement on a SPE due to this and thus this where the PPE will probably do most of the heavy lifting even though you can get crafty and still get some of the AI task done on the SPE's I think. This is a conjecture at this point but this is a safe assertion given what we know at the point I think. (this belief is primarily driven by how costly branch mis predictions should be IIRC 18 cycles etc)
A task like AI in the macro-scale will be could be handled by the PPE in the Cell while in the micro-scale the PPE will farm out smaller parts of the overall task to the SPE's. Given how the SPE's seem to be number crunchers and AI is more of logic or rather task with allot of branches this although this seems plausible I cannot say to what end. The line between the macro and micro scale is something I am not apt enough to define.
A task like rendering is can be highly parallelized and is computationally intensive. That is to say rendering is a task that can be split up into stages where each SPE's can take some stage and pass it's result to another SPE working another another stage of the task and you can setup a pipeline or 'stream'...sort of like a factory's assembly line. This leads me to believe Cell will have a hand in rendering as it's just begging to get used to that end. In fact, I have reason to believe the Nvidia GPU is there to ease the devs learning curve with the Cell and to perform pixel processing alone as Cell are not especially appear equipped to that end in being general purpose processors and not simply GPUs.
If nothing else think like this. The PPE can handle larger more complex tasks but isn't the number crunching monster an SPE is. The SPE's are what give the Cell it's wicked raw power but...that power may be hard to get to in some instances as I think I've made clear by now.
The Xbox2 CPU will not have SPE's and this could be it's greatest weakness...or strength given the task.
The Xbox2's CPU is supposed to have 3 PowerPC cores and I suspect it is true that these cores are brand new too like the Cell's PPE is. That is to say they will support the PowerPC ISA but they have been built from the ground up and are not based on an existing core architecture explicitly but rather use what works but tries allot of new things.
What is important here are two things. These cores will be more capable than SPE's in that they will have intelligent branch predication (good for AI etc and general tasks allot of common PC software performs) among other things. The other thing is that these cores will at least be dual issue SMT and perhaps even 4 issue (again the 4 issue is pure spec on my part). So if the Xbox2's CPU has 3 cores it will definitely be able to handle 6 threads simultaneously and could handle up to 12 if each core is a 4 issue SMT capable.
Unlike on the Cell these cores can handle AI more independently on a per thread basis because there should be no need to farm out portions of the task. This is something to note.
Now given each core has been redesigned they most likely have the potential to clock to very high speeds like the Cell's PPE does (it has been confirmed that all cores clock at the same frequency PPE's and SPE's in the Cell...I...think). The Cell's PPE right now is in the 4GHz range and a schmoo chart supplied at the ISSC by the STI (Sony, Toshiba, IBM) consortium shows that it can be pushed fairly well into the 5GHz range though it gets really power hungry and hot when this is done so they'll need to find ways to curb this in the future no doubt. (going to a smaller fabrication process and whatnot)(I think it's been found that the schmoo had some errors but I've not yet heard that 5GHz was out of the question)
But the rumor has been that the Xbox2's CPU's are clocked at 3.5GHz and now 3.0GHz? (who's to say just yet?)
This makes sense if a core in the Xbox2's CPU are more complex and thus more capable than the Cell's PPE. It's not impossible for the Xbox2's CPU cores to have stronger VMX/Altivec (something Apple is all over) etc as the slower clocked Power5 shows is quite possible. If this is not the case and then the Xbox2's CPU should be able to operate at the same clock rate as the Cell can but I think it's fair to say MS wanted the cores to be a bit beefier in order to offset the Cell somewhat as best it could.
With this thinking I propose that the Xbox2's CPU is composed of newly designed cores similar to but beefier than the Cell's PPE, and that it's deeper capabilities are what bring it's clock rate back down into the 3GHz+ range. (more transistors = more heat and leakage issues = reduction in overall clock rate) This is not necessarily a disadvantage in all cases.
Heres's what the score board could look like:
-The Cell can handle 10 threads in the 4GHz range of operation range but 8 of those threads will be on SPE's that won't take kindly to things that aren't computationally intensive and require allot of branching operations
-The Xbox2's CPU can handle 6-12 threads in the 3GHz+ range of operation where all 8 of those threads can be general tasks that require allot of branching such as AI and to a lesser extent physics and allot of common software
-The Cell should still have an advantage in cases where raw computational power are needed and given how in the overall an Xbox2 core will not be the killer number cruncher an SPE in the Cell is...perhaps significantly...perhaps not so significantly but I predict it will be a fairly noticeable margin in any case.
