Talk:PowerPC 970

The listing of every Apple product that uses a G5 in it seems rather unneeded. This is a page about the PowerPC 970, not about apple products.

Intel edit?
A user posted information about Apple's recent transition to Intel. Is that really realevent to the article? Personaly, I don't. I will revert the edit. --CoolFox 19:22, Jun 6, 2005 (UTC)

do not enter —Preceding unsigned comment added by 76.84.149.40 (talk) 21:10, 31 August 2007 (UTC)

what about the processor itself
-power consumation -unit(activec alu...) -etc...

i also am interested in the TDP of the different G5 models... Plonk420 00:36, 20 November 2005 (UTC)

I found an IBM document dealing with the power consumption of different G5 versions, and also how different voltage would affect the power consumption. It's out there on the IBM website and I'll see if I can find it again. /Håkan 2007-02-25

Endianness?
There are claims that unlike previous PowerPC processors, the 970 series cannot handle both high-byte/low-byte and low-byte/high-byte ordering of values. This is the same thing as the Big Endian and Little Endian business. The PowerPC could switch from the Big Endian standard (which is what the English language and Arabic-math systems work in) to the Little Endian format popular with the aged x86 and MOS 6502 architectures. This switching ability was perhaps first implemented as a way to faciliatet faster x86 emulation.

In any event I'm agog that the ability to handle both formats would be taken out. Frankly I find it hard to believe. I'd like to see confirmation and (ideally) justification. Please tell me the PPC series isn't heading off towards Little Endianness!

Morons
Some person, obviously with a huge sense of humour changed the image for this page to an image of a toilet. I changed it back to the image found in the previous revision.

Insides of a Rev.?
This isn't a Revolution that the page is reffering to is it? —The preceding unsigned comment was added by 70.112.76.158 (talk • contribs) 01:25, 11 May 2006 (UTC)


 * No, it is not. "Rev." is short for "Revision" in the caption.  As the caption says, it's a G5 iMac. Guy Harris 01:54, 11 May 2006 (UTC)


 * It would be a good idea to have a link to the broadway page. Statements by IBM, reports from Ars and various other sources indicate that it is the low power consumption G5 processor. —The preceding unsigned comment was added by 64.6.0.220 (talk • contribs) 23:22, 15 September 2006 (UTC)
 * On the contrary.. the latest rumors (since there are no official word from either Nintendo nor IBM) suggests that it's a evolved Gekko, ie a G3 part. If you have a source where IBM states something, I'd be glad to reconsider. Either way it shouldn't be long until the official specs are revealed. -- Henriok 18:44, 16 September 2006 (UTC)


 * I'll try to find the quote from the IBM guy, here are a few of the ars technica articles where they report that IBM offered broadway to Apple as a laptop G5 http://arstechnica.com/news.ars/post/20060520-6877.html http://arstechnica.com/news.ars/post/20060802-7407.html. I don't think its very likely IBM would have offered Apple a chip that is simply a G3 as a potential laptop chip. Also the language in the quote from the IBM rep (which I can't seem to find) he made it sound like it was something new, and not simply a lower powered G3 class chip. I think the final nail in the coffin would be the $250 pricetag of the Wii. Unless Nintendo are taking $75 or more in pure profit on every console sold or IBM is charging Nintendo individual prices on bulk orders the processor in the Wii has to be something more than a G3. —The preceding unsigned comment was added by 64.6.0.220 (talk • contribs) 20:39, 16 September 2006 (UTC)
 * Hannibal is an excellent journalist and on any other day, his word is law, but not in this case. He's clearly uninformed and says so himself. He speculates but the rumors are quite firm for the time beeing. It's not a decendant of 970 in Wii, and the only thing we have to go on that's _NOT_ speculation is that it is indeed an evolutionary step from the previous Gekko. No more, no less. Speculation of how Nintendo choses to price their console is neither here nor there. It would be an enourmous leap to draw any conlusions regarding lineage of the processor based on their pricing strategy of the entire platform. -- Henriok 23:01, 17 September 2006 (UTC)


