Submitted by PPCNUX-Team member Arno found a source that posts prices of the PowerPC970 (G5) CPUs. They conclude that that could not be the reason that there’s no Apple machines running on PowerPC anymore. “G5 Quads for everbody! Thanks to IBM’s pricing policy, the open-armed Power.org community, and Genesi’s sustained commitment to the Power Architecture this could become a truely realistic option in the not-so-far future…”
The G5s lesser siblings are quite strong in the embedded market, but apart from the usual suspects, i.e. enthusiasts, what oppertunities exist for the G5 these days?
* There are no signicficant operating systems that run on the G% that doesn’t do so on the x86/x86-64 architectures. Most do so even better than on the G5 (e.g. better driver support, better tested and so forth).
* While other PowerPC CPUs are lean on power consumption and heat generation, the G5 is not.
* There are few suppliers of quality north- nad southbridges for PPC. Just recall the the problems both Eyetech and Genesi had with their respective motherboards.
So, now my question is simply this, what qualities does the PowerPC G5 platform have that negates these three points in order to make it competitive?
NB: I’m writing this from PowerMac G5 which is easily the best computer I’ve ever owned. Also note that I very much like to see the PowerPC and other CPUs fare well in the market as a dominant hardware architecture is as bad as a dominant operating system.
Edited 2006-09-01 15:39
Have these people never heard of VOLUME selling? Would Avnet ever be able to keep up with Apple’s demands? 5000 units is nothing; try 100 times that. At least. Per quarter. And units faster than 2.5GHz, because current Core 2 Duo processors are blowing these G5’s out of the water — to my great dismay. I used to be such a fan of the PowerPCs… Then I saw the light with MIPS.
And this has been beaten over and over: the videogame processors are NOT the same as the desktop-oriented CPUs. Unfortunately the whole plethora of existing software can’t just be recompiled with some IBM toolchain that takes care of instruction scheduling on in-order, single- or dual-issue CPUs. Nintendo, Sony and Microsoft are starting clean-room on the whole “powered by PowerPC” videogames; not Apple. Not since 1996.
The problem is that the G5 is to slow, and use to much power, to be used effective as a mobile chip.
“Too slow” is a bit excessive. It is extremely fast at floating-point arithmetic and SIMD processing, and it has only been with the Core 2 Duo that Intel has begun to draw close.
It’s major weakness was slower than average integer performance. As the majority of programs use integer arithmetic far more often than floating-point arithmetic, most programs run faster on Conroe, and even Yonah, than they do on the G5. That said, here are some areas where floating-point performance matters; examples include as scientific computing and graphics work.
Your right about that last bit. These cheap G5s will be great for people doing signal processing apps, especially if they do manage to get the chipset up and running with DDR2-667 memory.
…that did not get over that Apple switched to Intel.
You want some reasons why Apple switched? ok, here they come:
1. Where are the 3GHz?
2. Where is a trusty roadmap at IBM what comes after the G5?
3. Where is the guarantee that IBM can deliver in quantities (mainly new models)?
4. How can you think you know about Apples conditions they get from Intel?
Enough?
Apple switched to x86 because the mobile prospects on the G5 were not looking good.
Intel has a strong mobile lineup and thats what apple needed. The laptop market is a hot one and critical for apple’s long term success.
The G5 though a great chip is history now. Without Apple, I don’t believe it will gain a respectable market share to justify its production. The 970 is like the old well built vacuum cleaner that was heavy and sucked up a lot of energy.
Well on the other hand, maybe IBM wants to make suitcase sized “laptops” incorporating the 970 chip, power supply, water cooler.
finally someone states the obvious. Give me a G5 laptop and then I will say maybe Apple was quick to transition. BTW, laptops give Apple their biggest margin and that is where they have significant market share, even despite runing on crappy G4s for a long time. So “G5 Quads for everbody!” I guess he meant for those people looking to run phososhop all day long or use it a as a server. certainly not laptop-only people, such as myself.
This article is shortsighted. With PPC Apple had to develop chipsets themself. That cost a lot of money and I wouldn’t wonder if the costs of the chip easily top the price for the cpu.
Those PPC-fanatics are really annoying…
I think this place is getting better. People commenting with facts instead of pure fanaticism.
Pricing shouldn’t be compared with MHz, they should be compared relative to performance. A 2.33 GHz Xeon 51xx is about 60% faster than a 2.5 GHz 970MP on SPECINT. If we normalize by performance, we’re comparing a 2.5 GHz 970MP to a 1.6 GHz Xeon 5150. At $230, the Xeon is substantially cheaper.*
And as MatzeLoCal pointed out, the cost of motherboard and infrastructure shouldn’t be underestimated. Designing your own chipset and motherboard isn’t cheap, and if Intel can provide a ready-made solution that’s most of the way there, it saves you tens of millions of dollars in R&D.
*) Of course, that’s not really a completely fair comparison. The 970MP is essentially the same as the circa-2003 970. Meanwhile, the Xeon 5150 is a brand new design on a spanking new process. That’s why Apple switched. Not because of any specific weakness with the G5 as a cheap, but with the inability of IBM and Motorola to keep pace.
Edited 2006-09-01 17:19
Apples doesn’t need to develop chipsets anymore with Intel. They got cheaper cpus that run not as hot and even faster. And graphics included, saving some cents for the GPU slot. And Apple will sell more computers when fueled by Intel chips, that should bring costs furher down.
