Predict what processor will be used for the Xbox 360

[Shifty Geezer]

PS3 is the driving force behind Cell. Without a new Playstation, work on Cell would never have started.

Yes and no. KK stated he started dreaming of a cellular, organic approach to computing at PS2's launch. He envisaged (from what a I gather) a super-network of processors in many machines around the globe, all sharing processing. This wasn't just for games, but entertainment as a whole.

I bet, when Sony sat down with IBM and Toshiba, PS3 wasn't the only platform they considered designs for. IBM would have added 'we want this type of scalable processing for mainframes' and Toshiba would have said 'we need this level of downscalability to fit into other devices.'

As a single platform, sure PS3 will be the largest application of Cell, but overall I imagine Cells might well appear in more numbers in consumer electronics of various sorts. PS3 is just a slice of the Cell pie; a pie that incoroporates professional TV/film production work, portable handheld computing, living room media equipment and so forth, and was as designed for such from the beginning as it was for next-gen gaming.

 

[pc999]

Dont you read DeanoC posts

 

[blakjedi]

I hope guys dont mind my laymen extrapolations...because this is xecpu thread i thought id do the mind exercise to trying to figure where the >than 1 terflop in targetted performance J. Allard was talking about is going to come from... theres only one of three options

1. j allard is a liar - I wouldnt bet on this one. He may be exaggerating to stop the Sony spin cycle before X2 approaches. the key to Allards statement is targeted performance. The point that leads me to is if you threw every ounce of computing power in the x360 at a problem. you "might" "could" in a perfect world get over a teraflop. Nothing wrong with this approache however considering Sony's numbers also reflect peak usage and raw data rates not avg.

2. The R600 (which I think is whats really in the x360) is much more of a beast than we think - think 800+ Gigaflops. Though I think the GPU will be R600 class I certainly dont think it is capable of of more than say 500 GF peak raw fillrate.

3. The XeCPU either has features more akin to some of the conjectures in this thread and/or there is more than one XeCPU in the box. Unfortunately even at 162GFLops X 2 thats only 334 GF leaving 600+ to other silicon...

So the math is still undone at this point and I havent seen anyone come close to filling the silicon gaps, as it were, on this site or anywhere else.

 

[Carl B]

I think if you look around this site, and others, you'll see plenty of those gaps filled in. The best thing to do is probably go find the claimed flops performance for the original XBox - it should be evident almost immediately where Allards numbers are derived.

And it's definitely the R500.

 

[blakjedi]

I think if you look around this site, and others, you'll see plenty of those gaps filled in. The best thing to do is probably go find the claimed flops performance for the original XBox - it should be evident almost immediately where Allards numbers are derived.

And it's definitely the R500.

Nah iv'e looked. Pretty much read everythread dealing with this issue since October of 04. The thread which fills in the most is the "second gen cell processor thread" which i cited earlier. Which is exactly in line with what I've suggested in this thread... 6 VMX units... but the numbers dont add up no matter what... i dont think looking at xbox is a good indication either...

as they are completely different design parameters and needs. the design parameters are radically different between Ps1 and PS2 so i would not use PS1 as model of where Sony was going when looking at the chipset design for PS2.

Only the PPU discussion so far fills in the power gaps but i dont believe that there is a PPU in X360 so i didnt even add that in as a variable.

 

[Carl B]

Well, let's put it this way - the original XBox claimed well over 80 GFlops of performance, of which about ~3 of those came from the CPU. Now if we scale these numbers forward into modern hardware, I honestly have no doubt that somewhere over at Redmond Mathematics Headquarters, they've got the flops rating of the GPU alone at a good bit over a terraflop - so when talking about 'system performance,' and if using the same method to derive it this time as they did last time, I'd honestly be shocked if the 360 were NOT over a terraflop of 'targeted' performance.

 

[vblh]

I assume that those quotes for the X2 cpu are based on the G5 processor. The G5 is based on the IBM Power4. Problem is I've read that the X2 cpu is based on the Power5. So if the G5 is basically a cut down Power4 don't you think all the calculation being made are going to be way off? I think we need to find out the difference in processing power between the G5 & Power4 & Power5. Only then you can understand how Allard can claim 1Tflop of targetted computing power. Also, the supposed "leaked" white paper claims that their goal was to focus on general computing power not any one uber chip.

