PPC976: Possible Xbox2 processor?

[bbot]

I was checking out the macosrumors.com site when I read an article dated 1/18 about future versions of PowerPC. The article listed several processors, but one stuck out:

"*PowerPC 976 - POWER5 Dual Core, 65nm SSOI (Strained Silicon On Insulator) process, VMX2 instructions, 4GHz+"

Which seemed to perfectly fit teamxbox.com's description of Xbox2's processor. What do you think?

 

[London Geezer]

What do you think?

I think we already have LOADS of threads based on rumours and pure speculations, most of which get locked when someone thinks his dreams are better than someone else's dreams...

I also think we should just leave this speculation-mood for a while and get on with the little info we have or the consoles we have at the moment, to then return talking about next generation hardware when we actually know something more about them.

For God's sake, we have a new "oh look this might be in [PUT NEXTGEN CONSOLE OF CHOICE HERE]!!" thread every day, with subsequent speculation-based-flame-war... I mean, we already have flame wars on actual facts, i don't think we need any more bitching on thoughts and opinions, which are diverging by nature.

 

[hey69]

i know one thing. a 4+Ghz procesor is gonna be WAY to much expensive to put in a console , even by end 2005 -begin 2006

edit: 4ghz+ Powerpc procesor that is

 

[Paul]

4Ghz+? And people think Broadband Engine at 4Ghz is going to be imposible?...

 

[Evil_Cloud]

Why not 4Ghz? Soon 4Ghz will replace 3,2Ghz, do I don't think it's impossible...

 

[Paul]

Can't compare clock speed on different archiectures, X86 vs PowerPC.

 

[London Geezer]

See, that's why i said this and many other threads are just useless.

 

[gurgi]

Yeah, I say we argue about what is or isn't a valid thread instead.

 

[cthellis42]

Well, if timetables are stuck to, the 970's used in G5's should be hitting 3ghz by summer 2004 (Jobs' announcement of getting there "within a year"). They haven't moved yet, but hopefully they're going to speed-bump the G5's soon, and we'll better be able to judge from there. (Presumably the whole line will move to 90nm chips like the Xserves as well.) If they can get to 3ghz by mid-2004 then it might also be possible to get to 4ghz by mid-2005 with a chip revision and an even smaller process, but we're talking utterly bleeding edge here. What's the price per chip for PPC970's right now? Opterons? How on earth would that fit into a console, unless Microsoft is looking for an efficient process to dump their cash reserves?

 

[Brimstone]

I was checking out the macosrumors.com site when I read an article dated 1/18 about future versions of PowerPC. The article listed several processors, but one stuck out:

"*PowerPC 976 - POWER5 Dual Core, 65nm SSOI (Strained Silicon On Insulator) process, VMX2 instructions, 4GHz+"

Which seemed to perfectly fit teamxbox.com's description of Xbox2's processor. What do you think?

Seems possible, but at this point a lot of things seem possible. But why say multicore when that is a dual core? I guess it comes down to an individuals interpetation of syntax.

Also if Microsoft is making a switch from x86 to something else, wouldn't they want to make the change to the latest and greatest architecture?

 

[cthellis42]

Also if Microsoft is making a switch from x86 to something else, wouldn't they want to make the change to the latest and greatest architecture?

They would, but cost is always the balancing point. It's not like there's an unlimited budget or the ability to blow one's wad just on the CPU in assured multi-million consumer devices with guaranteed huge competition.

 

[Brimstone]

Also if Microsoft is making a switch from x86 to something else, wouldn't they want to make the change to the latest and greatest architecture?

They would, but cost is always the balancing point. It's not like there's an unlimited budget or the ability to blow one's wad just on the CPU in assured multi-million consumer devices with guaranteed huge competition.

IBM has to get a return on their investment also in CELL and the Fishkill Fab. Proliferating CELL into as many markets as possible is important if they ever really want to challenge x86. PowerPC didn't fair so well against x86, I don't think IBM wants the same thing to repeat again with CELL.

 

[Deadmeat]

...

"*PowerPC 976 - POWER5 Dual Core, 65nm SSOI (Strained Silicon On Insulator) process, VMX2 instructions, 4GHz+"

I don't think this is it for Xbox Next. Xbox has no use for Altivec and MS would like to skip that and cut some cost.

4Ghz+? And people think Broadband Engine at 4Ghz is going to be imposible?...

Massive heat generation from 16 APUs and short pipe design makes it difficult for EE3 to clock high.

 

[Panajev2001a]

No use of Altivec ?

Are you going to run Physics completely on the GPU or convince me that flexible SIMD engines like Altivec are no use for game physics ?

 

[Guden Oden]

Re: ...

Xbox has no use for Altivec and MS would like to skip that and cut some cost.

Uh, right, just as current XBox has no use for SSE? I'd like to hear your reasoning for making such a statement, if you don't mind.

I think most programmers in particular would agree on SIMD instructions being very useful in a processor, for a variety of tasks, physics and AI being just some of them.

Btw, I see you declined my invitation to list your credentials in an earlier thread. You just keep on making what appears to be rather ignorant posts...

Massive heat generation from 16 APUs and short pipe design makes it difficult for EE3 to clock high.

As the exact make-up of the EE - including number of APUs and pipeline length of said APUs - has yet to be revealed, I wonder how you can sound so confident in your predictions. One might note however, that if the pipelines are short and thus not highly clocked, 'massive' heat generation from the chip seems unlikely. Especially if the thing is fabricated with SOI technology, which I do believe it will be.

Care to clear up this apparent contradiction?

