Xbox 2 specs leak - Page 2

And what space would that be? What 4th core?


Games use around 99.5% of the available CPU, so you pretty much have to code the game to use it. Everything else that is expensive is in the GPU or DMA controllers.

The reason why people are doubtful about G5s being used in the Xbox is that a powerful desktop chip has never been used in a games console - at the same time being used in computers...

This is ridiculous when you take into account what has been used in consoles so far.

let me bring up a new point:
heat

so we know apple had a hard time fitting 2x2GHz 90nm G5s into a 1U enclosure.
Now a console is about half, if not a fourth the size of a 1U rack. And you're telling me they're going to stick 3 (three, tre, trois, drei, golme, tri, etc.) into a box, approximately the size of an Xbox?

at what speed, I ask? at 500MHz?

Unless they want to make a howercraft that is so loud even a 7.1 sound setup won't be able to match it, they will have to think of something absolutely revolutionary, IF the rumors have ANY truth to them.

I think he means that the cores will be placed in a 2x2 matrix, where 3 slots are used by PPC-cores and the fourth are being used by the supporting ASIC.

Couldn't the cores be placed three in a row? Since yields aren't 100% it must be easier to find three working cores in a row, than the suggested 2x2 design.

They don't have to be the same shape, and there are other options for the 4th corner... like a shared cache, for example. If you look at the existing XBox's design you'll note that the "system ASIC" is actually part of the GPU, which has a more pressing need for massive bandwith.

As for repeated comments about heat... the whole point of using the 9xx series design is that it is the most power efficient high-end processor available presently. Scaled down to a 65nm SOI process with IBM's cooling and PowerTune technologies, this thing's heat should be less of an issue than ATI's GPU.

I was imagining a large rectangular die with three cores arrayed like so:




Where * is a core. That leaves room for a good-sized ASIC's worth of stuff to be integrated on board.

I realize that this design is neither necessary nor inevitable - you could just have three cores in a rectangle, or something like that - but the first thing I thought of was a SoC design, and that was the layout that popped into my head.


Ah. In that case, you're splitting the game into three concurrent threads, which is not actually all that hard if you factor your code properly.

For the benefit of others, the main reason games aren't threaded now is that, with the main architectures consisting of single high-MHz CPUs, threading doesn't make a lot of sense. But the code can still be well-factored: For instance, even though the PC version of Giants was single threaded, they wrote clean enough code that Omni was able to thread it easily to run on DP PowerMacs.

Please excuse me 'cause I haven't read any of the previous posts.

Yes- I don't think the Xbox Next will be running 3 2GHz G5's. I'm thinking they'll be at a much lower clock speed (to keep heat and cost down).

How low do the 970's clock? Was it 1.4GHz? Maybe by late 2004 they'll be cheap enough.

I think this is a fake leak. Microsoft FUD, and tweaking Intel at the same time.

The Xbox2 will be a cheaper-to-make design than the current Xbox. It probably won't even use a PPC - IBM can make X86 chips, they used to make Cyrix's.

Why didn't the Xbox take the market by storm? Because (a) the Playstation had more games and backward compatibility, and (b) both Nintendo and Sony engaged in a war of price-cutting attrition. Is Microsoft so stupid as to make a more expensive box which is not backward-compatible? Of course not... altho they may want people to THINK that.

Now that IBM Semiconductor has been folded into the server group, and IBM committing to Linux over Windows, it appears that IBM is no longer doing the sort of bet-hedging that it used to. They're going to consolidate their efforts around POWER/PowerPC now that there is no pressing need to support Windows on the server side, and x86 offers no other advantage, and any number of disadvantages (including the simple fact that IBM owns the IP to PowerPC, the engineers to develop chips to that standard, and the means to fab them).

IBM knows that Intel is stumbling, and they know better than to assume that Intel will stumble for very long before recovering. So they're moving aggressively. The contract with Microsoft - which is long since confirmed (not CONFIRMED!!!) - is a major coup in this strategy.

