PowerPC 970 (G5), “ultra-light”

07.11.2018 - Apple, History, PowerPC
PowerPC 970 (G5), “ultra-light”

The last part of the code names of the project PPC 970 (GP-UL) meant “ultra-light”. GP was the code name for CPU POWER 4. Critics accused Apple and IBM of cheating, a PPC G5 was not even just “easy”, but where is the deception in the GP-UL? The fate of the fifth generation PowerPC is widely known, everybody knows what “did not work” that Apple, in desperation, switched to Intel is a reflection of these events in the public consciousness in a straight line and without mercy for the fallen. They say that the Apple story is a story of lies, stupidity and a fantastic gain.

A PowerPC G5 Saga proves it. The figures of the epic, for two years (from June 2003 to June 2005), “changed testimony” and “Dodge”.

First, the fifth generation PowerPC promised the sky with diamonds and incredible achievements “in a year”, then almost prevented them with dirt, claiming that for the development of one unit of capacity (what’s this?) PowerPC consumes 5 times more energy than Intel processors.

But it was all easier and more banal. The potential of the PowerPC G5 has not been implemented and a hundredth part of it, and developed this generation is not worse than the previous three, but the situation in the world changes sometimes too fast.

The developers of the PowerPC is simply not enough time. In 2005, Intel worked on the new architecture of processors (Core) that combines advantages of RISC and CISC, the project was close to completion. Those who are devoted to the details of this project became clear: PowerPC chances for a quick and easy victory no. Among the initiates was Steve jobs.

If not Intel Core, PowerPC defeat would be temporary.

IBM, Freescale, and P. A. Semi (the leading developers of PowerPC in 2005), of course I would take revenge. But Apple, the main consumer of PowerPC in the industry, that in 2005 neither the time nor the desire to wait for revenge was gone. Apple had an alternative.

PowerPC 970 – the first presentation

Date and circumstances of the project GP-UL… unknown. But we know where and when this project was announced publicly. October 15, 2002, at the conference center name Machinery in San Jose, on the second day of the microprocessor forum MPF.

Introduced the PowerPC 970 IBM Microelectronics, senior architect of the PowerPC processors Peter Sandown’s. PowerPC 970 he called “the first in a new family of 64-bit high performance PowerPC processors”.

The origin of the PowerPC 970 was almost Royal: it was developed on the basis of POWER 4, one of the well-established processors heavyweights for particularly important and responsible task. Its design capacity was sacrificed for reliability, intentionally: applications error would cost too much, and this paranoia was appropriate.

PowerPC 970 paranoid measures were excessive, “ultra-lightweight”, compared to its ancestor, the processor, nothing prevented IBM engineers to push the gas pedal to the floor and to disperse their offspring to unimaginable POWER for 4 clocks.

In the popular creation of “light” means “weakened”, but the reality is paradoxical (it is the main property): PowerPC 970 was not inferior to his ancestor responsible for almost anything, and for several very important properties even superior to him.


As POWER 4 and PowerPC 970 was a “real 64-bit processor”. 64-bit was his entrails: and registers, and line data. Moreover, everything related to AltiVec (vector processor modules) was 128-bit… the Townsfolk were delighted.

PowerPC 970 (G5), “ultra-light”

However, the increase in bit depth had nothing to do with performance, nor to power – but the power of the psychological effect was a necessary. However, the main advantage of the POWER 4 and PowerPC 970 was support for 32-bit code. That is, no 32 – and 64-bit modes or options of programs is not required.

From the point of view of the user and the application programmer, both varieties of programs just worked.

The number of registers of the imagination: a 32 64-bit integer register 32 64-bit registers for floating point, 32 128-bit register to the vector unit and registers 48 of each of these data types for “rename”.

8 Executive modules – two for integer operations, two for floating-point arithmetic, vector operations and two for memory (load and unload). Different pipelines for different data types, different number of stages: from 17 to 25 for an integer vector.

Technical details would take several more pages, but instead we’ll just summarize: engineers at IBM, in collaboration with Apple (the “North bridge”, it was to find out I could not) have created a real masterpiece.

At the IBM plant in new York state, very automated and almost deserted shop in a few months (Apple event) was shown to the public – like the “Apple” production theme – soon began manufacturing these processors. By 15 October 2002, the new processor has existed in several dozen instances, and its performance was reported in the subjunctive mood.

From a physical point of view, the CPU was a great technological process of 0.13 µm, 8-layer SOI (SOI, silicon-on-insulator) copper-based, copper interconnects and other “Goodies”.

PowerPC 970, the invasion of reality

PowerPC 970 (G5), “ultra-light”

In the beginning of the century in the industry of microprocessors there was a fierce war. For the clock frequency. IBM wasn’t involved in it: for leadership fought by Intel and AMD. The most direct way to increase processor performance is increasing the clock frequency.

The way this is… sick and end, but is simple and effective – from time to time, and buyers are firmly hooked on these numbers: 2.5 GHz, 2.8 GHz, 3.0 GHZ…

But the emergence of the PowerPC 970, which at lower clock frequencies, due to the engineering art of its creators, was not inferior to the best specimens of the warring parties, has not escaped their attention. Perhaps it was this shock accelerated the development of the project of the new architecture of Intel processors, combines the best of RISC and CISC…

While everything looked rosy and promising. At WWDC 2003, Steve introduced the PowerPC 970 (is a trademark of IBM), also known as PowerPC G5 (trademark of Apple), and the first computers based on this processor.

It was a powerful Mac desktop intended for professionals. Once again the Mac and superior to “fast PC” is in all imaginable respects – including in terms of complexity of the cooling system.

The new CPU was “hot” not only in performance, but in General.

In a perfect world this would be no problem. The first goal (faster processor with premium features) was successfully achieved. The revolution was accomplished.

Now, the engineers of IBM were other primary task: to move to 90-nanometer process (and in the future and 65), to reduce the energy consumption at the same or even better performance, reduce heat… To overcome obstacles to these goals is the job of the engineer.

Perhaps, for the 2006-2007 year, they could succeed… But no luck.

In 2003-2004, according to experts, Apple will translate to a new and radically more powerful PowerPC fifth generation of the upper part of its product line, the PowerMac and PowerBook (encouraging news from IBM, could seem to hope for it), and in consumer computers (eMac, iMac and iBook) will be used more and more advanced and efficient modification of the processor of the fourth generation, until…

As we know today, experts have struck out twice in these projections. IBM was able to reduce the heat transfer one of the developed modifications of the PowerPC 970, but for use in portable computers, he didn’t answer: he was too greedy for that, even in the “sleep”.

The “home” desktop computers experts also missed. iMac G5 still came out, and made his appearance much noise. Thermodynamics of his body, too, was trivial, but about the energy consumption was modestly silent.

And the legislator in the field of portable computers within two to three years not disturb the peace of new masterpieces already had no opportunity. The consequences of the issues to one single company for industry processor was much more dramatic than these problems…

To be continued

Оставить комментарий

Notify of