Here's the big question...which one is more powerful?
If we're talking raw GFLOPS then I expect the Cell will still have the advantage as it is meant to be a computational beast and it still would retain a clock rate advantage too at that.
but when thinking about the usage of raw power...
Well this will depend entirely upon the situation. Even for gaming. With common computer software the Xbox2's CPU would seem to have the advantage to me because it's capabilities are more in tune with the tasks such software call for to be performed regularly. Most of us use software on our computers that wants something to be done when we specify or x happens. Most of us are not scientists that want our computers performing intense computations for the vast majority of time. Another aspect of the situation to look at is inter-dependency. If something can't be done until x happens or is computed then nothing else can happen until that condition is satisfied. The task at hand is dependent upon some other tasks execution here. Often inter-dependency situations are handled by branch and looping statements in code ignoring those which are totally reflective of temporal (time---it's not a simple decision based upon some regularly readily available data...it something that is more bound by how long it takes to produce the data in the first place) limitations. If the Xbox2's CPU is like I describe then it will be much better suited to handle inter-dependency on a per thread basis than the Cell's SPE are. Again a crude and simplistic analogy where the SPE's in the Cell are at a disadvantage the the Cell's PPE and maybe even more so to the PPE like cores in the Xbox2's CPU.(that's 2 threads on something that handles branches well vs. 6-12 threads on something that handles branches well)
To give an example that relates a little better to gaming the Xbox2's CPU should be able to handle AI and to a lesser extent Physics A GREAT deal better than the Cell can if I'm on track here. If AI takes up oh 20% and physics takes up another 20% percent of all the tasks an engine performs...this can be a significant advantage indeed over the Cell chip for the Xbox2 that could go a long way to evening up the score board. (just grabbed those numbers out of the air just so you know...it will be more dependent on the type of game it is)
Although the Xbox2 could have a part in rendering too I think the Cell chip is basically begging to be used towards that end. The Cell being the GFLOP monster that it is and highly parallel...when you think of GPUs it sure sounds like it borrowed allot of it's ideas from them (at least in concept). The Cell should be the better performer towards putting visuals on the screen. Geometry comes to mind among other things.
or will it?
...actually this will depend upon what Nvidia is doing for Sony with the PS3. If Nvidia delivers a GPU that is capable in the areas of both Vertex and Pixel processing with the Cell primarily adding to the vertex processing side of things the PS3 should have a CLEAR visuals advantage with respect to geometry on screen or thinks related the vertex operations. If the Nvidia GPU does only performs pixel processing alone and vertex shading etc is left to the Cell this creates the opportunity for a tie with the Xbox2 in the visuals department or possibly even a loss.
My reasoning for this is that the Xbox2 will be using the R500 GPU that will be a rendering beast unto itself and the fact the Xbox2 CPU can help in the visual department like the Cell can...just not to the same extent.
(when I say pixel processing...think lighting effects like HDR, water effects like per-pixel reflection, per pixel light interactions like normal mapping or virtual disp...think pixel shading which encompasses a wide range of effects and texturing)
I look at it sort of like this.
NOTE: I wish to make it clear that Cell can be used to whatever end a programmer can get it to including pixel processing like tasks which I have every faith will happen. I am speaking to Cell's easier and probably more utilized advantage.
Cell = extreme geometry engine = EGE
Xbox2 CPU = geometry engine = GE
Full Nvidida GPU = extreme geometry and pixel processing engine = EGEEPE
Specialized Nvidia GPU = extreme pixel processing engine or perhaps...times 2! EPE or EPEx2
R500 = extreme geometry and pixel processing engine = EGEEPE
case 1:
PS3 given full GPU by Nvidia: EGE + EGEEPE = EGEx2 + EPE
Xbox2: GE + EGEEPE = EGE * (1.1 to 1.5) + EPE
PS3 wins overall in the visuals department but primarily in the geometry department where a virtual tie in the pixel processing department is not out of the question.
case 2:
PS3's GPU is for pixel processing: EGE + EPE = EGEPE
Xbox2: GE + EGEEPE = EGE * (1.1 to 1.5) + EPE
Xbox2 wins overall in visuals but probably not to a great extent...including geometry. This is the least likely case I feel though.
case 3:
PS3 is given a pixel processing monster: EGE + EPEx2 = EGEEPEx(1.3 to 1.5)
Xbox2: GE + EGEEPE = EGEx(1.1 to 1.5) + EPE
In this case it could be a virtual tie with perhaps one or the other having a slight advantage in either vertex or pixel processing...but you'd be hard pressed to pick a winner or for that matter...to notice it...at...all.