If I recall correctly the Cyrix's chip was not a "true" X86 chip, though compatible. They may have received some of the technical designs from early tech sharing agreements like AMD did. However, and someone will probably correct me here, I think that it was backward engineered. I'm not sure if IBM currently holds an agreement with Intel to manufacture chips that are compatible with the current crop of Pentiums, but I don't think that they do.

It is quite possible that even in the first revision of the 970's the yield at 1.6 was large enough that IBM didn't even bother offering a lower speed chip. If this is the case even if Apple downclocked the chips to make them work in a specific enclosure they would still cost the same as 1.6's. Along the same track of thinking IBM may not offer anything slower than a 1.8 or even a 2.0 in the 970FX's.

Enough of this! Again I quote from the pressrelease:

"Microsoft has licensed leading-edge semiconductor processor technology from IBM"

IBM has no leadning edge x86 technology of any sort.

"the new Xbox technologies will be based on the latest in IBM's family of state-of-the-art processors"

The latest in IBM's family of state-of-the-art processors is the PowerPC 970, or perhaps POWER5. Either way, it's a PowerPC-processor they are talking about. IBM doesn't have any other family of processors.

That could still be a whole host of different chips and clockspeeds. Could as well be a PPC 750xx at 1GHz, for all that is worth.


All this means is that IBM is going to fab them. Ie using their 90nm or 65nm process, with SOI and all that stuff.
If it wasn't for the second sentence you posted, it could as well be a totally proprietary chip nobody has ever seen before.

I keep seeing people argue that the chip will be "much cooler at 65nm". Look at the bloody Prescott core: it's hotter than Northwood, despite the smaller process. smaller dies get hotter too, yaknow. And then you'll have 3 of these beasts in a console, along with high speed ASICs, Memory and GPU?
That's going to require a bloody megaton of cooling, unless they want to make it a watercooled solution. And pardon me, if I doubt that.

Look at the Sony PSX One: that thing also had 3 RISC chips doing all the work, just that they were only clocked at 33MHz or so. I guess the Xbox2 is going to have a similar approach. Probably a special adapted low-cost, low-clockspeed version of the 97x chip, making a massively parallel system. But I don't see a massively parallel system with ultra-high clockspeeds.
That's too expensive, even for MS, to hot, especially for the living-room and too bloody overkill. They'll want to sell a third generation Xbox within 10 years maximum, they're not going to make one now that will last for the next 50 years.

And don't forget, this is not a Sony PS3 idea, where huge CPU power is needed because everything is done in the CPU, even the graphics. The graphics are going to be handled by an ATI chip, the CPU will hardly be stressed doing a little AI and a little data shoveling, let alone 3 of them.
Plus, rumorsites have had an extremely low hit-rate recently

This is only true for the current iteration of the P4. If you look at the 90nm 970FX vs. the 130nm 970, the picture changes.

Where the original 970 used around 51W at 1.8Ghz, the revised chip takes only 24W at 2Ghz. This is because IBM has introduced new techniques (like SOI) to drastically cut leakage current - something intel has not done.

intel and AMD are currently locked into a full-throttle, pedal-to-the-metal death match about Ghz where IBM is much more conservative in order to be able to deliver more balanced chips.

We'll see at least one revision of the 90nm process resulting in power usage cuts before the 65nm chips on strained silicon are going to reduce wattage even further. Combined with more advanced power saving features, maybe a simpler bus interface (non-elastic), different cache layout etc., IBM should be able to reduce power dissipation to about 6 - 10W/core in late 2005. This might, incidently be the reason why MS chose to jump ship - intel would not be able to deliver the same processing power per Watt.

Don't forget further that 970 is based on Power4 (a 125W monster), whereas the future consumer PPCs might be based on Power5 (25 - 40W target).

