The PowerPC 980 Roadmap

[jettredmont]

Originally posted by yoman
I just read at

http://www.macminute.com/

That intel is planning to hit 4Ghz in 2004. I wonder if IBM will be able to catch up.

Given that a dual-2.0GHz G5 smokes 3.2GHz P4's by most estimations, I imagine a 3.0GHz G5 (with 1.5GHz FSB) will likewise smoke a 4.0GHz P4.

Will Intel ever deliver the P5? That might change the equation slightly, as the P5 is supposedly more efficient (instructions per cycle in real use) than the P4.

 

[Rocketman]

Originally posted by MrMacman
:
Wake me up when they get a Better name like...
The G-006
the G-007!

Processor improvements are good.

Multiple replies

Why G5/G6?

Same reason why they used to have a G3/G4. A commodity line and a premium line.

As each generation is released, the prior one is phased out. G5 kills G3 and G6 kills G4 (Motorola). As we all know Apple/Jobs and Apple's customers themselves are in a BIG HURRY to dump Motorola.

G6 is the "dump Motorola" line. Everyone is in a huge hurry for it.


I stated early that the G5 would be about 2.6 Ghz in Jan-Mar 04. I stated early that the G6 would be what was referred to by Jobs as the 3 Ghz late 04 release.

That much is fairly clear from public statements, IBM releases and chip industry events. What is particularly interesting is that some of the "guidance" that enterprise needs to know what to expect in the fututre, is now being released by Apple in an uncharacteristic way. If the information flow becomes even incrementally more about "future products" then Enterprise will have the comfort level needed for adoption of the product. They also need a ramp-up in real service with the sale, and charging fees for that is just fine.

The G5 should really have just as long of legs as the G4 has, or more. That gives it a multi-year life cycle. The G6 when it is released frankly will require more expensive associated components to really give it the resources it can feed. So we can expect maybe even a higher price point than the G5 at release.

For example the G5 was released as a 64 bit clean 32 bit system. Deja vu.

The G6 is likely to be primarilly 64 bit with good 32 bit backwards compatibility.


Overall this is an astonishing series of events as compared to the past several years. This is beginning to realize the actual early promise of the PowerPC.

What I am hoping is the G6/980 is more scalable to 4x and 8x processor systems so true compute power can fit in 3u or 6u deskside computer ala some of the IBM AS/300's and such. Remember the supersize Apple server?

This will also be a server monster and a cluster and grid monster.

When Internet 2 becomes more reality in 5 years or so, true grid computing (WAN) will become practical.


And who says a processor (G6/980) that was sampling in October 2003 would NOT be released in some astronomical Apple system in March 2004?

Rocketman

 

[ITR 81]

Originally posted by tortoise
Actually, you ARE mistaken. While they licensed the MIPS ISAs (not core designs), it was for their other processor lines and unrelated to their desktop computer business. AMD has been making high-performance embedded RISC CPUs for far longer than they have been dabbling in the x86 market, and for a number of different popular ISAs and some original ones (like the AMD 29k RISC processors). High performance embedded systems CPUs based on other ISAs is still a major line of business for them. MIPS32 and MIPS64 are popular embedded target ISAs, and licensing those ISAs supports their embedded CPU business.

Their earlier x86 processors were an IA32 ISA on an AMD core. The Opteron (and related) are an AMD64 ISA on an AMD core, with support for the IA32 ISA. There is no MIPS technology in their consumer CPU line. AMD has a lot of experience designing cores and designing ISAs, with the added experience of making many different popular ISAs run on their cores. AMD CPU cores have a long history that predates their x86 offerings and which were designed by AMD. The only thing special about the AMD64 is that this may be the first time they have marketed their own ISA design since the 29k-series of embedded processors.

AMD bought the license in April 29, 2002 just right after they bought Achelmy from MIPS to compliment it which was bought on Feb. 2002. At that time AMD and MIPS said it was to be used for embedded processors but AMD also stated it could be used in there entire processor line. That being said everyone assumed back in 2002 the AMD 64 was going have instruction sets from this license. I was gone fighting a war for over a yr so I guess I might have missed some of the last details...and yes I know about AMD's history on RISC as my college had old AMD Nx586 machine we use to play with.

 

[ITR 81]

Tech Tv also reported today of the news of the 980 90Nm processor coming in March of next yr. at 3GHz.

I wonder if the rumor is just getting around or if news is leaking because of someone from Apple or IBM.

 

[rog]

This is great news. Just imagine a year from now, every mac at least 2 GHz and a G5 or better! Keep it coming IBM!

Anyone know of the other improvements in the 980 other than clockspeed? Bigger L2 perhaps?