I feel case 3 or case 1 are most likely. It's not difficult to think that Nvidia would supply it's next gen GPU for the PS3 so case 1 makes sense enough as it is. Case 3 will require you to go to beyond3D and look at one of their articles. There an Nvidia representative speaks about how it's approach to the unified shader model differs from ATI's. Long story short ATI is moving in the direction of having it's rendering pipelines being able to handle vertex or pixel processing on the fly i.e. this is a HW implementation. Nvidia is moving toward still retaining seperate pixel and vertex engines that are highly specialized to each task and using software to handle unified shader support. Nvidia claims it's still better to have two engines that are highly tuned rather than one that can do both but is less efficient on both fronts. The the jist I get from it anyway. What is important is that if Nvidia is approaching the future this way then it's not difficult to see that it could simply peel off it's "next gen" pixel processing engine/pipleline and supply it for the PS3 where ATI can't provide a specialized pixel processing part because R500 simply won't be designed that way. I'm not so sure Nvidia would take the stance it's taken thus far if PS3 support were not a factor in the first place. I have my suspicions they would be doing the same thing ATI is and furthermore that they will in the not too distant future...unless Cell's really start taking off in the PC market. Then it's a whole new ball game. It also stands to reason that if Nvidia is going to only supply something for pixel processing that it be twice as powerful as normal in using all pipelines to that end. I say that because Sony must realise that ATI is supplying Nintendo and MS with a full GPU that is FULLY optimized for rendering as that is it's purpose. To demand less than this from Nvidia leaves the door open to get out gunned or even flat out owned. Given how much money they've invested in the Cell and thus the PS3...it's hard for me to see them making such an obviously boneheaded mistake.
let forget specs for a moment and speak about ease of use:
It's no secret at this point that the Cell chip is going to be hard to work with as will any highly parallel architecture will be next round.
It is my thinking though that getting the potential out of the Cell is going to be a more difficult task than gettign the potential out of a CPU like unto the one I describe could be the Xbox2's. The Cell will require programmers to do allot of tasks in software that are done by the hardware on the Xbox2 CPU. One such thing is direct memory management of the SPE's memory. Another such thing is branch prediction in software if you really don't want your SPE's sitting idle all the time. Another larger problem for programmers may be trying to take advantage of the Cell for graphic rendering vs. rendering on a GPU they know and understand. This isn't going to be like how software rendering was handled before because SPE's are not like CPU cores programmers have contended with before. It is not unfair to say that the Cell processor may be using a fair amount of it's horsepower running software to handle capabilities it SPE's do not have in hardware for coders. How significant this is is unknown at this point but I think it is fair to note that the difficulty in programming the Cell may further diminish the lead it may have in real world performance over the Xbox2's CPU as I describe it.
The Xbox2's CPU should be something programmers will not have near as much difficulty in coming to grips with and given that it's cores will utilize memory virtualization via caches etc it's hidden potential should be more attainable. Also unlike the Cell SPE's a core in the Xbox2 CPU should spend more time on average towards actually completing a task than running code to replace missing hardware capabilities.
Thank you if you've read everything up to this point.
Here's where the Revolution comes in....
IBM is also working on the Rev's CPU just like they are for the Xbox2 and the Cell.
ATI is also working on the Rev's GPU just like they are for the Xbox2.
It's a simple and believable assertion. I believe Nintendo will get similar parts from both partners. I expect there will be some superficial differences here and there...and I hope that any scaling back on the spec sheet if done at all is kept to a minimum.
I do not believe the Rev will use 2 or Dual G5 processors as they are to big, to hot, to power hungry have more capability than it needed.
That's it. Pure and simple. I mean do you think Nintendo is going to have the same people working on their parts and watch as they don't get equal or better treatment from them? I don't...or at least I haven't until recently. It's been said countless times that Nintendo doesn't make under-powered systems and I will hold onto the hope that when the goodies are virtually right in front of them...they will go for the nookie.
Also, this would make the most sense for IBM. They have already dumped a ton of money into the Cell. It only makes sense that they re-use what they can from all that R&D. To design three "completely" unique CPU's simultaneously seems allot less plausible to me than simply being selective with what you lend with respect to what you've already done. I know MS would have paid the money to get any revelations IBM came across in making the Cell's PPE applied to their own core and I at least have the hope that Nintendo would do the same. I would also think that the PPE or rather the core that is compliant with the PowerPC ISA is an intellectual property etc that IBM would retain while the SPE's which seem to be born of the PS2's Emotion Engine would belong more so to Sony. (Toshiba basically brings it's manufacturing expertise with DRAM chips to the table...someone has to make that Rambus XDR and the interface to it...that would have most likely been Toshiba's role) What I'm saying is that IBM should have enough freedom to given Nintendo and MS cores for their CPU's that can contend with or even out gun the PPE in the Cell chip. What IBM can't do on it's own is sell Cell's to Nintendo and Microsoft. This may not be necessary in that the Big N and MS can get what they need from the Cell tech without actually using them. Rambus can license it's Yellowstone and Redwood tech to MS and Nintendo in turn as well I think.
