IBM Gobi, Mojave, and PowerPC 970 and Beyond

[MacRumors]

Background

In December 2002, ThinkSecret first reported on faster PowerPC G3 processors in the works -- these high speed chips were said to provide speeds in excess of 1GHz, and were codenamed "Gobi". Beyond this, few other details were provided.

We heard again about Gobi in January 2003, when MacEdition reported that the Gobi processors would be in full production during the first calendar quarter of 2003. MacEdition also revealed that the next generation processor would be code named "Mojave" and start at 1.6GHz based on a 0.10-nm process. First quarter 2003 has come and gone, without any sign of the Gobi processor.

More recently, an interesting forum post claimed to have inside knowledge of future processors coming from IBM. This post reported that the 750-derivative G3s with Altivec capability as well as Front-Side Bus speeds up to 450MHz and chip-speeds up to 2GHz is due as early as this summer. However, assuming the processor described is "Gobi", this information conflicts with previous numbers (described above).

New Info?

Of most interest, however, is an anonymous and completely unverified submission which trys to provide some interesting claims and details on these upcoming processors. As an unverified source, the information should be taken as such, and is provided for interest's sake:

- The PowerPC 750GX is sampling and will be available in the Fall with speeds up to 1.4GHz.
- The 750GX is said to be codenamed "Gobi"
- "Mojave" is an Altivec enhanced G3 to debut at 1.5GHz and will be sampling at the end of the year. This processor is said to be known as the 750VX.

The only available information regarding the 750GX appears to be this product page for the CompactPCI board and its accompanying data sheet (pdf). IBM's site only provides an unhelpful Power Notes PDF encompassing the 740 and 750 lines.

It's unclear at this time if the 750GX is truly "Gobi" or simply a small revision to the 750FX processor. The CompactPCI data sheet implies that the two processors are pin-compatible and would suggest that the 750GX could not support any advanced bus technology.

And Beyond...

Other future timelines for the Power4/5 and PowerPC 970/980 lines were also included with the unverified submission, and are also of unknown accuracy:

- Power5/PowerPC980 work is on target.
- Power5 due later this year, PowerPC 980 in Q3 2004 @ 90nm process.
- PowerPC 980 (4.5GHz-5GHz top speeds)
- 2006: Power 6, PowerPC 990. 990 is targeted at 8GHz.

[Hawthorne]

This may sound stupid (hardly a first for me...), but isn't a G4 by definition a G3 with the Altivec? In that case, what are these 'new G3 w/ Altivec' chips?

[dguisinger]

Originally posted by Hawthorne
This may sound stupid (hardly a first for me...), but isn't a G4 by definition a G3 with the Altivec? In that case, what are these 'new G3 w/ Altivec' chips?

Rofl....so true

[Abstract]

What's stranger to me: Why would IBM continue to produce the G3 for the Mac if they're developing the 970? The next generation Mohave cpu will have clock speeds that overlap the 970, so is the Mohave going to simply be a chip released with the understanding that it is inferior to the 970 but still very fast?

[Sublime]

I think that is correct. I bet there will be some cache gimping to take the confusion away. /boggle
Anyway I hope they throw all that junk out and go straight 970's, except for the mobiles.

[arn]

well...

take it for what it's worth...

Someone with more knowledge than I regarding IBM's processors might be able to make some more sense of it...

arn

[reedm007]

If I remember correctly, the G4 was based off of the PPC 604 architecture whereas the G3 was based off of the PPC 603 architecture. IBM has always pushed the G3 over the G4s architecture, because G3s at the same clock speed at G4s should, in theory, be better number crunchers.

(correct me if I'm wrong here, I'm going off loose memory, and it's 2:15 AM )

[MacBandit]

Well this sounds like the new iBook processor to me. That is if Apple carries on this stupid marketing method of crippling the consumer products.

