Topic: Fact Only Thread: Rev Tech Specs

[BlackGriffen]

CPU: made by IBM
GPU: made by ATI
LSI (bridge chip, I believe): made by NEC

Barring yet another custom variation that uses the PPE in the PS3/Cell and XBox360, it looks likely that Nintendo will be using some variant of the PPC970 (aka G5).

Given the historical rate of improvement at IBM, I expect that any 970 Ninty might use would have to be on this list:

• 970MP (dual core, 1.4-2.5 GHz, 1MB L2 cache per core [2 MB total], optionally turn off one core for low power usage)
  
• 970FX low power (uses 13W* at 1.4 GHz, 16W* at 1.6 GHz, runs up to 2.7 GHz, 513 KB L2 cache)

*Power usage figures are typical, not maximum - hardware design needs to handle maximum.

Those are Nintendo's 970 options. It is not safe to speculate about the power consumption of the MP model based on the FX - the size of the caches are different, for starters. Suffice it to say, given the small size of the Rev, I would be surprised to see an MP at any faster than 1.4 GHz if we see one at all. If Nintendo goes single core, I would guess that 1.8 GHz would be the top speed they could use. Maybe 2.0 GHz if they push it.

Remember, however, that Nintendo may not take this route. They may opt for something based on the same much simpler cores that MS and Sony are using.

BlackGriffen

[bmfrosty]

http://cube.ign.com/articles/100/100445p1.html

and

http://cube.ign.com/articles/100/100543p1.html

are the two parts of an interview conducted with the engineers at IBM who designed the Gekko for the Gamecube.

It wouldn't surprise me at all if Broadway is an evolved Gekko, run on a 90nm process, with a larger cache, a higher clock speed, and possibly with two cores on the same silicon with an interface for dealing with it as a two core piece, or alternatively run as two seperate cores in a single package.  Remember, the Gamecube was the best designed (arcitecture not visual style) system of the 6th Generation.  I doubt that Nintendo will stray too far from it for the 7th Generation.

Here's another arcitecture article here:

http://www.eet.com/story/OEG20010516S0056

As for the Hollywood GPU, I don't expect it to be derived from the flipper.  It will be something new with a feature set that is a superset of the flipper features.  The beauty part is that if we are dealing with a dual cpu, and the Flipper and the Hollywood have different instruction sets, the second CPU in the revolution can be used to transform Flipper instructions to Hollywood instructions, allowing fast and good backwards compatibility with the Gamecube.

I expect good things from the Revolution.  I expect that it will support a much more rapid environment for code development than either of it's two closest competitors will be able to provide.  The real hope and dream this generation involves whether or not Microsoft and Nintendo are able to keep Sony from taking a huge lead like it did in the 6th Generation.
 

[OptimusPrime]

I Also believe Hollywood and Broadway will be huge evolutions of the Gekko and Flipper. It makes backwards compatibility easier (unlikesome other upcoming console) and makes sence that the Rev uses the same API as the GameCube. Besides the Gekko was a very tiny chip that didn't produce much heat and Nintendo is always raving about a efficient low-powerconsuming console: evolved Gamecube is just the answer.
Nintendo is going to be demanding hyperefficiency from its hardware, something both X360 and PS3 are hugely going to lack. Sure they are 10 times more powerfull then their predecessors, but thats theoretical, it will take developers 4-5 years to get that if they go trough all the problems with multicore and multithread programming and that cotst lots of time which means a lot of money... in comes the big difference with developers (who want to spend that time) and publishers (who want to make as much of money and as fast as possible).

[BlackGriffen]

I, for one, will be stunned if Nintendo goes with a Gekko/G3/PPC750 lineage processor. Why? Sorry to burst your bubble, folks, but the thing is weak as hell!

Let me count the ways:
* No G3 has been clocked faster than 1.1 GHz, ever. The pipeline is simply too short to clock higher.
* The execution resources on the G3 are anemic: 1.5 integer units, 1 floating point unit, 1 load/store unit, and zero vector units. The thing is comparable to a Pentium II in both age and oomph! The couple extra special floating point instructions for the Gekko aside.
* The front side bus (the thing that hooks the CPU to the rest of the system) on G3s (likely also Gekko, but that's information I don't have) is likewise too slow.
* It has zero ability to execute instructions out of order.
* Does not have the hardware necessary to play well with other CPUs - ie no multi CPU or multi-core.

Changing any of these things would require enough of an investment on Nintendo's part that it would be cheaper to just go with one of the other solutions IBM provides. Hell, Nintendo would be better of with a Motorola/Freescale G4 than with a G3, hands down.

