Power4 w/SIMD Rumors hit the Mainstream

in Future Apple Hardware edited January 2014
<a href="http://arstechnica.com/wankerdesk/3q02/power4-apple.html"; target="_blank">http://arstechnica.com/wankerdesk/3q02/power4-apple.html</a>;


When IBM announced the Power4 architecture a while ago, there was an initial rush of enthusiasm in the Mac community along the lines of "Power4 is very big, it's very fast, and it's PowerPC... ohpleasohpleaseohplease..." This enthusiasm was quickly quashed as those in the know applied the wet blanket of reality to the community's hopes: Power4 is expensive, it lacks Altivec, and it's just so far beyond the reach of the desktop in terms of price and power consumption that talk of putting hardware like that in a Mac would be like talking about putting a... (well, you know, I just can't do this without a car analogy, and I refuse to add to the total number of computer-car analogies in the world. So forget it. On with the news story.)

So when the following session description from the Microprocessor Forum 2002 calendar hit the Mac grapevine, it was as if a world-class team of respected medical researchers had announced a very promising cure for old age. Reactions were a mix of guarded optimism, outright skepticism, and old-fashioned dancing in the streets.

Session One: PC Processors

Kevin Krewell, Senior Editor, Microprocessor Report; General Manager, MDR

Breaking Through Compute Intensive Barriers - IBM's New 64-bit PowerPC Microprocessor

Peter Sandon, Senior Processor Architect, Power PC Organization, IBM Microelectronics IBM is disclosing the technical details of a new 64-bit PowerPC microprocessor designed for desktops and entry-level servers. Based on the award winning Power4 design, this processor is an 8-way superscalar design that fully supports Symmetric MultiProcessing. The processor is further enhanced by a vector processing unit implementing over 160 specialized vector instructions and implements a system interface capable of up to 6.4GB/s.

Beside the fact that an IBM-made, SIMD-sportin', 64-bit Power4 derivative for the desktop is a cool thing to contemplate, there are a few very promising things about this announcement in terms of the whole Apple processor picture that have been pointed out in this thread. First, this processor is, in fact, a desktop processor aimed at the PC space (it's in the "PC Processors" section along with Intel's desktop stuff), and right now the PowerPC desktop PC market consists of exactly one company: Apple. Of course, thing's don't necessarily have to stay that way; IBM could very possibly have some kind of grand Power4 + LinuxPPC (or AIX?) scheme for desktop PC domination in mind for this chip. But regardless of who this chip is for, Apple or just IBM, the fact that it's being announced alongside Intel's Banais and targeted at the PC desktop means that it'll probably be affordable and operate under reasonable power constraints. So that dries up a big portion of the aforementioned wet blanket.

The next promising thing is that the Altivec instruction count is 160, just like the unnamed SIMD instruction set in the calendar blurb. So if this chip implements Altivec, then there goes the rest of the wet blanket. As for the other promising things, just read through the thread and decide for yourself which speculations are plausible and which are wishful thinking.

There's some speculation in this eWeek article that this chip can't possibly the Next Big Thing for Apple because if it were IBM wouldn't be able to talk about it publicly. I don't really buy that, because Moto has been known to disclose processors at MPF that wind up in Apple's machines. But I also don't think it's a given that Power4 = G5. First, as I mentioned before IBM is serious about Linux, and this could give LinuxPPC a real boost. But of course, this chip is aimed at the desktop PC market, not the workstation market, and there's not much enthusiasm currently behind Linux on the desktop. Second, this chip could be a very long ways off, so even if it does ever make it into the Mac it might not be for a while. Then again, IBM could just as easily be ready to launch it in October; we won't know either way until it's announced. Finally, there are just too many unknowns to speculate on such a large shift for Apple: from 32 bits to 64, from Moto as their desktop PPC supplier to IBM, and from their current bandwidth starved system architecture to something completely different that could support the Power4. There's also so much that we don't know about Moto's roadmap, and what (if anything) they've got up their sleeves. The Power4 might not make the transition to the desktop very well, and whatever offering that Moto has out when this chip launches could possibly give better performance, especially on Altivec-intensive code. It'll all depend on what the microarchitecture and clock rate look like on the new IBM chip.

What we do know is that regardless of how the Apple-on-Power4 thing shakes out, this represents a big boost for PowerPC and it suggests that the architecture's desktop prospects will no longer languish due to Motorola's neglect. That right there is plenty of cause for enthusiasm about the new chip, whether or not you think it's the future for Apple.

Finally, just so you guys know, when I last mentioned that the Mac Ach was the best source for Apple news on the Web, I got some emails asking for a link to this mysterious Mac news site. Well, here it is: it's Ars's own Macintosh Achaia forum. Check it out, register, and subscribe to support it.


Looks like it's more than just Message Board speculation.


  • Reply 1 of 4
    bigcbigc Posts: 1,224member
    A lot of words about a chip with no details. Time (October) will tell what it all means other than IBM is making new desktop chips.

    To me, the fact that MrNSX sayst that IBM is making something that Apple can use means more than most of the other stuff in the thread.
  • Reply 2 of 4
    *l++*l++ Posts: 129member
    What I find interesting in what has transpired so far is that IBM says that it comes with over 160 new instructions. This happens to be the same count that Altivec has <a href="http://developer.apple.com/hardware/ve/summary.html"; target="_blank">(Apple's Altivec web page)</a>

    MMX/SSE, and so on have a differenrt number of instructions. I reckon that they might be identical instructions. I would think that if IBM did not license Motorola's Altivec instructions, they can still have the same instructions, but would have to use different binary opcodes for them. This is a problem, as it requires a recompile. But at least you do not have to rewrite your vector optimized code.
  • Reply 3 of 4
    msleemslee Posts: 143member
    how the hell is arstechnica mainstream? btw: the ars forum members were covering this long before moki hinted at anything
  • Reply 4 of 4
    fran441fran441 Posts: 3,715member
    Please use one of the current threads dealing with IBM's PPC development.

Sign In or Register to comment.