IBM AltiVec

[arn]

Architosh posts some speculation/interpretation about two posts regarding gcc.

The tentative conclusion is that IBM's PowerPC64's supports Altivec. Speculation of the new chip started after a description of an upcoming chip appeared on the conference schedule for Microprocessor Forum 2002. Since then, rumors of Apple-IBM talks over the new chip emerged. PenguinPPC initially reported that the vector unit on the IBM chip was Altivec compatible, but that post has since been removed from PenguinPPC, which questions its original accuracy.

 

[bobky]

Woot!

 

[vniow]

Well, I'm not suprised. It's still good news.

 

[amnesiac1984]

will my brand new DDR powermac be upgradeable to 64bit IBM power4 based chips in a year or so? Otherwise I'll just shoot my own foot while eating my hat.

 

[avkills]

Yes!

This is great news. I have no doubt that if IBM delivers, Apple will deliver and the Wintel world will be in tears.

8-way superscalar, 64bit, Alitvec and speeds up to 2Ghz. Bring it IBM, and let us put the final nail in the Wintel coffin.

-mark

 

[ffakr]

Originally posted by amnesiac1984
will my brand new DDR powermac be upgradeable to 64bit IBM power4 based chips in a year or so? Otherwise I'll just shoot my own foot while eating my hat.

I'd almost guarantee it is not compatable. IBM will not make a high speed 64 bit processor that runs on a 64bit MPX system bus.... Unless upgrade makers could do some daughter card hocus pocus (like getting a G3 to work on a 33MHz bus nubus machine). It'd castrate the chip though.

 

[vniow]

I'm not so sure that it wouldn't be just a pop in daughtercard. They redisigned the whole case, created a new motherboard with DDR, flipped it around so air can have a pass-through from front to back, a giant heatsink and fan. They didn't do this for no reason. The upgrades for all that are miniscule; a pair of 1.2 GHz G4 with 167 Mhz bus does not need that much modification. Ive got a feeling that this case is for something more.

 

[avkills]

Current PMacs

I doubt the current Motherboards would support the chip to its full potential. For one, it looks like the chip uses HyperTransport for the FSB. But then again, Apple made the B&W motherboard so a G4 could just pop in.

My gut feeling is it would be a new motherboard. But then again, switching MB designs on the Xserve and the Current Towers so fast would be kinda strange I think.

Oh well, whatever happens, Apple had better use this chip whether it has Altivec or not. Motorola needs to be taught a lesson here.

-mark

 

[barkmonster]

I think 2Ghz is highly optimistic as a "starting at" speed, If they debut at 2Ghz maybe we'll see macs with 1.6, 1.8 and 2Ghz chips with a dual core 2Ghz chip as the Ultimate model so it's effectively a dual 2Ghz.

Apple are long overdue for a massive speed hike from either IBM or Motorola, if we're at 1.25Ghz right now and the jump from the PPC7410 to the PPC7450 gave us an increase in clockspeed of over 45%, a 2Ghz 64bit PowerPC chip with Altivec would be the perfect upgrade for the powermac line in February or March of 2003. 1.8Ghz wouldn't be too shabby either.

There's also the fact that these chips will be running most of the code in 32bit till people recompile applications that can take advantage of a 64bit chip, that gives us even more speed to look forward to in the future. It could be like the transition from G3 to G4, there were few applications that took advantage of altivec at first, after a year or so there were a lot of optimised applications using it. I can only imagine what a true DDR motherboard, Altivec AND a 64bit chip with fully optimised code could do for software synthesis and plug-in effects.

 

[Catfish_Man]

Re: Current PMacs

Originally posted by avkills
I doubt the current Motherboards would support the chip to its full potential. For one, it looks like the chip uses HyperTransport for the FSB. But then again, Apple made the B&W motherboard so a G4 could just pop in.

My gut feeling is it would be a new motherboard. But then again, switching MB designs on the Xserve and the Current Towers so fast would be kinda strange I think.

Oh well, whatever happens, Apple had better use this chip whether it has Altivec or not. Motorola needs to be taught a lesson here.

-mark

HyperTransport is not a system bus, it's for linking components on the motherboard together. IBM supports the RapidIO bus, which is similar to HyperTransport, but more in use in embedded processors (it's also better than HT according a very knowlegable person on arstechnica). Since this new chip is based on the Power4, I imagine it will use either a cut down version of the Power4's bus, or an on chip memory controller. Teaching Motorola a lesson would be stupid. Being stuck with IBM would be just as bad as being stuck with Motorola. Apple needs to have two suppliers competing. Using a chip without Altivec would be equally/more stupid, because 10.2 is MORE altivec optimized than 10.1 was, and not using Altivec would trash all the optimization work of the past 4 years (probably resulting in something that ran significantly slower than the current powermacs).