We know that they went with IBM, though, so that leaves two realistic options: PPC970/G5 and PPE/Cell. They both have their up sides and down sides. A review:

PPC970:
+ Beefy execution resources: 2 integer, 2 floating point, 2 load/store, and 1 AltiVec
+ Out of order execution (OOOE) - makes it easier to program for, Cell and Xenon, by comparison, will have steep learning curves - also significantly improves performance of code with branches (think AI, game control, etc).
- May only be able to squeeze one into the tiny Rev.
- Can only run one software thread at a time

PPE:
+ Will be able to run at higher frequency.
+ SMT = can run two threads at once
+ Competitors will be using it, making ports easier
+ Guaranteed dual core at least.
- Anemic execution resources: 1 integer, 1 floating point, 1 load/store, and 1 AltiVec
- Zero OOOE - rumors have developers complaining that the slow down for branchy code is as much as a factor of 10 compared with standard PC cpus of similar speed! Expect it to get better as coders learn some tricks, but the gap will never completely close; the PPE is optimized for multi-thread throughput (which is why each core runs two threads despite its slim execution resources) on the cheap and not single thread speed.

That's it. If Nintendo goes with a PPE don't expect their processor to be any better than the XBox360s. Indeed, going this route Nintendo would likely opt for two cores instead of three, IMHO, with a clock speed of 2.5 to 3 GHz. If Nintendo goes the PPC970 route, expect numbers in line with what's at the top of the post. That is, unless IBM comes up with a miracle - something they haven't been known for these days.

Nintendo could, hypothetically, go asymmetric like Sony (Cell = 1 PPE + 7 SPE; asymmetric because not all cores the same) and have one PPC970 and one PPE in the chip. I don't think it's likely, but I can't think of an explicit reason why it couldn't be done. Just several reasons why it wouldn't be done: harder to program for, IBM would probably ask for a money hat to crowbar the two together, etc. Still in intriguing idea...

BlackGriffen

[BlackGriffen]

Also, re: APIs remaining the same. All that means is that the machine will be able to handle OpenGL (a standard feature of all graphics cards) and have a PowerPC processor that has been tweaked to handle Gekko's special instructions.

It also means a bunch of other minor stuff like the ability to handle the GC audio and any specialized bits from the Flipper. These things are not a big deal - the audio stuff will, I imagine, be incorporated into the NEC part and ATI will naturally be able to include any of the other special sauce from Flipper into Hollywood.

BG

[ABlueflameA]

Uhm, BlackGriffen, have to disagree with you a bit there.  

1. I used to own a P2 and there is no way in hell that it could play the games that are found on Gamecube, especially, but not limited to, RE4.

2.  The gamecube processor is based off of powerpc archetecture, meaning we're back to RISC vs. CISC.  Sure, IBM's may have larger numbers associated with them as far as mhz or ghz go, but a smaller numbered Mac will still beat the higher numbered PC all 7 days of the week.  Now, as not to compare apples to oranges, AMD processors also generally can be over 500mhz lower than a comproble pentium processor and still outperform the hell out of it.

The only other important thing that I think should be mentioned is that black and white numbers DO NOT tell you how good of graphics or how good of games the system will be able to play.  Case in point would be the PS2 which in theory is godly strong, but in actuality is such a pain in the ass to program for that getting 100% of the available power out of it is nearly impossible.  Meanwhile with other systems, it is possible to actually get close to using 100% of the system.

Oh, and after all the advertising and outright preaching Sony has done about the Cell processor, there's about zero percent chance that it will be found in Revolution, and thus is not a realistic option.  Which, may be good, as I hear that it too may be complicated to program for.

-Blueflame

[KDR_11k]

A P2 might be able to play RE4 if you throw a Radeon 8500 in there, remove the OS and optimize the hell out of your code. That's not a realistic situation for PCs but for a console, it is.

The Cell is said to be very easy to code for except for the in-order logic of the PPE, the vector units are addressed separately from the PPE so it's not really like coding for a multiprocessor machine. The X360 OTOH requires six equal optimized threads at least to get performance out of it, for the Cell you make one optimized thread and code some sections in your code in a way that it can be shipped off to a SPE.

Personally I'm expecting to go Nintendo with a lower CPU count, probably a single processor because they're a software company first and foremost. Singlethreaded code is much easier to write or debug. Perhaps a modified Cell (less SPEs, more special instructions). Sony might have advertized the hell out of it but the project is still a collaboration and it was built for much more than just the PS3. I don't think they're even allowed to deny licensing to potential competitors.

