Ibm Gpul2
A MacRumors report today cites an eWeek article about IBM's successor to the 970, called GPUL2. My question is, if this is a die shrink of the 970, is this what'll end up in PBs? Will this be what we refer to as the 980? Is this what Jobs was referring to when he said we'd be at 3 GHz in a year? Be sure to read the eWeek article linked from the MacRumors article.
Comments
I'm all for IBM getting aggressive about Linux on PPC. We will be the beneficiaries of fast processors and a more stable lineup.
This viewpoint is of course based on speculation on top of speculation, so I don't really know what to think.
Is it reasonable that the die-shrink would take another 10-11 months from now? Or, alternately, is it reasonable that we'd have the 980 so soon?
not sure if that was post-980 or if die shrink was a prereq to dual-core at usable power/heat levels
*heads off to search for linkage*
And as far as the wait for a G5 PowerBook is concerned, you'll have a far more reasonable expectation of the timetable if you simply look back at previous transition of like kind. The G4 towers came out in 1999, I recall, and yet we've only had the PowerBook G4 since 2001 I believe. Yup, a Google search validates my memory -- the first PB G4 came out in January 2001. Now that we're dealing with IBM one expects the processors to progress more quickly, but realisitically the migration down to the PB will still take time.
G3/233-introduced 1997.11.15
G3 PowerBook-introduced 1997.11.10; discontinued 1998.05.04
G3 AIO-ntroduced 1998.04.03; discontinued 1999.01.01
iMac 233-announced 1998.05.06; North American release on 1998.08.15 at $1,299; replaced by Revision B in mid-October 1998
Dates from LowEndMac web site. You will notice that the desktop and laptop G3 were both released in the same year, followed up by a quick migration to the true "consumer" mac released less than a year later. I dont think we will see a portable G5 this year, but if IBM can mirror their succes in manufacturing the G5 that they did with the G3 then I see the possability, and need, for Apple to release a PowerBook and iMac using the G5 next year. If there is a hint of truth to the GPUL-2 then the PM's may even be ready to move to a G6 in about 18 months anyway.
The G5, however, will obviously need more of a redesign. That's why you'll see an upgrade taking longer than people might like.
GPUL = G5 970 @ .13µ ranging from 1.2 GHz to 2.8 GHz
GPUL2 = G6 980 @ .09µ ranging from 2.8 GHz to 4.5 GHz
(Ranges are rough estimates)
My question is, will there be a die shrink of the 970? If so, mightn't it be in the first quarter of '04? If there's a die shrink, could the die stay the same size by doubling the L2 cache? That's what Moto is doing with the 7455 to 7457 revision, going from .18µ to .13µ.
If, as Jobs said, the Power Mac is going to 3 GHz by July of '04, won't that be the 980 at .09µ? If so, does it make sense to invest the time, money and effort to die shrink the 970 just a few months before the 980 comes out?
Wasn't it Joswiak who said at WWDC that there wouldn't be a G5 PowerBook anytime soon? Didn't he play up the importance of Motorola? If Moto is coming out with the new 7457 for use in PBs and iMacs, would it make sense for Apple to contract for the chips for only 6 months?
What about this scenario:
7457 in PBs by August, in iMacs by Sept. and both are speedbumped with faster G4s 6 months later. G5 PMs speedbumped to 2.4 GHz by end of Jan. using same .13µ process. PM moves to G6 at up to 3 GHz in July. PBs and iMacs bumped up to faster G4s in March, April and then move to low end G6 by late summer.
It may be that Apple continues to call the different versions of the 9XX series the G5. The G4 didn't get a name change throughout all the revisions from the 7400 to the 7457 with all of their speed changes and die sizes.
So many questions, so few answers.
Originally posted by @homenow
After the long road with the G4, I would hope that Apple pushes the G5 into their product line like they did the G3 rather than the 4.
G3/233-introduced 1997.11.15
G3 PowerBook-introduced 1997.11.10; discontinued 1998.05.04
But the reason Apple could do that is the G3 was a low power, low cost, designed-for-laptops chip.
The G3/233 _desktops_ were NOT great. The 604 machines (and their varients) were better in some areas, and FAR better in other areas (like 6 slots!) The main disadvantage was PRICE. Apple was basically in full retreat/dig-in-and-weather-the-storm mode.
There has to be some kind of work done before a next-generation laptop. Either the G5 needs to be cooled off/shrunk, some other chip is being designed, IBM's G3's are extended with AV, or Apple goes to Motorola.
Surely if the GPUL2 were the 980 (based on the POWER5) then they would milk it for everything they could. I don't see a mere mention of an update as a hint to a POWER5-based chip. I also don't see how, if IBM can make blade servers with the 970, Apple can't put at least one in an Xserve, unless they're waiting for Panther Server to be ready ...