 

[Carl B]

I assume that those quotes for the X2 cpu are based on the G5 processor. The G5 is based on the IBM Power4. Problem is I've read that the X2 cpu is based on the Power5. So if the G5 is basically a cut down Power4 don't you think all the calculation being made are going to be way off? I think we need to find out the difference in processing power between the G5 & Power4 & Power5. Only then you can understand how Allard can claim 1Tflop of targetted computing power. Also, the supposed "leaked" white paper claims that their goal was to focus on general computing power not any one uber chip.

Power 4, Power 5 - they're both equally wrong. Talk originated back in the day that the 360 CPU would be based off of 'Power,' and so people started deriving where IBM would be in it's traditional family of chips at the time and what might make it's way into the XBox. But it turns out that everything is pointing to the XBox CPU being an in-order chip very much more akin to the PPE unit inside of the Cell than to anything else prior in the Power family, so it's best to let these thoughts of Power 4 and Power 5 go.

 

[vblh]

Understood, but my point was simply that the figures for the X2 cpu were based on the G5 chip. As for the ppu from the cell & X2 being the same i'll wait till the official info comes out. The cashe size & the number of vector registers may be the same but that wont make them the same. X2 won't have the spe so they have to do something to make up the difference. Besides, i doubt MS would be willing to use the same processor as Sony. It's all about bragging rights & the two of them will want to say they have the better tech. Well that's how i see it anyway.

 

[Carl B]

Understood, but my point was simply that the figures for the X2 cpu were based on the G5 chip. As for the ppu from the cell & X2 being the same i'll wait till the official info comes out. The cashe size & the number of vector registers may be the same but that wont make them the same. X2 won't have the spe so they have to do something to make up the difference. Besides, i doubt MS would be willing to use the same processor as Sony. It's all about bragging rights & the two of them will want to say they have the better tech. Well that's how i see it anyway.

Well, I'm just going off of the fact that the 360 CPU is supposed to be in-order. If it is, then it doesn't matter if it's necessarily like Cell or if it's not - it's just not going to be very similar to the Power4 or Power5 is all I'm trying to say; probably more similar to the PPE core inside of Cell.

 

[blakjedi]

The real question is this... are six VMX units with two VMX per core "better" than 8 SPU's tied to a single core...?

Answer that question correctly and we'll have a better answer about XeCPU's power versus the Cell. Not that it makes a difference cause games are what its all about.

 

[Carl B]

Well, flops-wise the Cell would still win out, but the truth is that depending on the frequencies of both the PS3 and the 360 CPU's, the number of SPE's on the Cell, and the exact form the 360 ends up taking, it's pretty much impossible to talk about the clear-cut differences between the two except in a very generalized manner. Flops isn't everything, afterall.

 

[Mythos]

The real question is this... are six VMX units with two VMX per core "better" than 8 SPU's tied to a single core...?

Answer that question correctly and we'll have a better answer about XeCPU's power versus the Cell. Not that it makes a difference cause games are what its all about.

Drop it.

The consensus is that the over 1 tflop claim for X360 is marketing speak.

 

[Shifty Geezer]

@ Blakjedi : If so I doubt Cell would exist. According to the news we have, Cell was designed when bods from IBM, Tosh and Sony got together to talk about what a multimedia processor needs to do and how to do it. An awful lot of money and time was spent on a clean sheet design to solve this problem, resulting in Cell. If half a dozen VMX units can solve the multimedia requirements, at design time why didn't IBM say 'Hey, guys. If we just chuck a handful of these things on some silicon we've got a solution,' and save everyone a lot of time and effort (and money)? What you're suggesting is MS spending less money and involving less research to use existing tech that launches earlier than Cell, and yet outperforms it. That flies in the face of logic, surely? How can STI have ploughed so much into Cell to end up with a less effective solution then current tech?

There's only two outcomes to Cell.