 * Even if the pricing is irrellivant, it is highly unlikely IBM would consider offering a PPC750 part to Apple as a processor. "that's _NOT_ speculation is that it is indeed an evolutionary step from the previous Gekko" I haven't heard that stated by anyone in the know as being true so it is also questionable in my opinion. Really we shouldn't be listing broadway either as a PPC750 or as a PPC970 part till we know for certain. Truthfully a PPC970 chip with the gecko extensions could be considered 'an evolutionary step from the previous Gekko', certainly more than 'Gekko at a higher clockspeed' could be. —The preceding unsigned comment was added by 64.6.0.220 (talk • contribs) 23:04, 18 September 2006 (UTC)

970GX missing in action
IBM "outed" the 970GX in some documentation, and it has been seen on a couple of roadmaps in the past.. but AFAIK, it's missing in action.. IBM does not list it as an available part and I haven't seen it in the wild. -- Henriok 14:42, 4 July 2006 (UTC)

Move
I belive, that like all of the other apple-used proccessors, this article should be named PowerPC G5, like all of the others. I will move it, and if anyone dissagress, please feel free to move it back. 71.225.248.188 21:35, 13 November 2006 (UTC) (i will log in to move)
 * Gaaah :( This article should REALLY be named in the way that it was PowerPC 970. It is the OTHER articles that shoud be renamed to better reflect what the actual names are, ie PowerPC 750 and PowerPC 7400. Using codenames (G3) and Apple's marketing (G4 and G5) shouldn't really be a base of which we should name these articles. The names _are_ PowerPC 970, PowerPc 750 and PowerPC 7400, and the names of the respective articles should reflect this. Really.. this is a no brainer. I can't believe that the other articles have kept their names for so long. Rederections from Apple's names are appropriate though. -- Henriok 22:20, 13 November 2006 (UTC)

terra soft powerstation
Seems to be used in the Terra Soft Powerstation. is this worth adding. http://www.terrasoftsolutions.com/products/powerstation/ --87.127.117.246 (talk) 18:55, 10 June 2008 (UTC)

odd sentence comparing g4 to g5 in laptop potential
"Indeed, the 970FX only generated around 30 W TDP at 1.5 GHz, which while slower in clock speed over Motorola/Freescale's older G4 processors, would have maintained the advanced design of the 970 like full 64 bit design and better clock scalability."

This is odd, boasting of "better clock scalability" makes no sense when the first part of the sentence concedes the g5 would have to be underclocked to work in a laptop due to thermal issues. And the only advantage of 64 bit design would be using more RAM, more RAM which would not fit into a laptop.

xenon is based on the PPC970
the xenon and PS3 PPE is based on the PPC970, evidence of this was also seen on the devkit of the xbox360 which was a dual and later Quad-Core PPC970 based Apple Powermac G5 (pm which Microsoft had the Branch-Prediction disabled on, because the xenon is a PPC970 without branch-prediction) Markthemac (talk) 00:06, 11 October 2010 (UTC)
 * No, it's not based on the PowerPC 970 core (which is nearly identical to a POWER4 core). The reason MS used PowerMac G5s as dev boxes is that PPE and 970 both are 64-bit PowerPC-processors with AltiVec. The similarities ends there (there's A LOT more than just a disabled branch predictor differentiating the designs). The PPE is new PowerPC core/pipeline design, a fact backed up by several sources from IBM, Sony and Microsoft. Your statement is backed by speculation and original research. -- Henriok (talk) 07:18, 11 October 2010 (UTC)

sorry your wrong, the PPE it's a standard Power4 based PPC64 (which the PPC970 is), and i was wrong about branch-prediction but it's still a G5 based PPC64bit. in fact it's right here. http://www.ibm.com/developerworks/power/library/pa-fpfxbox/?ca=dgr-lnxw09XBoxDesign Markthemac (talk) 17:07, 11 October 2010 (UTC)
 * 100% PowerPC ISA 64BIT just like the PPC970, it runs the same it smells the same heck it is the same except for a smaller L1/L2 and slower FSB Markthemac (talk) 17:17, 11 October 2010 (UTC)
 * lets just keep it at Power4 based PPC design, because that's what both the xenon and PPC970 are. Markthemac (talk) 17:42, 11 October 2010 (UTC)