It also confirms that the 970 will be deployable into 1U systems such as Apple's xserve. Low end Blades are indeed 1" modules in 7U Chassis now, it will be interesting to see IBM sales of these compared to their Xeon offerings - though they have a sales advantage being able to run Windows Server Software.
Originally posted by hasapi
- though they have a sales advantage being able to run Windows Server Software.
And disadvantage of not being able to use OSX
Chips ahoy!
Speaking of intergenerational relations, the Blade has received some snippets of new intelligence about the successor to the PowerPC G5; aka PowerPC 970; nee GPUL (GigaProcessor Ultralite).
The GPUL2, which was recently fingered as the centerpiece of a mid-2004 line of Linux servers from IBM, is also a presumptive player in Mac models due at the same time; however, the BladeÕs sources caution that, whirring of the propaganda machine aside, the Rev B processors wonÕt mark much of a departure from the Power4-based underpinnings of the current chip.
Specifically, the next-generation processor will be based on IBMÕs Power4+, not the Power5, the BladeÕs moles attest. ÒGPUL and GPUL2 are considered the same architecture, and if the marketing drones donÕt fiddle with it, GPUL2 should be marketed under the 970 series,Ó they chirp. ÒThe 980 series (if indeed it will be called 980 at all) will be the chip referred to as P5UL (Power5 UltraLite).Ó
It seems to me that GPUL2 will be the .09µ version of the 970, will be out by July 2004, will be up to 3 GHz, and will still be called the G5.
BTW, I'll bet the Blade scans these pages. 8)
Originally posted by Rolo
The Naked Mole Rat has an NMR Report out today that mentions GPUL2:
It seems to me that GPUL2 will be the .09µ version of the 970, will be out by July 2004, will be up to 3 GHz, and will still be called the G5.
BTW, I'll bet the Blade scans these pages. 8)
M.Isobe over at the Ars Macintoshian Achaia forum in
this thread says:
""P5UL (for Power5 Ultralight), may or may not be branded 980, but will be a Power5 derivative."
"No. The codename of POWER5 is not "P5", but "GR". Furthermore, the present 970 has already been POWER4+ based specification, such as 130nm process, 512KB L2 block and 3:1 bus ratio support."
Seems the "Blade's" info is a little off.
MM
From IBM's press release
1.2 GHz IBM POWER4+ processor with peak power consumption of 35 watts per processor.
Well, this is old news, but POWER4+ seems to be a low-power mod of POWER4 on a smaller process (130nm). From the PDF I gather that POWER4+ performs very much like higher-clocked POWER4. So, considering that PowerPC 970 is on 130nm, I think it's based on POWER4+. Am I wrong?
Originally posted by MartianMatt
M.Isobe over at the Ars Macintoshian Achaia forum in
this thread says:
""P5UL (for Power5 Ultralight), may or may not be branded 980, but will be a Power5 derivative."
"No. The codename of POWER5 is not "P5", but "GR". Furthermore, the present 970 has already been POWER4+ based specification, such as 130nm process, 512KB L2 block and 3:1 bus ratio support."
Seems the "Blade's" info is a little off.
MM
You are right there is not much difference between a power 4 and a power4 plus : more speed, a greater L3cache.
The next PPC 9XX chip will be based upon the power5.
The question is : is the next chip from IBM will be the powerPC 970 on 90 nm process with some improvements like a bigger L2 cache (1MB), or directly a new design derivated from the power 5.
The rumors seems to be in favor of a PPC 970 on 90 nm process first.
Originally posted by Powerdoc
The question is : is the next chip from IBM will be the powerPC 970 on 90 nm process with some improvements like a bigger L2 cache (1MB), or directly a new design derivated from the power 5.
The rumors seems to be in favor of a PPC 970 on 90 nm process first.
With all due respect for M.Isobe, he is not an IBM employee and does not have any more access than any of the rest of us. Where he got the code name, I have no idea, but I suspect that TGB's info is more accurate.
I think we will next see a 90 nm PPC970, possibly with a slightly updated Altivec engine (not sure about cache), followed by a 90 nm SMT version of a Power5-based chip (P5UL) about one to one and a half years from today.
Originally posted by Tomb of the Unknown
With all due respect for M.Isobe, he is not an IBM employee and does not have any more access than any of the rest of us. Where he got the code name, I have no idea, but I suspect that TGB's info is more accurate.
I think we will next see a 90 nm PPC970, possibly with a slightly updated Altivec engine (not sure about cache), followed by a 90 nm SMT version of a Power5-based chip (P5UL) about one to one and a half years from today.
IBM has been steadily increasing cache size with changes in process or sometimes with out. The 750CX series went to 256KB L2 from 0KB in the 750. The 750FX went to 512KB and the 750GX will go to 1MB. An educated guess would show the 970 going to 1MB cache with a move to 90nm.