 

[jettredmont]

Brilliant Marketing Strategy

Apple will rebrand the Motorolla 74xx chips as "G5-Ultra" and the IBM 970 chips as "G5-Ti". Maybe those old 750FX chips IBM still has laying around would sell better if they were "G5-MX" instead?

That way, all the PowerBook users can feel like they're getting a cool new processor ...

Hey, works for nVidia and ATI ...


sounds kinda off-topic, but the point is this: Apple stands by its trademarks and brand names a little more than most companies. I definitely don't see Apple going to G6 soon, not because they don't think the public can figure out that "G6" is kinda like "G5" except "one more", but because when you give the marketplace a whirling target of increasing numbers all they see is a blur, which destroys branding.

Customers haven't even had a full year of G5-desire yet. Too early to bring out the G6.

 

[tortoise]

Originally posted by rog
Anyone know of the other improvements in the 980 other than clockspeed? Bigger L2 perhaps?

I'm hoping for a very slick on-chip memory controller. That would rock, and close a couple gaps.

 

[yamabushi]

The 980 will likely be able to support a huge amount of L2, like the Power5. However the actual amount of L2 in G6 chips for Apple will probably be quite modest.

 

[roy_g_biv]

i'm not sure i can see apple moving to the G6 so soon...

i remember some readers here disputed branding the 970 as the "G5", given it was not a "5th generation processor" like the G4 was fourth generation powerpc. keeping the G5 moniker for the 980 - the POWER5 derivative - just sounds like it makes sense.

and i don't think a shiny new "G6" is needed to market apple computers. as long as the G5 brand remains competitive with intel or AMD processors in speed and scalability, i don't think consumers will care too much about what the actual G-designation is used. (nor do i think a G5/6 line, or a G5/4 line need be set in stone) maybe apple can take the 90nm 970, stick it in an imac and call it a G-something. G-anything, i don't even care... maybe something funny like a "G-sumer", or maybe a "G-Unit" and target that 18-34 demographic.

but then again, in the end it's just fun to speculate.

 

[jaedreth]

G5 / G6

Steve said the *G5* will reach 3GHz by the end of next year. Not the G6. Keep in mind, the term G5 is Apple's. Sure, it's the next chip design, but it is not a completely new architecture, it's just the sequel.

So this whole talk of "moving to G6" is silly, and assumptive. It's the same chip despite whatever name. Besides, we want the G5 to show itself as the best out there.

I wouldn't expect Apple to change chip generation monikers until at least Power7, unless IBM comes out with some new technology that redefines all the rules before then. (Which they may.)

Jaedreth

 

[~Shard~]

Re: G5 / G6

Originally posted by jaedreth
Steve said the *G5* will reach 3GHz by the end of next year. Not the G6. Keep in mind, the term G5 is Apple's. Sure, it's the next chip design, but it is not a completely new architecture, it's just the sequel.

So this whole talk of "moving to G6" is silly, and assumptive. It's the same chip despite whatever name. Besides, we want the G5 to show itself as the best out there.

I wouldn't expect Apple to change chip generation monikers until at least Power7, unless IBM comes out with some new technology that redefines all the rules before then. (Which they may.)

Jaedreth

I agree - until IBM comes out with some revolutionary or noticable advanced architecture, I don't think we will see the "G6" name applied to any machines. I think there is a lot of potential with the 970s and 980s, and the addvancements will be relatively quikc (compared to Motorola at least!), so I think staying with the G5 for the next year or 2 is perfectly acceptable. I hope Apple waits until IBM breaks completely new ground with a chip that will deserve to be called the G6.

 

[Gyroscope]

Yeah ,they ll stick to 970 as long as they can milk your money from it. Chip development doesn't come cheap so why would they jump to new design when 970 hasn't been around for more than 4 months or so, to pay for itself. Besides there is no official confirmation about 980??(whatever) existence as of yet.

Sweet dreams.

 

[suzerain]

simple numbers

About the Intel 4 Ghz thing. If they can get their chips up to that, great. I'm all for speed, anywhere.

But, they are currently at 3.2 Ghz, and the 970 is at 2.0 Ghz. Let's assume the P4 is faster at some things, and the 970 is faster at other things. But, ultimately, these two chips seem to me to be about even (in reality, the 970 is better in FP ops, and the P4 is better on INT ops...but for simplicity, let's say they are even for now).

So, we have this breakdown of "performance ratings"

P4 @ 3.2 Ghz: 100
970 @ 2.0 Ghz: 100

This translates into a "Ghz efficiency rating" of 31.25 for the P4, and 50 for the 970. In other words, the 970 is approximately 60% more efficient per clock!

If, one year from now, the P4 is at 4.0 and the 970 is at 3.0...well then the same "Ghz efficiency rating" ought to translate into these performance scores:

P4 @ 4.0 Ghz: 125
970 @ 3.0 Ghz: 150

Or, that means at next year's speeds, we have a G5 that is 20% faster than the then-current P4.