So why for the pure fact of plain business rules are thoses Intel based Mac minis a lot more expensive than the old, pure selling, outdated and with expensive parts and design manufactured G4 Mac minis?
The cost of a chip is largely due to R&D costs, not manufacturing costs. A Xeon 5150 costs probably about $50 to manufacture. It costs $800, because it cost Intel hundreds of millions to develop the architecture, and billions to build the cutting-edge fab the chip is manufactured in.
At this point in the G4’s lifecycle, it is very likely much cheaper than a Core*. It’s based on an old design (the R&D cost of which was recouped long ago), and its manufactured in less than state-of-the-art fabs. In comparison, the Core Duo/Solo is a relatively new design, and manufactured on a leading-edge 65nm process. The G4 is Motorola’s workhorse embedded chip. The Core* is Intel’s flagship laptop chip. It’s not surprising the latter costs a lot more.
So in comparing cost, you’ve got to compare like to like. It’s probably the case that a 970MP based system costs Apple substantially less than a new Xeon based system. After all, the basic 970 design is three years old at this point, and manufactured on IBM’s mature 90nm process. Moreover, the chipset is a mature design. Meanwhile, the Xeon is a brand new design, on a cutting-edge 65nm process, with a brand new chipset. However, consider what’d happen if Apple had based the Mac Pro on a hypothetical Power 6 derivative. In that case, we’d have a brand new chip on a brand new process, but without the Xeon’s economics of scale. That’d cost Apple much more than they’re paying for the Xeons in the Mac Pro.
Those PPC-fanatics are really annoying…
I do think that’s a little short-sighted.
Why?
There’s something about human nature, well in some people at least, that is not taking the status quo for granted, or accepting monopolies, or always accepting that higher quantity is higher quality.
Some people that know a little of how these CPUs are designed, happen to like, apparently, the idea of the PPC architecture. It’s not as if there is a law of nature that x86 must rule.
In fact, the dead end Ghz race (that has now finally stopped) is one outcome of the Intel dominance (in which PPC was dragged along). When various CPU architectures are around, people aren’t so interested in these meaningless races. The 32bit-64bit transition could have happened long ago, for instance.
PowerPC G5 got too hot, so it was dumped by Apple. But Apple doesn’t even make most of its money selling computers anymore. Why would it be the end for PPC then? Intel has produced bad CPUs in the past and they seem to be still around, I’ve been told.
It would be a waste to throw away PPC use, experience and knowledge, if hundreds of developers are porting to it, and desire to be able to continue to do so in the future.
Is speed really all that’s relevant? Would put Toyota out of business. The G4 I’m using now is the nicest, coolest and quietest computer I’ve worked on.
It isn’t fast, but neither am I.
always accepting that higher quantity is higher quality.
By “higher quantity”, you mean “higher performance, lower price, and higher availability”.
Because that’s what this is about. PPC fanatics are enthralled with the “beauty” of the architecture, while everyone else wonders how they can love a lower-performing, more expensive, and lower-volume product.
Edited 2006-09-01 18:14
I thought it was always known that the G5s were a cheap chip to make. I believe going back there was info showing how the G5 cost less to make then the G4 and the intel competition at the time. None of this is news. The intel switch never had anything to do with price.
Since Apple has abandoned the PPC platform (except for replacement parts as part of legal requirements for extended support, maybe) who is going to be purchasing the PPC 970 mutants for desktop/common computer use now and into the future? Those other one or two makers of PPC motherboards don’t have that large of scale to even come close to what Apple what purchasing, and not even Apple had a large enough purchasing economies of scale to satisfy IBM to the point where progress was made in a timely fashion. Thus, unless IBM suddenly starts making servers with many of these mutants included (IIRC they prefer to use the big brothers of these chips in their server lines) then the reality is that after the relative glut of PPC 970 mutants have been largely sold out of inventory, there is very little incentive for IBM to bother producing many at a low price (and they may not be able to, because of overhead of making short runs as needed, and a lack of desire to keep a bunch in stock, lest they become old stock that can’t be sold readily) or for that matter, invest any more R&D into developing faster versions for non-server use.
In less than 3 years, chances are the PPC 970 mutants will still be at their current level of development, and still be no cheaper than they are now, unless IBM simply sells remaining old stock at a firesale. If you have paid attention to the prices of older types of RAM, for example, this is the sort of thing that happens when something is no longer in much demand: the supply dries up, and the price goes up quite a bit, if you can even find new stock: PC-100 SDRAM is a great example of this. So, too, are many older processors, though quite a few of them have found new life as the basis for embedded use as cores of microcontrollers. However, most embedded processors don’t need the power that the PPC 970 variants offer, and most embedded devices aren’t able to tolerate the heat and power envelope of the PPC 970 mutants, either. Thus, IBM still has a lot of income they can derive from the older G4 (and before) mutations, because for most embedded applications, they’re more than capable of what’s required, and fit within a reasonable power and heat budget.
The practical reality is that the PPC 970 and mutants will likely be the last hurrah for the desktop PPC processor line, while IBM specializes on mutations of the PPC architecture for embedded applications (game consoles and others) and servers.
2-way Opterons includes four memory controllers and two northbridges.