Remember! The guts of the machine will have no bearing upon how the interface to it is setup. If Nintendo wants gyroscopic controls...vr goggles...or a life size dancing Rob the robot to come with the Rev...this will not prevent them from doing this. Personally, I like controllers, I don't want to hit bongos or wear a full body rumble-suit-power gloved thinga-ma-jigger but if that's what floats your boat Nintendo can still deliver no probs. Of course I reserve the right to latch onto anything truly Revolutionary as being better than bottled sunshine and will reject having ever been in doubt of anything if that should become the reality.
I admit that at first the PS3 was looking like the elephant in the room and I was having a hard time seeing any way for the competition to compete with it. Thanks to a mere suggestion by an IBM fellow in a discussion about the Cell and the next gen I began to re-think things. I think what I've theorized is plausible. If it's the actual reality then all is not lost for MS and Nintendo in the next round. I do think the PS3 will probably have the most impressive specs in terms of big numbers it can push. I am very optimistic and hopeful now that when it comes to putting specs to good use that the Xbox2 and the Revolution could be on par with the PS3. Perhaps even outrunning it.
In any case I think next gen things should be quite interesting and for once in a long time I feel...there is chance that we all can be pleasantly surprised at how things turn out.
With that I close and give a link to an article that suggests I may be right in at least guessing what comprises the Xbox2's CPU.
http://arstechnica.com/news.ars/post/20050309-4686.html
I've seen some speculate that a Physics Processing Unit will be the third core on the Xeon CPU.
I have my doubts. A PPU crosses to far into specialization...but since it's a console that's not so solid an argument. I'll go to some other places then. To me it would make sense to use it but place it elsewhere in the system. Putting a PPU on the CPU would make IBM's task that much harder etc. It would also eliminate the advantage of having another core AND and PPU unit. It could provide some significant visual advantages though such as cloth interactions, water that acts like fluid, more particles on screen, better explosions etc. This would be a nice advantage if no one else used a PPU. It would also mean you could not let PPU accelerated physics seep into game play for multi-platform games in the console arena. Yet with XNA and the possibility of PPUs in the PC realm you could still port an Xbox exclusive to the PC market.
It is my hope the all next gen consoles use a PPU and that PPUs themselves find a place in the gaming world.
...ok NOW I'm shutting up
NOTE: I do reference or look at things on the macro scale throughout the formulation of this theory. I do this admittedly out of self preservation at times but also to try to keep it simple as best I can. I know...I'm not very good at it, but at least I'm trying right?
Anyway...
Most of what I say next is a theory based upon rumors floating here and there. I cannot back up everything I say with a direct link to where I got it from because some of it is my own conjecture and some of it is stuff that I took notice of months ago but did not make an effort to keep track of. I will say this. Most anything I say can be dug up over at Rage3D, Beyond3D, Anandtech, Tom's Hardware, Hard Ocp, and a myriad of other technical sites that I frequent. If you don't take my word for it you can dig it up for yourself. I'll provide what I can if asked, but I'm not going to spend the next two weeks digging for stuff so if I can't find it fast...well that's just how the cookie crumbles.
pieces or rather rumors:
-Xbox2 CPU to be dual or triple cored that is compliant with the PowerPC ISA (instruction set architecture)
-IBM is also making Nintendo's CPU
-cores will be similar to the PPE ('newly' designed from the ground up PowerPC core that has full compatibility with the PowerPC ISA...this thing is not as capable as the beefier Power5 core or derived from it as once thought)
-cores should then be at least dual issue (like the Cell's PPE) or (perhaps 4-issue SMT pure spec on my part) (simultaneous multi threaded---means the core can handle more than one thread at a time)
Cell facts:
-Cell's PPE core is dual issue SMT at this moment
-Cell SPE's (synergistic processing engines) are not nearly as capable of handling branching statements as the PPE core. This is bad for allot of common software and things such as AI and physics on the large scale (you can handle smaller chunks of such tasks if you're crafty). However each SPE is like unto a mini computer unto itself within a network of other SPE's and a PPE where each SPE has it's own personal set of memory etc.