Now, go further...

Say instead it is a 980 (G6) and the AppleInsider article is accurate (speed increases of up to 40% per clock over the 970).

Then, the comparison becomes:

P4 @ 4.0 Ghz: 125
G6 @ 3.0 Ghz: 210 (max...probably less)

Now, there are two things to keep in mind. It is possible that the 4.0 Ghz speed Intel is targeting is not for the P4, but for the Prescott chip, which presumably will have much greater efficiency per clock than the P4.

Secondly, AMD's Opteron is really where the performance edge in the PC world is right now, even though Intel has all the mindshare among consumers. So, obviously, for now Apple and IBM ought to be competing with AMD until Intel unveils its "Next Big Thing" (i.e., Prescott).

But, when you start to think about a 3.0 Ghz 980 potentially outperforming a 4.0 Ghz P4 by 60% (!), noting that it is capable of dual-processor configurations, it really makes you feel optimistic about 5.0 Ghz G6s, which would perform with the speed of a P4 running at...are you ready...11 Ghz!!! Is intel going to get to 11 Ghz by early 2006?

Gee, somehow I doubt it. Of course, Prescott will change all this...whenever it comes.

 

[Fukui]

Originally posted by tortoise
The AMD cores are very well engineered and are also somewhat more "general purpose" in design than PPC cores, which tend to have an emphasis on numerical codes that makes them less optimal for many tasks than the AMD core.

What are "Numerical Codes" ?

 

[jiggie2g]

What I think Apple will most likely do is use these 2 processors 970/980
to be completely independent of those bums at Motorolla (who Cut Apples Legs just as we were leading the Speedrace)Labeling the 980 the "G6" and using it in it's Pro Market(Power Mac and Eventually Power Book) . while the G5/970 becomes a Consumer chip (iMac 1st then/eMac and Eventaully ibook). having these 2 Chips Gives Apple Flexibility so they won't have 2 cripple the consumers machines worring about them eating into Power Mac/Power Book sales. and the best part is that we will Finally Be Equal to or better then Performance wise to windoze machines at Both High and Lowend Markets. Intel knows if IBM can get the G Series Chips within 500mhz or less of thier precious Pentium it will be game over apple will start eating into the windoze market share then they'll be the ones talking about "the MegahertzMyth" . I my self am Waithing for a 2GHZ G5 20in iMac ...maybe b4 Next Christmas...

 

[jiggie2g]

Well The G6 is every bit as Different to the G5 as the G3/G4 were to one another . Built on 90nn Fab ,PowerTune(making it speed Scaleable to improve batterlife and power consumption) Improved Altivec , Dual Pipelines ,Independent Bus for Each processor in Dual models(Crossbar Switch) , DDR 2 , so Different Technology . as fot the wait between the G3and G4 you have to remember that at that time 1997-1999 we were still Head Perfomance wise , if anyone remember the days when G3's were 3x the Perfomance of a P2 at equal clock speeds forcing Intel to Bring the P3 out so quickly. except now the situation is reversed now Intel is the Leader and Apples Market Share has tuned to Crap as a Result of Motorola's Incompetence being a Major Factor .so Yeah I do expect to see a G6 by summer or 3rd Quater 2004,we are the ones who are behind and have 2 play catch up. the G5 put us neck and Neck with Intels Best but only on the Highend Market, now all the other Mac lines will have 2 follow suit. Jobs will not be satisfied until he can take on the PC on all ground and equal footing. that way atleast apple has a Chance to get out the 3% market rut thier in. either way it will sure as hell be fun to Watch how it goes down, Let the Speed Race begin...Again..lol P.S. Sorry for the Double Post

 

[Kamu-San]

PPC PC CLone?

What I'd like is a cheap, PC-technology based clone with, instead of a Pentium, a PPC 970.
I really like Apple stuff and I'll buy it just for OS X, but for a lot of things Linux is sufficient and Apple hardware is too expensive to just run Linux.

I'd love to have a cheap PPC970 based Linux-box at home.

 

[Eric_Z]

Re: Crossbar Switch

Originally posted by ennerseed
Appleinsider also reports the new machines may include a Crossbar Switch: "and will use a crossbar switch to speed communication between processor"

I believe (don't know what else it could be) this is like SGI's implementation of a Crossbar Switch which could bring major speed improvements.

Or perhaps something like this [MAI Logic, se page 9-10 and 18-19]?

 

[Analog Kid]

I can't believe how much bandwidth goes into discussing when Apple will change their labels to say G6...

 

[ktlx]

Re: simple numbers

Originally posted by suzerain
About the Intel 4 Ghz thing. If they can get their chips up to that, great. I'm all for speed, anywhere.