[mim]

This site - while only going off rumors and such for speculated on chips - has some fairly good info on the difference between processors. Warning - there are a lot of "not known" entries.

http://www.bayarea.net/~kins/AboutMe/CPUs.html

I can't claim to have huge amounts of techical knowledge about the difference between G3 and G4 though, but from what I do know a G4 is not a G3 with alivec slapped on. The G3 is a different design - has less transistors, smaller pipe, etc.

The more time goes by, the more I get the feeling that the 750 (G3) is actually the more advanced design. Apple recent 'fussless' speed bumping to the iBook - making it's clock speed higher than the 2x price powerbook - says to me that they're having trouble finding low enough power G3's to put in the iBook, and altivec is the only thing giving any high-end low-end definition.

Apple need to get the whole line onto Altivec asap. Gutless G4's have held them back long enough. They'll be just as happy as us to see the end of them. Long live the G3!

a.

[backspinner]

Originally posted by Macrumors
In December 2003, ThinkSecret first reported

How do they do that, reporting in the future?

[whooleytoo]

Originally posted by Hawthorne
This may sound stupid (hardly a first for me...), but isn't a G4 by definition a G3 with the Altivec? In that case, what are these 'new G3 w/ Altivec' chips?

IF this rumour is true, my guess is that Apple is looking for IBM to deliver a "new G4" with higher clock and bus speeds; for the low end machines, while going for the 970 for the high end.

This would free Apple from Motorola completely. Given the bad blood between the two from Apple cancelling the cloning program, and Motorola cancelling the G5, this wouldn't surprise me.

Mike.

[centauratlas]

Ok, two things:
1. December 2003?

2. Compare:
"'Mojave' and start at 1.6GHz based on a 0.10-nm process"
and
" Power5 due later this year, PowerPC 980 in Q3 2004 @ 90nm process."

Those are 900 times different in size!

Assuming the post was an accurate in quoting the email, something is fishy - the submitter didn't know enough to get the units correct and yet the other info is presumed correct?

[HasanDaddy]

Well - I'm gonna bring this discussion back home now:

Should I buy the 17" LapZilla now, or wait?

THANKS!

[RalphNumbers]

Sounds like they're looking for a cheap low end chip, without the frills one might expect of a high end system (like SMP, L3, a double FPU, SIMD, 64bit addressing, etc) and with a correspondingly smaller die and therefore lower price.

I'm really hoping that this means that they'll be transitioning entirely away from Motorola ASAP.
I never much liked the G4, it focused on all the wrong areas IMHO. Motorola has a history of strength in DSP design, and they leveraged that into a pretty damn cool SIMD unit, but they lost alot of the simplicity and clockability of the PPC in doing so, and they made only minor improvements to the scalar ops. This, coupled with Motorola's mediocre fab tech ( especially compared to IBM), lead to a dead end, as far as desktop CPUs are concerned.

Integer scalar op speeds stagnated, to the point where I wouldn't be surprised if these G3s are at least as fast in that domain as equivalently clocked G4s if not faster. These G3s will probably be a bit behind in FP ops, but I'd say it's a decent price to pay, considering the market niche it'll be aimed at with the 970/980 on the high end.

[Snowy_River]

Originally posted by centauratlas
... Compare:
"'Mojave' and start at 1.6GHz based on a 0.10-nm process"
and
" Power5 due later this year, PowerPC 980 in Q3 2004 @ 90nm process."

Those are 900 times different in size! ...

I really think that this is a trivial typo. I have seen this exact same typo made so many times on these message boards that I have completely lost count.

One thing that I'm glad to see is that there is no mention of anything like a "60 nm process", as we are quickly closing in on the limit of how small current manufacturing methods are capable of. Beyond this we get into the range where quantum variations start to make significant disruptive influences. The new hope for smaller processes is the use of carbon nano-tubes. To the best of my knowledge, the only labs that have successfully made a transistor array out of carbon nano-tubes (i.e. a simple microchip) are, hmm, guess where? IBM! Gee, that's promising...