-there is one PPE and 8 SPE's in the prototype Cell and although Cell is customizable and these numbers can be played with by design it's most likely this is what will end up in the PS3 so I'll use this setup. (The scalability of the Cell architecture is the largest single caveat to my whole theory just so you know and to be fair)
Here's where the rubber meets the road. The Cell chip at this point is capable of handling 10 threads simultaneous with 2 of them being on the PPE and 8 others on the SPE's. The thread on the PPE most often will be the one that's telling the SPE's what to do or the one that's handling branching and other things the SPE's aren't good at. Make note of this and I'll explain why this is important later. The SPE's threads will mostly be things that are burdened with intense computation and things that have predicable code. (as a branch mis predication is quite costly for an SPE and SPE's are not good predictors from what I've gathered). The threads or rather 'APULETS' will also need to be things that can be run independently of each other. That is to say that one process is not dependent upon the results of another process or a single task that can be split up into chunks. Stuff with allot of branches or immediate inter-dependcy is going to knock the SPE's for a loop due it's lack of intelligent branch prediction and limited ability to unroll loops. I need to state that inter-dependency problems can cripple ANY architecture be it single core or highly parallelized. I am simply asserting that SPE's due to various reasons will have a more adversely affected if the simply sit idle while dependencies are dealt with. With a single core even though the problem is still there the processor is not sitting idle. Coding in the first place will be harder for multi-core as you have even more reason to stave off inter-dependencies in the and you must find ways to leverage all the cores in when they cannot be avoided. I am in line with most common thinking that it won't be as simple a task and will require a change in thinking with respect to coding.
Things like AI may be hard to implement on a SPE due to this and thus this where the PPE will probably do most of the heavy lifting even though you can get crafty and still get some of the AI task done on the SPE's I think. This is a conjecture at this point but this is a safe assertion given what we know at the point I think. (this belief is primarily driven by how costly branch mis predictions should be IIRC 18 cycles etc)
A task like AI in the macro-scale will be could be handled by the PPE in the Cell while in the micro-scale the PPE will farm out smaller parts of the overall task to the SPE's. Given how the SPE's seem to be number crunchers and AI is more of logic or rather task with allot of branches this although this seems plausible I cannot say to what end. The line between the macro and micro scale is something I am not apt enough to define.
A task like rendering is can be highly parallelized and is computationally intensive. That is to say rendering is a task that can be split up into stages where each SPE's can take some stage and pass it's result to another SPE working another another stage of the task and you can setup a pipeline or 'stream'...sort of like a factory's assembly line. This leads me to believe Cell will have a hand in rendering as it's just begging to get used to that end. In fact, I have reason to believe the Nvidia GPU is there to ease the devs learning curve with the Cell and to perform pixel processing alone as Cell are not especially appear equipped to that end in being general purpose processors and not simply GPUs.
If nothing else think like this. The PPE can handle larger more complex tasks but isn't the number crunching monster an SPE is. The SPE's are what give the Cell it's wicked raw power but...that power may be hard to get to in some instances as I think I've made clear by now.
The Xbox2 CPU will not have SPE's and this could be it's greatest weakness...or strength given the task.
The Xbox2's CPU is supposed to have 3 PowerPC cores and I suspect it is true that these cores are brand new too like the Cell's PPE is. That is to say they will support the PowerPC ISA but they have been built from the ground up and are not based on an existing core architecture explicitly but rather use what works but tries allot of new things.
What is important here are two things. These cores will be more capable than SPE's in that they will have intelligent branch predication (good for AI etc and general tasks allot of common PC software performs) among other things. The other thing is that these cores will at least be dual issue SMT and perhaps even 4 issue (again the 4 issue is pure spec on my part). So if the Xbox2's CPU has 3 cores it will definitely be able to handle 6 threads simultaneously and could handle up to 12 if each core is a 4 issue SMT capable.
Unlike on the Cell these cores can handle AI more independently on a per thread basis because there should be no need to farm out portions of the task. This is something to note.
Now given each core has been redesigned they most likely have the potential to clock to very high speeds like the Cell's PPE does (it has been confirmed that all cores clock at the same frequency PPE's and SPE's in the Cell...I...think). The Cell's PPE right now is in the 4GHz range and a schmoo chart supplied at the ISSC by the STI (Sony, Toshiba, IBM) consortium shows that it can be pushed fairly well into the 5GHz range though it gets really power hungry and hot when this is done so they'll need to find ways to curb this in the future no doubt. (going to a smaller fabrication process and whatnot)(I think it's been found that the schmoo had some errors but I've not yet heard that 5GHz was out of the question)
But the rumor has been that the Xbox2's CPU's are clocked at 3.5GHz and now 3.0GHz? (who's to say just yet?)