There's an awful lot of ifs and whens in there.
Experience with 65nm processes is very low, as is any information on strained silicon developments in this area.
Note that Prescott is currently the only chip using it, and it's hotter than Northwood.

Frankly, we don't know enough to make claims about what a future chip might or might not do.
Eitherway, if they want a triple core/chip version running at even nearly desktop CPU speeds, they'll have to resort to massive active cooling, resulting in a even noisiert box than the current Xbox.

I thought Prescott is 90 nm...


Sure, if we talk about processes below 90 nm. But at 90 nm, according to the available information about the new IBM and Intel chips, the situation is quite clear. The IBM PPC design is right now the winner, at least when we talk about power reduction.

Indeed it does but I think it's pretty certain that will be PowerPC based witch was the point I was trying to make.

To me it seems that Intel ****ed up greatly with their 90 nm fab. strained silicon an all, but IBM have given us no reason to doubt them. The excellent thermal performance of 970FX surprised us all. 50-60% reduced power consumption when moving to a new fab was prevoiusly unheard of. Shrinking it further, thus reducing the voltage needed, and applying strained silicon should reduce power consumption even further. By how much, I cannot say.

IBM said that strained silicon wouldn't make any difference at 90 nm while using SOI, but on 65 nm it's another story. Seems to me that IBM were right.

Prescott also has double the number of transistors and a higher transistor density. One of the main reasons the PPC970 saw a decent drop with the die shrink is they didn't increase it's size and the transistor density is in fact lower. Had they done what Intel did and double the transistors it would have been another story.

Is it conceivable that the chip IBM will make for the XBox might be fast enough that the PPC instruction set could be used as a sort of microcode to run x86 software?

Presumably not since VPC performance is so abysmal, but I'd like to hear it from someone with authority.

Oh, and in regard to the "PPC can't run Direct X" dispute (which is an excellent point btw), what about this:

http://www.coderus.com/

After all, it would hardly be like Microsoft to aquire a small struggling company with interesting technology in order to avoid some hard work, would it?

I dont think that they would really need to. Microsoft did build a version of Windows NT for Power PC processors. Windows XP is a derivative of Windows NT as I recall, so part of the work is already done. All they would need to do is update the code neccessary for the games to run, and the effort to do this would probably be a lot less than trying to get the hardware and software "tuned" well enough to run the software through emulation, though that might be a good solution for legacy game compatability if the hardware is fast enough to match the Xbox 1 in performance under emulation.

I really hope the linux community will crack the xbox 2... this kinda setup would be great for a desktop. Cheap too.

(on top of that you could run MOL)

In any case everybody seems to assume that these processors are going to the same ones that go into Mac's. They are not. This will be a different processor than you will be finding in your 2004/05/, or /06 Macintosh.

It is, but if you read my post again, you'll see that I was referring to strained silicon, not 90nm. ::/

As for Mac DX: afaik that is merely a translation layer, not a DX implementation for the Mac.
Plus it seems to be OS 9 only.

As for NT PPC:
The last and only version of Windows NT for PPC was version 3.5. It was never released to the broad public either, only a few select people ever got their hands on it.
Going from a 3.5 codebase to a XP codebase is almost rewriting everything from scratch, plus the people who developed the PPC version probably no longer work for MS or have lost all training in PPC programming. Not even talking about 64bit PPC programming for a G5.

The good thing about it: if they port DX to PPC, chances are somewhat higher the Mac will eventually also get it, which would be a terrific chance for having more games come out for the Mac.
I just don't see MS maintaining a DX build for 2 different hardware platforms all along.
In fact I could number a series of things that speak against moving away from x86, but aparently they already made that decision, so there's no point arguing about that anymore.

Where I see no problem is Xbox 1 backwards compatibility:
Assuming they get soem of VPC technology to run on such a 3 core setup, they really only have to find a way to tape into the graphics accelleration provided by the ATI chip and their problems will be solved.
My Dual 1.25GHz G4 already emulates a Pentium MMX 667Mhz+ under VPC 5. I really don't see any problems in a 3x2GHz, as some think it's going to be, emulating a meager P3 733, IF it doesn't have to do the graphics stuff.