1) It's a sham. A 1:8 Cell processor is no more powerful than an equivalently clocked P4 on the same die size. Both can stream 48 MPEG 2 clips simultaneously. All this money spent is a bluff, creating a belief in a new technology that can't get anymore power per square millimetre of silicon than any other solution. Any other multicore solution, from Intel, IBM or AMD, will have the same performance/die area ratio. As such XeCPU is comparable in performance to PS3CPU.

2) It works. A 1:8 Cell significantly outperforms other processors in media/stream based applications, in a way existing technologies can't match because of inherant implementation trestrictions. That's why Cell was created instead of just multi-VMXing an existing design. As such, PS3CPU outperforms XeCPU substantially.

I'm more willing to believe in the second case. It just doesn't make any sense to me otherwise!

 

[Titanio]

SPEs and VMX units aren't really directly comparable imo. SPEs dont' depend on their core like VMX units do, they have their own memory which VMX units don't (registers aside), VMX units can't talk to one another (AFAIK?), VMX units don't their own access to main memory (again AFAIK?), VMX units can't run general purpose code. SPEs are closer to a "regular" core than a VMX unit, I think.

And as Shifty points out anyway - STI's goals could have been achieved by just sticking another VMX unit onto a Power core, I'm pretty sure IBM would have stepped in at some point and made the suggestion

 

[blakjedi]

Hey Mythos I am not enamored with or disillusioned with the Cell either way. Its interesting that people believe Sony's one Terflop comment but not MS'...

This is the predict what processor is in the 360... since no one knows... *shrug*

Shifty - personally I agree with your statement especially with how you've framed the argument. But that means one of two things by extension-

1 - N-rev has no chance in hell that they will devise a CPU that equals or surpasses Cell anytime soon based on tech from IBM. Also this means that every design team at IBM agrees that Sony Cell is the best CPU design possible based on IBM's technology for the last five years.

Thats possible but unlikely...

2 - Cell can matched/beaten using IBM tech just organized differently. Matched/beaten may just mean that for the needs that Nrev or MS have, whatever those needs are, Cell WOULDNT work the best.

None of us have the answer or would publicly comment that they do at this time. 8) Again it doesn't matter to me per se, but since we dont know much about the XeCPU it cant hurt to be reasonable and imaginative at the same time.

 

[Titanio]

Hey Mythos I am not enamored with or disillusioned with the Cell either way. Its interesting that people believe Sony's one Terflop comment but not MS'...

Errr...no one here believes Cell can "do" 1Tflop :?

 

[Shifty Geezer]

As far as I can recall, the 1 teraflop figure was for Cell's capabilities or something. It wasn't specifically applied to PS3 by KK, but the media started pinning it as a benchmark figure for the system. No-one here's expecting 1 TFlop in PS3, at least not conventional processing.

It seems to me that MS's 1 Teraflop figure was a knee-jerk marketting rection to Sony's. Any number you can use, we can use better! I don't think the numbers are at all relevant as they neither offer fair comparisons of the systems (because both parties will be making up whatever numbers they can to out-value their opponents, including even the system's number - 360>3!!) nor provide useable benchmarks that give an indication of what the systems' realworld performances will be like. Kinda like this generation, where fans toot numbers showing one system's better than the other, but I can't see much difference between them all myself.

 

[blakjedi]

As far as I can recall, the 1 teraflop figure was for Cell's capabilities or something. It wasn't specifically applied to PS3 by KK, but the media started pinning it as a benchmark figure for the system. No-one here's expecting 1 TFlop in PS3, at least not conventional processing.

It seems to me that MS's 1 Teraflop figure was a knee-jerk marketting rection to Sony's. Any number you can use, we can use better! I don't think the numbers are at all relevant as they neither offer fair comparisons of the systems (because both parties will be making up whatever numbers they can to out-value their opponents, including even the system's number - 360>3!!) nor provide useable benchmarks that give an indication of what the systems' realworld performances will be like. Kinda like this generation, where fans toot numbers showing one system's better than the other, but I can't see much difference between them all myself.

Agreed and well said.

 

[gurgi]

Naw, cell was designed for PS3. It's budget was justified with all these other proposed functions. I'm not sure we'll ever see tv's or toaster ovens on the market with cell chips in them, honestly.