This makes sense if a core in the Xbox2's CPU are more complex and thus more capable than the Cell's PPE. It's not impossible for the Xbox2's CPU cores to have stronger VMX/Altivec (something Apple is all over) etc as the slower clocked Power5 shows is quite possible. If this is not the case and then the Xbox2's CPU should be able to operate at the same clock rate as the Cell can but I think it's fair to say MS wanted the cores to be a bit beefier in order to offset the Cell somewhat as best it could.
With this thinking I propose that the Xbox2's CPU is composed of newly designed cores similar to but beefier than the Cell's PPE, and that it's deeper capabilities are what bring it's clock rate back down into the 3GHz+ range. (more transistors = more heat and leakage issues = reduction in overall clock rate) This is not necessarily a disadvantage in all cases.
Heres's what the score board could look like:
-The Cell can handle 10 threads in the 4GHz range of operation range but 8 of those threads will be on SPE's that won't take kindly to things that aren't computationally intensive and require allot of branching operations
-The Xbox2's CPU can handle 6-12 threads in the 3GHz+ range of operation where all 8 of those threads can be general tasks that require allot of branching such as AI and to a lesser extent physics and allot of common software
-The Cell should still have an advantage in cases where raw computational power are needed and given how in the overall an Xbox2 core will not be the killer number cruncher an SPE in the Cell is...perhaps significantly...perhaps not so significantly but I predict it will be a fairly noticeable margin in any case.
Here's the big question...which one is more powerful?
If we're talking raw GFLOPS then I expect the Cell will still have the advantage as it is meant to be a computational beast and it still would retain a clock rate advantage too at that.
but when thinking about the usage of raw power...
Well this will depend entirely upon the situation. Even for gaming. With common computer software the Xbox2's CPU would seem to have the advantage to me because it's capabilities are more in tune with the tasks such software call for to be performed regularly. Most of us use software on our computers that wants something to be done when we specify or x happens. Most of us are not scientists that want our computers performing intense computations for the vast majority of time. Another aspect of the situation to look at is inter-dependency. If something can't be done until x happens or is computed then nothing else can happen until that condition is satisfied. The task at hand is dependent upon some other tasks execution here. Often inter-dependency situations are handled by branch and looping statements in code ignoring those which are totally reflective of temporal (time---it's not a simple decision based upon some regularly readily available data...it something that is more bound by how long it takes to produce the data in the first place) limitations. If the Xbox2's CPU is like I describe then it will be much better suited to handle inter-dependency on a per thread basis than the Cell's SPE are. Again a crude and simplistic analogy where the SPE's in the Cell are at a disadvantage the the Cell's PPE and maybe even more so to the PPE like cores in the Xbox2's CPU.(that's 2 threads on something that handles branches well vs. 6-12 threads on something that handles branches well)
To give an example that relates a little better to gaming the Xbox2's CPU should be able to handle AI and to a lesser extent Physics A GREAT deal better than the Cell can if I'm on track here. If AI takes up oh 20% and physics takes up another 20% percent of all the tasks an engine performs...this can be a significant advantage indeed over the Cell chip for the Xbox2 that could go a long way to evening up the score board. (just grabbed those numbers out of the air just so you know...it will be more dependent on the type of game it is)
Although the Xbox2 could have a part in rendering too I think the Cell chip is basically begging to be used towards that end. The Cell being the GFLOP monster that it is and highly parallel...when you think of GPUs it sure sounds like it borrowed allot of it's ideas from them (at least in concept). The Cell should be the better performer towards putting visuals on the screen. Geometry comes to mind among other things.
or will it?
...actually this will depend upon what Nvidia is doing for Sony with the PS3. If Nvidia delivers a GPU that is capable in the areas of both Vertex and Pixel processing with the Cell primarily adding to the vertex processing side of things the PS3 should have a CLEAR visuals advantage with respect to geometry on screen or thinks related the vertex operations. If the Nvidia GPU does only performs pixel processing alone and vertex shading etc is left to the Cell this creates the opportunity for a tie with the Xbox2 in the visuals department or possibly even a loss.
My reasoning for this is that the Xbox2 will be using the R500 GPU that will be a rendering beast unto itself and the fact the Xbox2 CPU can help in the visual department like the Cell can...just not to the same extent.