[BlackGriffen]

KDR already addressed most of your points, ABfA, but I felt that one was worth expanding on. You're argument that a P2 could handle GC games and I'd bet you're right. The reason you're probably right, though, has nothing to do with architectural prowess and everything to do with the fact that it appears that Intel never made them at a speed faster than 333 MHz and the Gekko is near 500 MHz. If you got a PII that runs at about the same speed and was tweaked with Gekko's special fp instructions, I would be flabbergasted if it couldn't stand in for the Gekko. Sure, the game would have to be recompiled for the new ISA and any assembly replaced, but that's about it.

Look, I'm not saying that the Gekko was outright bad for the GC. It has done it's job admirably. It was fine for this gen, where it was competing with a Pentium III and the Emotion Engine, but it doesn't have the legs needed to perform well next gen. It has architectural features that were necessary at the time but that are severe shortcomings on the performance front now. Just like the Opteron outperforms Pentiums because of architectural features - not some supposed strength of its instruction set (ISA). Specifically, the integrated memory controller and the more abundant floating point execution resources help on this front.

Also re: ease of programming of the Cell. I'll believe it when I see it. The PS2 has three processor cores: the Emotion Engine general purpose processor and two vector units, V0 and V1. The Cell has the PPE (general purpose processor) and 7 SPEs (mainly vector units that look to be beefier than V0 and V1). If I recall correctly, Sony originally planned to just use the Cell like the PS2 just had the EE+V0,1. The Cell worked so well and was so easy to program for that they brought Nvidia on board to provide a graphics chip. In fact, that live demo at E3 was just running on the PPE and graphics chip - the vector units weren't even involved. In general, it just plain isn't easy to use multiple cores in parallel to do a job faster than one beefier core.

So, like I said, I'll believe that they can automagically make the Cell easier to program for when I see it.

BG

[jasonditz]

Since backwards compatibility and design simplicity seem to be the orders of the day, a G4-based CPU would make a lot of sense. Think Gekko with a higher clockspeed and an Altivec engine.

[ShyGuy]

Wow, was it that long ago that Pentium II CPUs were common place? Intel made the PII processor up to 450mhz, not 333mhz.  

[KDR_11k]

BG: Apparently the API handles the SPEs, for the PS2 you had to address the VUs with assembly language. Makes the thing a whole deal easier.

[BlackGriffen]

Quote
Originally posted by: jasonditz
Since backwards compatibility and design simplicity seem to be the orders of the day, a G4-based CPU would make a lot of sense. Think Gekko with a higher clockspeed and an Altivec engine.

Except that a G4 is not just a G3 + Altivec. You remember how I said that the G3 had 1.5 integer units? Well, that 1.5 was meant to mean "1 full integer unit + 1 simple integer unit". The difference between the two being that an integer unit can do anything, whilst the simple one can only do a few operations like add and subtract. The reason that it's important to distinguish the two is that a G4 has 1 full unit + 3 simple integer units.

The G4 also has a longer pipeline. If you imagine a CPU as being like an assembly line, then dividing the work up in to more stages means that each stage takes less time. Thus the processor can be run at a higher frequency. I'm vague on exactly how many stages are involved, but I believe the G3 has 4, the G4 started with 7 and may have gone up to 14, and the G5/PPC970 has more than 20.

Another big difference is how the CPU communicates with the rest of the system, called the front side bus or FSB. The G4 FSB is plain better than the G3 one (more bandwidth and lower latency). Sad thing is, the G4 FSB still sucks by today's standards. That's why Freescale (formerly part of Motorola) is going the route of AMD's Opteron and adding a memory controller to the next G4s. Think of it as cutting out the chip that normally sits between the CPU and RAM.

The biggest problem with a G4 theory is that the G4 is Freescale's baby and nobody else can make it without permission.

Re: PII speed. Ooops. I didn't do a lot of in depth research, as you can tell, and just went by the top speed Intel makes available now. At 450MHz the PII is probably close enough to being as fast as the Gekko to make it plausible.

Re: KDR. Very interesting, if true.

Also, there are now more details available on the new single core PPC970s. This PDF has a table of maximum power usage at various temperatures. Important data follows (power in watts @ 85 C, power in watts @ 105 C):

• 1.0 GHz: 12 W, 13 W
  
• 1.2 GHz: 15 W, 16 W
  
• 1.4 GHz: 18 W, 19 W
  
• 1.6 GHz: 20 W, 21 W
  
• 2.0 GHz: 47 W, 50 W
  
• 2.2 GHz: 56 W, 60 W

For the non-power optimized parts:

• 1.6 GHz: 27 W, 29 W
  
• 1.8 GHz: 35 W, 37 W
  
• 2.0 GHz: 56 W, 60 W
  
• 2.2 GHz: 71 W, 76 W

Just judging by what I understand to be acceptable in a laptop computer (ballpark max 30 W), Nintendo could go with 1.6 GHz single core (1.8 GHz if the push it) or, just guessing that MC power ~ 2x, 1.2 GHz single core (1.4 GHz if the push it).