[Snowy_River]

Originally posted by RalphNumbers
Sounds like they're looking for a cheap low end chip, without the frills one might expect of a high end system (like SMP, L3, a double FPU, SIMD, 64bit addressing, etc) and with a correspondingly smaller die and therefore lower price.

I couldn't agree more. And I think that this is a tactic that makes a lot of sense...

[MisterMe]

Originally posted by reedm007
If I remember correctly, the G4 was based off of the PPC 604 architecture whereas the G3 was based off of the PPC 603 architecture. IBM has always pushed the G3 over the G4s architecture, because G3s at the same clock speed at G4s should, in theory, be better number crunchers.

(correct me if I'm wrong here, I'm going off loose memory, and it's 2:15 AM )

Your memory parallels mine, but we reach different conclusions. The PPC 604/604e has been a workhorse for IBM. For all the talk on this forum and others about how great the POWER4 is, most of IBM's offerings are based on the POWER III. The company continues to sell entry-level RS/6000 workstations based on the PPC 604e. The PPC 603 was noted for its low-power design and inferior FPU. IIRC, since the G3 is based on the simpler PPC 603, it was easier to scale up its clock. That has a direct and positive effect on integer math. However, my own experience is that the G3's floating point performance is barely superior to the PPC 604's. That the G3 is even competive with the PPC 604 is due solely to the G3's faster clock. Clock cycle for clock cycle, the G4's floating point blows away the G3's.

[chrysrobyn]

Originally posted by reedm007
If I remember correctly, the G4 was based off of the PPC 604 architecture whereas the G3 was based off of the PPC 603 architecture. IBM has always pushed the G3 over the G4s architecture, because G3s at the same clock speed at G4s should, in theory, be better number crunchers.

(correct me if I'm wrong here, I'm going off loose memory, and it's 2:15 AM )

The way I understand it is far more linear.

601 -> 602 -> 603 -> 604
                |-> 750 -> 7400

Some of the development went on in parallel, but the 7400 just doesn't have the dispatch that the 604 did-- the 604 was more power hungry because it could dispatch more and therefore be doing more and using more power. The 603, however, turned out to be almost ideally suited to lower power uses. Basically, it was far easier to take the 750 and tack on a vector unit than it was to take the 604, modernize it 2 technologies, update the caching structures and add the vector unit.

For a partnership that wanted to cut costs by keeping a processor that could go into the desktops in the initial technology and down to the portables after a technology and voltage shrink, the 603 was the more logical choice.

[gothamac]

One thing that I'm glad to see is that there is no mention of anything like a "60 nm process", as we are quickly closing in on the limit of how small current manufacturing methods are capable of. Beyond this we get into the range where quantum variations start to make significant disruptive influences. The new hope for smaller processes is the use of carbon nano-tubes. To the best of my knowledge, the only labs that have successfully made a transistor array out of carbon nano-tubes (i.e. a simple microchip) are, hmm, guess where? IBM! Gee, that's promising...


Moores Law retains grip on IT statute books - IBM
By IT Analysis
Posted: 08/05/2003 at 09:25 GMT

"The power of computers has increased by six orders of magnitude in the last 36 years and it will increase by a further six orders of magnitude in the next 36 years", claimed Nick Donofrio, IBM's Senior VP of Technology and Manufacturing to an audience of IT analysts at IBM Pallisades, writes Robin Bloor
of Bloor Research.

'Six orders of magnitude' is a math-speak for "a million-fold" so Nick was telling us on the one hand what we already knew, that Moore's Law has been operating since the late 1960s, and on the other hand, professing a belief that it would continue to operate for the foreseeable future.