(when I say pixel processing...think lighting effects like HDR, water effects like per-pixel reflection, per pixel light interactions like normal mapping or virtual disp...think pixel shading which encompasses a wide range of effects and texturing)
I look at it sort of like this.
NOTE: I wish to make it clear that Cell can be used to whatever end a programmer can get it to including pixel processing like tasks which I have every faith will happen. I am speaking to Cell's easier and probably more utilized advantage.
Cell = extreme geometry engine = EGE
Xbox2 CPU = geometry engine = GE
Full Nvidida GPU = extreme geometry and pixel processing engine = EGEEPE
Specialized Nvidia GPU = extreme pixel processing engine or perhaps...times 2! EPE or EPEx2
R500 = extreme geometry and pixel processing engine = EGEEPE
case 1:
PS3 given full GPU by Nvidia: EGE + EGEEPE = EGEx2 + EPE
Xbox2: GE + EGEEPE = EGE * (1.1 to 1.5) + EPE
PS3 wins overall in the visuals department but primarily in the geometry department where a virtual tie in the pixel processing department is not out of the question.
case 2:
PS3's GPU is for pixel processing: EGE + EPE = EGEPE
Xbox2: GE + EGEEPE = EGE * (1.1 to 1.5) + EPE
Xbox2 wins overall in visuals but probably not to a great extent...including geometry. This is the least likely case I feel though.
case 3:
PS3 is given a pixel processing monster: EGE + EPEx2 = EGEEPEx(1.3 to 1.5)
Xbox2: GE + EGEEPE = EGEx(1.1 to 1.5) + EPE
In this case it could be a virtual tie with perhaps one or the other having a slight advantage in either vertex or pixel processing...but you'd be hard pressed to pick a winner or for that matter...to notice it...at...all.
I feel case 3 or case 1 are most likely. It's not difficult to think that Nvidia would supply it's next gen GPU for the PS3 so case 1 makes sense enough as it is. Case 3 will require you to go to beyond3D and look at one of their articles. There an Nvidia representative speaks about how it's approach to the unified shader model differs from ATI's. Long story short ATI is moving in the direction of having it's rendering pipelines being able to handle vertex or pixel processing on the fly i.e. this is a HW implementation. Nvidia is moving toward still retaining seperate pixel and vertex engines that are highly specialized to each task and using software to handle unified shader support. Nvidia claims it's still better to have two engines that are highly tuned rather than one that can do both but is less efficient on both fronts. The the jist I get from it anyway. What is important is that if Nvidia is approaching the future this way then it's not difficult to see that it could simply peel off it's "next gen" pixel processing engine/pipleline and supply it for the PS3 where ATI can't provide a specialized pixel processing part because R500 simply won't be designed that way. I'm not so sure Nvidia would take the stance it's taken thus far if PS3 support were not a factor in the first place. I have my suspicions they would be doing the same thing ATI is and furthermore that they will in the not too distant future...unless Cell's really start taking off in the PC market. Then it's a whole new ball game. It also stands to reason that if Nvidia is going to only supply something for pixel processing that it be twice as powerful as normal in using all pipelines to that end. I say that because Sony must realise that ATI is supplying Nintendo and MS with a full GPU that is FULLY optimized for rendering as that is it's purpose. To demand less than this from Nvidia leaves the door open to get out gunned or even flat out owned. Given how much money they've invested in the Cell and thus the PS3...it's hard for me to see them making such an obviously boneheaded mistake.
let forget specs for a moment and speak about ease of use:
It's no secret at this point that the Cell chip is going to be hard to work with as will any highly parallel architecture will be next round.
It is my thinking though that getting the potential out of the Cell is going to be a more difficult task than gettign the potential out of a CPU like unto the one I describe could be the Xbox2's. The Cell will require programmers to do allot of tasks in software that are done by the hardware on the Xbox2 CPU. One such thing is direct memory management of the SPE's memory. Another such thing is branch prediction in software if you really don't want your SPE's sitting idle all the time. Another larger problem for programmers may be trying to take advantage of the Cell for graphic rendering vs. rendering on a GPU they know and understand. This isn't going to be like how software rendering was handled before because SPE's are not like CPU cores programmers have contended with before. It is not unfair to say that the Cell processor may be using a fair amount of it's horsepower running software to handle capabilities it SPE's do not have in hardware for coders. How significant this is is unknown at this point but I think it is fair to note that the difficulty in programming the Cell may further diminish the lead it may have in real world performance over the Xbox2's CPU as I describe it.