But, they are currently at 3.2 Ghz, and the 970 is at 2.0 Ghz. Let's assume the P4 is faster at some things, and the 970 is faster at other things. But, ultimately, these two chips seem to me to be about even (in reality, the 970 is better in FP ops, and the P4 is better on INT ops...but for simplicity, let's say they are even for now).

Since both processors need to be reimplemented at 90nm in order to reach those processor speeds, your comparison is not valid in really any way.

When the processors are reimplemented at 90nm, each design team will make changes to the processor. Who knows what each design team will do? Surely no one on this forum who is at liberty to speak.

For example, let's say Intel decides to just do a shrink with no significant design changes. Let's also say that IBM decides to add SMT and 1MB of Level 2 cache. Then the comparison will be skewed toward favoring a 3Ghz PowerPC.

The converse would also be true even though many on this forum are acting as if IBM will add all sorts of magical, wonderful features that will make its processor ideal at 90nm while Intel will do nothing more than reimplement its existing Pentium 4 at 90nm.

 

[geerlingguy]

Originally posted by g30ffr3y
i dont even have a g5 yet... they cant put out a g6... holy crap!!!

If you remember, the G5 was being mentioned about two years ago (just not quite as definite). About a year ago, more definite rumors were coming out, then it was released to the public a few months ago. I'd say we won't see "G6" until 2005.

Hopefully I'll be wrong.

 

[DrBoar]

The IBM 970 is not a stop gap solution any more than the 604 or G3 was.

Thoughout the Macintosh history starting with the 6800 in 1994 every leading edge cpu -020-040, 601, 604, G3 was replaced in about 2 years time. The G3 replaced the 604 in November 1997 and was replaced by the G4 in August 1999.

The freak here is the G4 that lead the race for 4 years ( I am using the term "leading" very loosly here). We should have had a G5 CPU back in 2001 instead of the Quicksilver G4. So there is big demand to replace the old old G4 so I wellcome the 970 and the 980 and any other IBM CPU that will them go faster

 

[psurrena]

cisc/risc

I'm by no means an expert on chips but shouldn't CISC chips be nearing their end as processors grow from 32 to 64 to 128?

As far as I know and please correct me if I'm wrong:

RISC will process a collection of data as close to 64bits at once. So it will collect 8 pieces of 8bit info, one 32bit and four 8bits, one 64bit and so on.

CISC can only process one 8bit, one 32 bit, one 64bit at a time.

SO! As processors move from 64 to 128 an so on, CISC chips with become overwelmingly complex. While I doubt we will move to 128 in our current form (cisc/risc) I think that 64bit will provide a challenge for Intel/AMD while IBM will finally be in a situation where they can move forward without too much sweat.

 

[Nemesis]

Re: G5 or G6? Also, Ars Technica article

Originally posted by illumin8
Illumin8, tell me more about your 15" AluBook. Is it good? Fast enough? Robust? Any special comments about it? Because I wish to buy one for myself ...

 

[Rincewind42]

Re: cisc/risc

Originally posted by psurrena
I'm by no means an expert on chips but shouldn't CISC chips be nearing their end as processors grow from 32 to 64 to 128?

As far as I know and please correct me if I'm wrong:

RISC will process a collection of data as close to 64bits at once. So it will collect 8 pieces of 8bit info, one 32bit and four 8bits, one 64bit and so on.

CISC can only process one 8bit, one 32 bit, one 64bit at a time.

SO! As processors move from 64 to 128 an so on, CISC chips with become overwelmingly complex. While I doubt we will move to 128 in our current form (cisc/risc) I think that 64bit will provide a challenge for Intel/AMD while IBM will finally be in a situation where they can move forward without too much sweat.

Hopefully whoever gave you that explanation of the whole thing will never say that to someone else .

First, CISC & RISC are largely academic descriptions today. There are no modern chips that are truly CISC or RISC. x86 flies a little closer to CISC and PPC closer to RISC simply due to their design heritages, and even then the x86 only looks like a CISC design at the ISA level. Basically Intel put an x86 translator on top of a RISC core back in the Pentium line and declared a victory for CISC that they have been trying to get away from ever since (hence the doomed Itanium).

In the end, there is no true CISC or RISC anymore. Traditional CISC chips have been taking on traits from RISC designs to get more speed, and RISC chips have been gaining ever larger instruction sets in order to do more advanced things. RISC has increasingly become less about instruction set size, and more about instruction set complexity - it's about the only distinguishing factor between CISC & RISC chips. No matter how large the instruction set gets on a RISC chip, each instruction tends to do relatively little and use very few resources. On a CISC chip the large resource hogging instructions that it accepts may be broken down into simpler less intensive instructions on the fly to be fed into a RISC-like core.