IBM PPC 970 to hit 2.5ghz

[BlackMonk]

Aye, I believe the PPC970 uses something called the "AIO" or "Apple I/O" bus instead of the G3/G4 Maxbus (maybe same bus for earlier 60x CPUs?)  450 MHz double-pumped for an effective 900 MHz rate.  And the interface, while not explicitly HyperTransport, seems to share some of the traits.  It would make sense considering that 1) Apple is part of the HT consortium and 2) word on the street is that Apple helped IBM design the system bus part of things.

IBM is an enterprise.  They sell 32-bit PowerPC CPUs.  They know that their customers do NOT want to rewrite all their applications in order to support a new CPU.  I would rather think that switching from a G4 to a PPC970 would entail either extremely minimal code changes or perhaps none at all.  IBM's customers would be angry if IBM broke legacy application support--look at how UltraSPARC, PA-RISC, MIPS, Alpha, and x86 are still around.  Their strength lies in part that a newer CPU works with applications written on an older CPU.

The POWER/PowerPC instruction set was defined as a 64/32 bit one something like 10 years ago, so it isn't like IBM is pulling this out of thin air either.

I think it's been known that the PPC970 SIMD instructions are "Altivec" for a while, now.  IBM might not have named them explicitly, but pretty much everyone knew.  Well, everyone who was following this CPU, that is.  

The only REAL issue with this is support.  Apple co-designed the interface.  IBM is integrating it into their workstations and blades and servers.  Who is going to develop and market a relatively low-cost motherboard?  It took HOW long to get these G3/G4 motherboards on the market?  And they are missing HOW many "modern" x86 features?

Maybe IBM will allow people to OEM their workstation motherboards and this won't be an issue.  Still, though, I'd think that the motherboard would be THE primary concern for using this new CPU with any future Amiga.

[BlackMonk]

Yeah, though considering the market they have with Apple, going their own way with the SIMD stuff would be shooting themselves in the foot, to an extent.  Why make the extra effort to make an incompatible solution when you can use an already pre-defined solution that there's a market for?  Knowhadamean?

But yeah, it's nice that they are solidly labelling it as Altivec.

Psy wrote:

Right but before Apple couldn't drop its prices becouse there was nothing to replace the aging duel 1.42 Power Macs.

Aging?  The dual 1.42 GHz PowerMacs only just came out like a month ago.  I'm not even sure if they are shipping yet.  Apple has been keeping up a 3-tiered pricepoint on their desktop "pro" line for a while.  I'd say even since the 7x00/8x00/9x00 days.  The new PPC970 systems would either displace all the G4 desktops or would be situated at the top prices, so around $3500 for the top of the line model.

downix wrote:

To take advantage of the 64-bit code, of course the OS and apps will have to be re-written.

To elaborate even further, apparently IBM has been contributing to the source code for Darwin, OS X's underlying foundation, for a while now.  And even a year ago there were references in the source code that IBM submitted that referenced SIMD instructions.  From what I gather as well, IBM (or someone) has been adding 64-bit extensions to bits and pieces for a while.  Since the underlying PPC foundation (Darwin) can be run on other IBM systems, the justification was that IBM was tweaking it to run on their POWER3/POWER4 workstations as an evaluation to replace AIX or some such thing.

Anyway, I don't think it will 1) be too much of an effort to get some real benefit from the 64-bit extensions (both memory and number crunching) and 2) there already seems to have been work done towards this anyway which would imply that they knew about the switch to 64 bits for a while and that they've been planning it for a while.

Heck, with the long lead-in time until the PPC970 release, I'm sure some Apple developers have prototype PPC970 systems that they can use to prepare the OS for when Apple releases the new systems.  Either way, Apple gets a performance boost regardless of the OS taking specific advantage of the new features.

Oh, and there's been some talk on one of the Mac messageboards I frequent that IBM tends to be conservative on their performance estimates.  And that the original clockspeed estimates were made last October so they've been refining their processes in the intervening months which helps account for the magically-higher clockspeeds.

Someone brought up concerns that those clockspeeds were guesses for moving to the .09 and .065 micron processes (did I move the decimal point too much?) but the press release specifically lists those speeds for the .13 micron process which is what the first run of PPC970 CPUs will be created with.

Anyway, enough o' my yappin'!