How can you say that the facts are right there, when the article you are referring to does not contain ANY reference to either "970", "POWER4"or "ppc64"? It makes absolutely no sense. You seem to be unaware that there are a multitude of 64-bit PowerPC implementations from three different vendors that bare no other relationship than the fact they implement the same instruction set (64-bit PowerPC), several of them preceding the POWER4. It's perfectly OK to design two processors independently of each another that implement the same instruction set architecture without implementing it in the same microarchitecture. This is the case of POWER4 and PPE, i.e. they are only related in that they share the same instruction set, just like it's the case of most other 32 and 64-bit PowerPC implementations. -- Henriok (talk) 20:18, 11 October 2010 (UTC)

same transistors same execution means it's based on the same chip, there isn't any vast difference even the power usage is almost 100% identical to a PPC970 (also talking on the development side it's 100% identical to programing on a PPC970 based mac) Markthemac (talk) 16:00, 12 October 2010 (UTC)
 * 1) They have not the same transistor count (The entire PowerPC 970 chip = 52 M, just the PPE+512 kB L2 cache ≈ 57 M), and not the same execution cores nor instruction pipeline (se below).
 * 2) It is identical programming since they both follow the same general instruction set: 64-bit PowerPC. In fact, at closer scrutiny they DO NOT follow exactly the same ISA. PowerPC 970 uses the PowerPC 2.00 spec while Cell uses the Power ISA 2.02 spec. Cell/PPE has memory tags and hypervisor support, 970 does not.
 * 3) The figures you state regarding power are made up, or at least unsubstantiated. A Cell draws ~80W @ 4 GHz. A 970MP ~100W @ 2 GHz. Hardly identical. And I can't even find figures how much a single core draws, just the complete chips which says nothing. A cell chip and a 970 is so obviously not the same _chip_.
 * 4) The transistor layouts of the processor cores are NOT the same as can be clearly seen on pictures: Xenon vs Cell = the same. 970MP vs POWER4/POWER5 = the same. The 970 core and the PEE is _cleraly_ not the same core.
 * 5) The differences in microarchitecture between 970 core and PEE are substantial.
 * 6) 970 is single threaded, PPE is 2x hyperthreaded
 * 7) 970 is out-of-order executing, PPE is strictly in-order
 * 8) 970 has a 15 stages long pipeline, PPE is 21 stages long
 * 9) 970 has a eight issue pipeline, PPE has a dual issue pipeline
 * 10) 970 can fetch eight instructions, PPE can fetch four
 * 11) 970 has eight execution cores (2xINT, 2xFPU, 2xLS, 2xVMX), PPE has five (ALU, FPU, LS, 2xVMX)
 * 12) 970 has three branch predictors (2x branch, 1x condition register), PPE has one
 * 13) 970 has 64 kB direct mapped L1 Instruction cache, PPE has 32 kB 2-way set associative L1 I-cache
 * 14) 970 has 32 kB 2-way set associative L1 Data cache, PPE has 32 kB 4-way set associative L2 D-cache
 * Can we please put this argument to rest? You are so clearly misinformed, and I've provided ample evidence to contradict every claim you make. And I've not even pointed to the several documents from IBM stating that PPE is a completely new core, designed as such by Sony's demand. Or from very reputable third party sources like Ars Technica, Anandtech and Real World Technologies making my case, that PEE share no heritage with previous designs, especially not POWER4/970. And those articles are from 2005! You are beating on a horse that's been dead for five years! -- Henriok (talk) 22:53, 12 October 2010 (UTC)
 * take a look at that PPC970 and put it sideways the CPU core itself is almost 100% identical Markthemac (talk) 01:01, 13 October 2010 (UTC)
 * No, it's not. The 970 core is not remotely like the PPE or Xenon cores regardless how you twist, flip or turn it. If you think otherwise, please show. The burden of proof lies on you. IBM, Sony, Microsoft, Ars Techica, Anandtech and Real World Technologies would be really interested in your findings, since that's be news to them. -- Henriok (talk) 07:56, 13 October 2010 (UTC)
 * yeah the xbox360 dev kit so wasn't a dual core PowerMac G5... Markthemac (talk) 00:09, 21 January 2011 (UTC)
 * But alas, they were. At least the early ones: http://www.xbox-scene.com/xbox1data/sep/EEEuEplpEAdQIZjXID.php -- Henrik