This could also turn out to be a good development for VPC for the Mac. (graphics accelleration anyone?).

Still, I won't believe any of this "stellar-specs-hype" until I see some official press releases on it.

Did you even follow the link I posted? Click it and the very first thing you notice is a prominent "X" icon in the top right corner. Look at the list of "Why MacDX" bullet points and you will find "Seamlessly Supports Mac OS 9.x
and Mac OS X".

I have to say, I hope you researched your other claims a little more deeply.

ok, I overlooked that. Still, it's a translation layer.

Geez, I haven't seen so many silly arguments since Motorola was supposed to ship the G5 in 2001.

Microsoft owns the source code to DirectX and they employ thousands of software engineers. You mean to tell me that they can't port it to PowerPC with months (or years) of development time?

IBM is designing PowerPC. Period. Microsoft is not tied to x86, and currently Intel & AMD are so focused on the PC/workstation/server markets that they have nothing appropriate for the game console market. The processor in the existing XBox wasn't particularly appropriate!

IBM has been making all sorts of noise about their "we'll help you design your own flavour of PPC and then fab it for you" program for a couple of years now. They did it for Nintendo. They are working with Sony & Toshiba. Why is it so hard to believe that Microsoft wouldn't take advantage of this? Not doing so can be seen as a disadvantage for them.

Do you really think Microsoft cares about backward compatibility? If you've got XBox games already then you've already got an XBox. If you don't, they'd rather sell you the new ones (and believe me, you'd rather have the new ones). And if there is an XBox game you really must have, don't you think that MS would rather sell you both systems? Most games only sell for the first 6 months they're on the shelves for anyhow. If MS wants to build the right machine for their target market, they will do so and then decide if it is possible (and desirable) to make it BC.

Sony cared about backward compatibility when they released the PlayStation 2, and Nintendo did with the GameBoy Advanced. Backward compatility makes initial cost of ownership in the upgrade more palitable for the consumer as well as bragging rights for more games available for the new system, therefore it is a good marketing stragegy. I would be willing to bet that the average consumer still purchased 2 or more games when they upgraded from a PS 1 to a PS 2, so Sony made just as much money off of the initial purchase as they did if they didnt have a system that was backward compatible.

I don't doubt they CAN, I doubt they want to.
Frankly, it would be like a second platform they'd have to maintain, unless they wanted to make it a 100% closed system with no way to update or implement new stuff. Actually looking at it from this perspective, it makes sense.

Still, it makes more sense if you think about IBM making a custom chip for them, not a 976 based one. Custom chip would mean closed system that nobody would hack a linux distro for and that nobody would use to make a true PPC DirectX of, to use it, for example on Mac OS X or Linux PPC. (Maybe I'm totally wrong when I think that Microsoft has absolutely no interest in bringing DX to either of these platforms and will do whatever they can do prevent that from happening).

So, assuming they're aiming for a custom, proprietary closed system platform, why does everyone think they're going to use the 976 chip that is going to have so many cores it's going to outperform a small-sized serverfarm?
It might as well be a very custom chip-fabric similar to the earth simulator, just several thousand times smaller and slower.

That would make tons more sense than having it based off 3 enterprise level high-end server CPUs working in tridem, causing the machine to melt.

basically what I'm trying to argue about is that this "article" is the wet dream of some geeky journalist, while the reality is going to be entirely different, such as pictured above.

I don't doubt MS is going with IBM or moving to PPC. I said I see several things that would have spoken against it, but aparently the egg-heads at MS beg to differ. Probably for a good reason we don't know yet.