Also, a caveat from the data sheet:

Quote
Important: The data in this table are based on the best available simulation data at the time of  publication and may berevised after hardware characterization. Because they have not yet  been correlated with production-qualified hardware, these estimates should only be used as  guidelines for design.  

In other words, YMMV.

Something interesting occurred to me, though. If you look just at the execution resources of the PPE, it looks vaguely like a Gekko + AltiVec + SMT - simple integer unit. The PPE's big advantage being that it's floating point unit is better than the Gekko's. In other words, we'll be more likely to see a PPE single or dual core chip than we are a Gekko evolution.

BlackGriffen

[BlackGriffen]

A thought recently occurred to me. The Elastic Interface Bus (EIB) that the PPC970 uses is rather complicated and sensitive. For a detailed description of the process of engineering a board for it go here. Long story short - a PPC970 actually needs another CPU in order to help it start its bus and run. My understanding is that in Macs this is a PPC440. I can't think of any reason, however, why Nintendo couldn't use a Gekko to do it. Doing so would certainly simplify backward compatibility - no longer any need to tweak the 970 design to run the Gekko's special instructions (an expensive process).

Now, whether Nintendo could use this extra Gekko for other things, I don't know. I'm not sure if they'd want to, to be honest.

So, I must amend my prediction. I think Nintendo will go with one of the following:

- single core 970 + Gekko
- dual core 970 + Gekko

If Nintendo can actually use the Gekko for running Rev software, it would make sense to run it at a higher speed than the GC's roughly 500 MHz. Current G3's available suggest an upper limit of around 1 GHz. If they do go that route then I would expect a Rev with even a single 970 core in it to perform comparably to the competition - worse in some areas, better in others, but well enough to compete regardless.

For those who think this is ridiculous, remember how Sony got backward compatibility in the PS2 - they used the PS1 CPU as a sound chip. Since an EIB requires a "wake up" CPU anyway I can't think of any reason for Nintendo not to use a Gekko of some form in this role. That is, assuming Nintendo uses a chip with an EIB (a PPE based chip could use it also, I think), which is not a given, but plausible if Nintendo doesn't want to spend money tweaking.

BlackGriffen

[Spak-Spang]

BlackGriffen:  Wow.  That last post makes alot of sense.  The part that makes the most sense is that it is actually cost effective for Nintendo to do it the way you described.  The Gecko chips are already in mass production for Nintendo in the Gamecubes, and a few minor tweaks and those chips can just be placed into the Revolution for Backwards compatiability and whatever other needs are necessary.  

Interesting.  

I wish we could just get the specs from Nintendo.  I can wait for the controller and other secrets...I just want to know 3 things:  Specs, Price, and Release date.  I really can't wait to get my hands on this system.

[BlackGriffen]

Thanks for the compliment, but after consulting with some people who actually know about this stuff (as opposed to just reading the forum posts of people who know about it for the last 5 years or so) I would put the chances of a Gekko + other chip at slim to none. The simple reason is that a Gekko would be massive overkill as a boot processor for an EIB. The part Apple uses is called an HC08 - a dirt cheap 8 bit processor that went for a buck or two at retail prices a few years ago. That plus the work necessary to actually make a Gekko available for use while booted in Rev mode would be substantial.

Not to mention the fact that if Nintendo wanted to tweak the main CPU in any other way, and it seems likely that they are judging from their own statements (custom CPU Broadway, etc), then the additional cost of making it able to support the Gekko's custom instructions is comparatively small.

So, there's a remote possibility it could happen, but it's down there in likelihood with Microsoft buying Nintendo.

BlackGriffen

[Rancid Planet]

Quote
Originally posted by: BlackGriffen
So, there's a remote possibility it could happen, but it's down there in likelihood with Microsoft buying Nintendo.

Yeah, I'm scared that Bill Gates may want a bigger parking lot one day.

[MrMojoRising]

Quote
Originally posted by: Rancid Planet

Quote
Originally posted by: BlackGriffen
So, there's a remote possibility it could happen, but it's down there in likelihood with Microsoft buying Nintendo.

Yeah, I'm scared that Bill Gates may want a bigger parking lot one day.

Then they truly would have paved paradise and put up a parking lot