The Xbox2's CPU should be something programmers will not have near as much difficulty in coming to grips with and given that it's cores will utilize memory virtualization via caches etc it's hidden potential should be more attainable. Also unlike the Cell SPE's a core in the Xbox2 CPU should spend more time on average towards actually completing a task than running code to replace missing hardware capabilities.
Thank you if you've read everything up to this point.
Here's where the Revolution comes in....
IBM is also working on the Rev's CPU just like they are for the Xbox2 and the Cell.
ATI is also working on the Rev's GPU just like they are for the Xbox2.
It's a simple and believable assertion. I believe Nintendo will get similar parts from both partners. I expect there will be some superficial differences here and there...and I hope that any scaling back on the spec sheet if done at all is kept to a minimum.
I do not believe the Rev will use 2 or Dual G5 processors as they are to big, to hot, to power hungry have more capability than it needed.
That's it. Pure and simple. I mean do you think Nintendo is going to have the same people working on their parts and watch as they don't get equal or better treatment from them? I don't...or at least I haven't until recently. It's been said countless times that Nintendo doesn't make under-powered systems and I will hold onto the hope that when the goodies are virtually right in front of them...they will go for the nookie.
Also, this would make the most sense for IBM. They have already dumped a ton of money into the Cell. It only makes sense that they re-use what they can from all that R&D. To design three "completely" unique CPU's simultaneously seems allot less plausible to me than simply being selective with what you lend with respect to what you've already done. I know MS would have paid the money to get any revelations IBM came across in making the Cell's PPE applied to their own core and I at least have the hope that Nintendo would do the same. I would also think that the PPE or rather the core that is compliant with the PowerPC ISA is an intellectual property etc that IBM would retain while the SPE's which seem to be born of the PS2's Emotion Engine would belong more so to Sony. (Toshiba basically brings it's manufacturing expertise with DRAM chips to the table...someone has to make that Rambus XDR and the interface to it...that would have most likely been Toshiba's role) What I'm saying is that IBM should have enough freedom to given Nintendo and MS cores for their CPU's that can contend with or even out gun the PPE in the Cell chip. What IBM can't do on it's own is sell Cell's to Nintendo and Microsoft. This may not be necessary in that the Big N and MS can get what they need from the Cell tech without actually using them. Rambus can license it's Yellowstone and Redwood tech to MS and Nintendo in turn as well I think.
Remember! The guts of the machine will have no bearing upon how the interface to it is setup. If Nintendo wants gyroscopic controls...vr goggles...or a life size dancing Rob the robot to come with the Rev...this will not prevent them from doing this. Personally, I like controllers, I don't want to hit bongos or wear a full body rumble-suit-power gloved thinga-ma-jigger but if that's what floats your boat Nintendo can still deliver no probs. Of course I reserve the right to latch onto anything truly Revolutionary as being better than bottled sunshine and will reject having ever been in doubt of anything if that should become the reality.
I admit that at first the PS3 was looking like the elephant in the room and I was having a hard time seeing any way for the competition to compete with it. Thanks to a mere suggestion by an IBM fellow in a discussion about the Cell and the next gen I began to re-think things. I think what I've theorized is plausible. If it's the actual reality then all is not lost for MS and Nintendo in the next round. I do think the PS3 will probably have the most impressive specs in terms of big numbers it can push. I am very optimistic and hopeful now that when it comes to putting specs to good use that the Xbox2 and the Revolution could be on par with the PS3. Perhaps even outrunning it.
In any case I think next gen things should be quite interesting and for once in a long time I feel...there is chance that we all can be pleasantly surprised at how things turn out.
With that I close and give a link to an article that suggests I may be right in at least guessing what comprises the Xbox2's CPU.
http://arstechnica.com/news.ars/post/20050309-4686.html
I've seen some speculate that a Physics Processing Unit will be the third core on the Xeon CPU.
I have my doubts. A PPU crosses to far into specialization...but since it's a console that's not so solid an argument. I'll go to some other places then. To me it would make sense to use it but place it elsewhere in the system. Putting a PPU on the CPU would make IBM's task that much harder etc. It would also eliminate the advantage of having another core AND and PPU unit. It could provide some significant visual advantages though such as cloth interactions, water that acts like fluid, more particles on screen, better explosions etc. This would be a nice advantage if no one else used a PPU. It would also mean you could not let PPU accelerated physics seep into game play for multi-platform games in the console arena. Yet with XNA and the possibility of PPUs in the PC realm you could still port an Xbox exclusive to the PC market.
It is my hope the all next gen consoles use a PPU and that PPUs themselves find a place in the gaming world.
...ok NOW I'm shutting up