Actually we know almost nothing anyway:
-IBM is going to fab a chip of their own design for MS
-ATI is going to fab a chip of their own design for MS
-Xbox2 will be more performant than Xbox1

That's about all we know for sure.
Now make of that whatever you want. Some will make it a 3 core, SMT enabled monster PPC 976 cluster with ATI R500 core and the option to bake bread while playing, luring millions of readers to their website, while others might take a more traditional (and imho sensible) approach and look more towards a mix of a PS1-Gamecube design with custom PPC chips working together with a custom ATI chip to provide excellent performance for specific uses (which is what games are all about). I'm sorry if I'm the only person not seeing a 3 core SMT PPC 976 solution as a "specific use solution". A brute force approach, rather, uneconomic by definition, expensive and overkill.

And everyone who thinks MS is going to spend an extra 1000$ on an Xbox2, just to satisfy the customer, hoping to make up for it in game sales, simply has no clue, really.

Sorry, no. My Windows NT 4.0 CD includes i386, PPC, Alpha, and MIPS, and it is not some special select release.

Programmer,

Just a random question for you...

I started a thread about this in the Software Forum but I dunno if you make it over there as much as here.. anyway... In the Merc news article I saw the following quote: "Internally, Microsoft has begun developing game prototypes, and it is using G5 systems to do so.". IF this was true... could anything be extrapolated about the quantity of QUALITY games that might be made available for the G5/OS X platform in the future?

I just thought that IF (a big IF I know) that statement is true then some of the same games that were made for the XBOX2 might be MORE EASILY be made to run on a G5 box too.. At least more easily than any current XBOX game.

Or am I just kidding myself... Something that most Mac users are pretty go at. (including me)

Dave

So that'll be a "no" then

It is 90nm, but Prescott is a different beast. It has over 100M transistors instead of ~50M in a regular Northwood because of the extra on-die cache and a few other design changes.

If MS were using Apple G5 machines to develop game prototypes on (big IF), they would most likely not be running MacOS X. Even if they were MS wouldn't take the games beyond the prototype stage on MacOS X and you'd never see any sign of them on the Mac platform. Sorry.

I think you're under a serious misapprehension of what a 3 core 9xx-based PPC on a 65nm process is likely to cost next year. And remember that nVidia shipped their leading edge nv25 on the XBox before an equivalent was available on the PC.

Tell you what -- let's just wait and see.

Interesting, that is total news to me.
Anyone have any info on that? will that only run on IBM Systems (RS6000 etc) or will it also install and run on a Mac?
Because if the later is true, that is really absolute news to me (and I bet several others too).
Ever bothered to install that build?

As for Programmer:

things may be much cheaper in a year, but I doubt it's going to be as cheap as YOU seem to think.
let's assume a 50%, maybe 60% price decline over the next 12 months.
Let's also assume a 2GHz chip costs around 200 maybe 250$ in quantities of 1000. Let's assume a triple core chip with all the interconnects etc costs around three times as much.
Now subtract 60% of the price: that's still a 250$ chip, GPU not included, RAM not included, connectivity not included, case, cooling, software, PSU etc not included.
I don't see that Xbox2 coming in at less than 400$ and that is a VERY steep pricetag for a console, especially in 1 year. And that is with very optimistic assumptions.

Waiting and seeing is what I've been proposing all the time now.

I was running it (NT4) on a PPC by Motorola a few years back. It would not install on a Mac, but that's just a minor issue. If MS wanted it to, they could get it to fairly quickly.

I think you're misunderstanding the general idea. I don't think anyone necessarily expects MS to port Xbox2 games to the Mac, but other companies that make Xbox2 games could.

No, the only people that will be doing XBox2 work right now are MS people. By the time 3rd party developers start MS will have a fully Windows-based solution which communicates with a dedicated developer box. No Macs involved. Some of the work would slightly help bring product to the Mac, but that's true of the GameCube as well and its such a trivial improvement that it doesn't actually matter in practice.

Okay, but you might want to take into account that I'm the one under NDA.





(bad assumption that a 3 core chip would cost 3 times as much, by the way)