 

[cthellis42]

IBM has to get a return on their investment also in CELL and the Fishkill Fab. Proliferating CELL into as many markets as possible is important if they ever really want to challenge x86. PowerPC didn't fair so well against x86, I don't think IBM wants the same thing to repeat again with CELL.

Still not sure if IBM can produce whatever of CELL it wants for anyone, or if the S/T/I co-operation lets them produce for their own purposes but not farm out. (Or restricts production for certain companies, or perhaps reached an agreement to not conflict in certain areas, such as Sony not producing for the server market while IBM wouldn't for a Playstation competitor... <shrugs> Can never tell.)

Meanwhile, PPC970 certainly fairs well against x86 counterparts. Old PowerPC chips certainly slowed, but then Motorola is no longer involved (with G5's at least), for which everyone is glad. Hehe... At least it compares well performance-wise. Sales-wise for entirely different reasons. (And even so, a whole ton people don't think about sold in the GameCube, and a whole ton more would sell with the Xbox2 and the N5 if they are indeed going that way...)

 

[Deadmeat]

...

Are you going to run Physics completely on the GPU or convince me that flexible SIMD engines like Altivec are no use for game physics ?

Power5 already has an excellent FPU and doesn't need the asm coding of Altivec to see a speed gain. This is why IBM doesn't do vector SIMD in its line of supercomputers.

Uh, right, just as current XBox has no use for SSE? I'd like to hear your reasoning for making such a statement, if you don't mind.

How many Xbox games do you know that makes an extensive use of SSE?

As the exact make-up of the EE - including number of APUs and pipeline length of said APUs - has yet to be revealed, I wonder how you can sound so confident in your predictions.

1. Large number of APUs require an APU implementation using bare minimum transistor count.
2. Limited transistor budget means a short pipe design.
3. So EE3 can't clock high because of above two combinatins.

Let's say each APU burns 3 watts @ 1 Ghz, a very favorable figure. How much is the total power consumption if all 16 APUs are active? 48 watts. Add the power consumption for two PPC440 + other circuits, and you are looking at a power consumption figure of 65~70 watts at mere 1 Ghz.

One might note however, that if the pipelines are short and thus not highly clocked, 'massive' heat generation from the chip seems unlikely.

Massive heat generation comes from the large number of APUs, not high clockspeed. CELL simply can't clock high as Pentium4s and superpipelined PPCs.

Still not sure if IBM can produce whatever of CELL it wants for anyone, or if the S/T/I co-operation lets them produce for their own purposes but not farm out.

IBM has no internal use for CELL. It is strictly a "We will license our design to SCEI for money" deal".

 

[Panajev2001a]

POWER5 has more than just a single FPU and in a game console resources would be better utilized if you had an Altivec-like unit that could help with Physics, A.I. and Vector Math for 3D calculations.

Knowing that for physics in games the need of DP FP is not there yet, the use of a SP FP Vector Unit is the most transistor efficient design.

Double core PPC970+ and SMT would make the most sense compared to a single POWER5 CPU.

Dual core POWER5 would mean 4 CPUs on the same chip and each would not be that small either: I thought you had somethign against machines with more than 1-2 CPUs

I see you keep going with your "I know what IBM wants to do with CELL" and your other theories Deadmeat: the interesting thing ( well not a surprise ) is that actually IBM has much more interest in CELL itself than "license it to Sony and Toshiba and collect royaltes (which is not how the joint-venture was set-up anyways )".

 

[Deadmeat]

...

POWER5 has more than just a single FPU and in a game console resources would be better utilized if you had an Altivec-like unit that could help with Physics, A.I. and Vector Math for 3D calculations.

The difference is that Power5 FPU can be coded in C++ and auto-optimized by compiler, whereas all vector units are asm stuffs.

Dual core POWER5 would mean 4 CPUs on the same chip

It is actually two. Looks like four because of SMT.

I thought you had somethign against machines with more than 1-2 CPUs

And Xbox Next has just two physical CPU cores. You can still obtain a good performance without multithreading on Xbox Next because most of OS services are multithreaded.

I see you keep going with your "I know what IBM wants to do with CELL" and your other theories Deadmeat: the interesting thing ( well not a surprise ) is that actually IBM has much more interest in CELL itself than "license it to Sony and Toshiba and collect royaltes (which is not how the joint-venture was set-up anyways )".

What uses would IBM have of CELL? CELL mainframe? CELL AS/400? CELL rackservers?? Get real.

 

[Vince]

Re: ...

Dual core POWER5 would mean 4 CPUs on the same chip

It is actually two. Looks like four because of SMT

Uhh, no? Power5 itself is based heavily in Power4's microarchitecture with the entire execution pipeline transfered over IIRC. It retains it's dual cores and adds SMT to each. All-in-all, Power5 is considered by even IBM as a lower-end server IC as compared relativly with Power4. It has some cool power saving functionality, preliminary Fast Track (which is useless in a console) and such, but it's definitly not the leap Power6 promises to be.

Which is curious as around the time Power5 development began in 2001, there was alot of IBM senior engineers moving to that other project. Although I'd hope they'll bring more than this to the table, atleast the speculation is moving in the right direction in preformance from the initial talk of Intel's IC.

What uses would IBM have of CELL? CELL mainframe? CELL AS/400? CELL rackservers?? Get real.

Actually, didn't Mfa post a publication from IBM which stated they intend to use Cell derivatives in their low-end servers a while back?

It wouldn't be surprising really, given the trend toward using low-cost common commodity IC's in servers. Hell, you of all people being the master of applying Blue Gene to every-frickin' thing possible should realize that this concept of low-cost IC's en masse isn't exactly alien to IBM. The stuff such as NCSA-UIUC's PS2 cluster and the imminent move towards high-end visualization using nVidia and ATI ICs only furthers this ideology.