64 GB in PowerMac G5
Hello, I'm currently trying to add a source to the 32/64 GB G5 rumor, but it tells me sources are unavailable whenever I try to add one. The pages I want to add are these two: https://forums.macrumors.com/threads/g5-quad-32gb-possible.1126338/ & https://www.reddit.com/r/mac/comments/1y5pvx/32gb_ram_powermac_g5_quad/ As stated before though, these are just rumors and no ones bothered to do a proper investigation on them. Still, I think the info should be preserved, someone may investigate this eventually. 70.170.67.244 (talk) 23:48, 30 March 2017 (UTC)
 * Thanks for the links, but they won't do as sources for Wikipedia. I do like them being here in the Talk page though! -- Henriok (talk) 20:12, 1 April 2017 (UTC)

“PowerPC 970GX”
Reverted anonymous edits claiming the existence of a “PowerPC 970GX” variant. Over the years the “970GX” seems to have become something of a self-perpetuating myth. A Google search suggests that at some point early on a secondhand press release from IBM mentioned the chip would ship, but despite my best efforts I have not ever been able to find any hard evidence that a “970GX” ever made it to tape-out, much less ever got manufactured, much less ever shipped. If this were an actual, real product, it should be super-easy to find at the very least a data sheet or any archived IBM page mentioning the product. (P. S.: If anyone can find a datasheet for the 970GX and link to it here, I will happily be the first to document that in this Wikipedia article.)  IBM seems to have tacitly admitted the product was vaporware when they retracted all mentions of the “970GX” from their own Web sites shortly after the secondhand press release started circulating in the media at the time.

Also, IMHO, all the artist’s renderings of the PowerPC 970 chips in the article should be removed and replaced by actual die photos in an article that purports to be encyclopedic (compare this with Wikipedia articles on Intel CPUs, which use actual photos of real Intel chips). I have not gone so far as taking this step because I am sure whoever put all the effort into these artist’s renderings (which I admit are pretty) is emotionally invested and would not take this well, so I’ll let that go.

I think it’s important to remember that at that time the PowerPC 970 was the basis of Apple’s fastest, top-of-the-line Macs, and like many Apple products, it seems that these rumors and speculation were amplified in an enthusiastic feedback loop among Apple fans and aficionados, but after all these years I have never been able to find any hard, independently verifiable evidence of the “970GX” ever having existed as a real, actual product. It does seem to resurface time to time on Apple fan sites, so I don’t expect the myth to die. Full disclosure: I run or have developed on Power Mac G5s and Xserve G5s based on Power PC 970, 970FX, and 970MP CPUs, so believe me, no one wishes there were such a fantastical chip as the 970GX more than me. Sadly, however, this seems to be more a case of wishful thinking among Apple fan and rumor sites. —PowerPCG5 (talk) 02:22, 4 April 2021 (UTC)

970 to 970FX Pipeline Length Change
According to and both the 970 and the 970FX have the same pipeline lengths, so I removed the talk about the difference between them. The confusion seems to come from the 9 fetch and decode stages. Admock (talk) 04:31, 6 September 2023 (UTC)