He has reasons for his convictions and, in a fascinating address, he referred to various areas of research that IBM was involved in which led him to conclude that Moore's Law will remain on the IT statute books. Here they are:
* Nanotube technology. Nanotubes are microscopic tubes constructed from carbon rings which can be used to build logic circuits. Currently this technology is between 50 to 100 times denser and therefore faster than current silicon. So in its current infant state, it offers about two orders of magnitude improvement and is expected to offer more in time.
* Nanodisk. IBM has built nano-machines that can store data on and erase data from a surface by puncturing a hole in it (or removing it by heating the surface up), using an array of minute cantilevered arms. This is effectively a nanodisk which is 25 to 50 times smaller than current disks and can probably be made even smaller.
* The Molecular Cascade. IBM has been building molecules using an electron tunneling microscope. One of the things it has built is a big molecule that can act rather like Babbage's computer as originally conceived with balls rolling down paths and passing through gates, except of course that the balls in this instance are atoms. It is thus possible to build a molecular computer, the smallest nano-machine yet conceived. This on its own would deliver the six orders of magniutude that Nick Donofrio is looking for.
* The Quantum Computer. A quantum computer is an extremely small photon driven device which can perform some kind of useful logical work, particularly in the area of encryption. A working device would be 6 orders of magnitude faster than current computers.

These were not the only futuristic developments that Nick Donofrio dealt with. He said that in the next 10 years IBM expected an explosion in secure sensor based computing. This is the broad extension of the use of sensor devices in cars in order to optimise engine performance, except of course that sensors will be embedded everywhere, allowing the optimised behaviour of just about any device you can think of in conjunction with any other. Clearly there are a host of applications in the home and in offices.

He also mentioned Web Fountain, the result of an IBM research initiative. This is an intelligent search technology, which he claimed had the ability to assemble a 'domain of expertise" which could then be queried. Think security and the idea of assembling a coherent body of knowledge on a terrorist organisation. IBM intends to offer this technology as a service rather than a product.

Finally he made some wry comments about IBM's Linux watch - a research product which IBM has gradually been evolving. Currently it has GSM, is able to record movement, read vital signs from the owners body and has wireless connectivity, and it also tells the time. He said that the watch was never intended to become a real product, as its form factor was very large, but he noted that large watches were currently becoming fashionable.

© IT-Analysis.com

[Freg3000]

All I can think about is a dual 8 Ghz PPC 990 computer in 2006.......drool.......

[Mr. Anderson]

Originally posted by Freg3000
All I can think about is a dual 8 Ghz PPC 990 computer in 2006.......drool.......

Dual? If the roadmap for these chips is to be believed, they're going to be multicore - so you might end up with 8 CPU cores in one chip!!

Then you're dual would actually be 16 CPUs all running at 8 GHz.....now that's something to drool about.

I'd actually believe that realtime radiosity rendering at small frame sizes might acutally be possible with a system like that - a personal goal I'd like to see...

D

[Frobozz]

Originally posted by Abstract
What's stranger to me: Why would IBM continue to produce the G3 for the Mac if they're developing the 970? The next generation Mohave cpu will have clock speeds that overlap the 970, so is the Mohave going to simply be a chip released with the understanding that it is inferior to the 970 but still very fast?

Well, it might be inferior in raw speed... but it may use very little power and be used in laptops and maybe lowend desktops like the eMac?

[knoxer]

I have a feeling they are specifically designing this as a laptop processor that can last for many many hours, and still be plenty powerful...

[Greenlightboi]

Originally posted by mim
This site - while only going off rumors and such for speculated on chips - has some fairly good info on the difference between processors. Warning - there are a lot of "not known" entries.

http://www.bayarea.net/~kins/AboutMe/CPUs.html

I can't claim to have huge amounts of techical knowledge about the difference between G3 and G4 though, but from what I do know a G4 is not a G3 with alivec slapped on. The G3 is a different design - has less transistors, smaller pipe, etc.

a.

THe G4 is a slightly modified G3 with altivec on the die. The G4 is, basically, a G3 but it isnt the exact core with altivec on the die. I would definitly take this rumor with a grain of salt, alot of things don't add up. The G3 is a good chip, but if you think its better then the G4, your crazy... the Altivev SIMD extentions allow data to be processed at 4x the normal rate (128 bit vs. 32 bit). Though the G4 is a badly designed chip.

[deepkid]

New Info?

Of most interest, however, is an anonymous and completely unverified submission