 

[daijones]

Read the rumors more closely and it sounds pretty bad for Apple. The rumors suggest that IBM is planning to release a Linux box under their own name. 1: an IBM brand linux box is going to be a big seller, taking sales off the power-hungry UNIX users Apple needs to be attracting; 2: if IBM want to sell this box, it would be in their interests NOT to sell the new chip to Apple, to reduce competition. I really hope Apple has maintained contact with IBM and has OS X compiled ready to put this chip into a box, because if IBM get much of a headstart with a 64-bit linux desktop the scientific market is lost to Apple before it even got going.

 

[noht*]

[shameless self-propaganda & view-count-fishing]
i told you so...
in this thread. [/shameless etc.]



noht

 

[synergy]

Originally posted by daijones
Read the rumors more closely and it sounds pretty bad for Apple. The rumors suggest that IBM is planning to release a Linux box under their own name. 1: an IBM brand linux box is going to be a big seller, taking sales off the power-hungry UNIX users Apple needs to be attracting; 2: if IBM want to sell this box, it would be in their interests NOT to sell the new chip to Apple, to reduce competition. I really hope Apple has maintained contact with IBM and has OS X compiled ready to put this chip into a box, because if IBM get much of a headstart with a 64-bit linux desktop the scientific market is lost to Apple before it even got going.

IBM high end boxes are just that, high end boxes. Not very cheap.
IBM sells CPUs to whoever will buy them as well. Apple is not in the high end server market either. The Xserve is not a high end server with max of dual processors. Some of those beasts come stacked with 10 or more processors.

 

[daijones]

The processor IBM are announcing in October is targetted at desktops and low-end servers: describes Apple perfectly. But it also describes a potential linux based range of products. I don't know any box manufacturers making PowerPC based boxes for linux, so I really hope the chip is going to Apple. It could be that IBM is thinking to use the chip in workstations, and also sell to Apple (and anyone who fancies becoming a linux box maker) to get the volumes up enough to make design/production more financially viable. But my point is, if alternatively IBM have decided to broaden their range of products to include low-end hardware, then this could be bad news for Apple, not good.

 

[MacCoaster]

For clueless people:

READ.

It says RS6000 down there. Hell:

Index: gcc/config/rs6000/linux64.h
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/rs6000/linux64.h,v
retrieving revision 1.25
diff -u -p -r1.25 linux64.h
--- gcc/config/rs6000/linux64.h 8 Aug 2002 17:55:37 -0000 1.25
+++ gcc/config/rs6000/linux64.h 24 Aug 2002 12:39:32 -0000

This is simply a diff file. --- means remove this line, +++ means add this line. Simply updated the stuff. Nothing big.

Doesn't mean it's for Macs or possibly for Macs.

As much as I want Apple to go IBM. Motorola is the only way they're going and we have to live with that.

 

[arn]

Originally posted by MacCoaster
For clueless people:

READ.

It says RS6000 down there. Hell:

So, what you are saying that the RS6000 has Altivec?

Still doesn't jive.

arn

 

[ddtlm]

MacCoaster:

I'm not sure how you conclude that IBM is not planning on offering a "Power-Lite" to Apple based on a two-line diff. Quite a leap of logic there.

 

[noht*]

about the rs/6000:
it can't 'have' or 'have not' altivec, as it's just a server architecture, not a processor; indeed, rs/6000 was rebranded the 'pSeries' more than two years ago. i guess it's still in the gcc config files for backward compatibility.

about the changes in gcc:
it's clear that this patch adds the altivec_target to the linux64 build of gcc. but why would this be done when there is no 64-bit processor that supports altivec? just for those users that want to compile osx/darwin/ppclinux apps on a rs/6000/pSeries...? kinda strange...

just some thoughts.

noht

[edit: "more than two years ago." -> it was fall 2000, actually...]

 

[AmigaMac]

I don't know any box manufacturers making PowerPC based boxes for linux, so I really hope the chip is going to Apple.

Actually there is one PPC/Linux vendor out there that is also an Apple reseller that builds its own boxes...

YellowDogLinux

 

[avkills]

Re: Re: Current PMacs

Originally posted by Catfish_Man


HyperTransport is not a system bus, it's for linking components on the motherboard together. IBM supports the RapidIO bus, which is similar to HyperTransport, but more in use in embedded processors (it's also better than HT according a very knowlegable person on arstechnica). Since this new chip is based on the Power4, I imagine it will use either a cut down version of the Power4's bus, or an on chip memory controller. Teaching Motorola a lesson would be stupid. Being stuck with IBM would be just as bad as being stuck with Motorola. Apple needs to have two suppliers competing. Using a chip without Altivec would be equally/more stupid, because 10.2 is MORE altivec optimized than 10.1 was, and not using Altivec would trash all the optimization work of the past 4 years (probably resulting in something that ran significantly slower than the current powermacs).

Ok, sorry about getting Hypertransport and RapidIO mixed up. All I know is it supports FSB transfers over 6GB/sec, which is damn fast.

I agree that using a chip w/o altivec for the Pro machines would be a mistake. But I also know that Motorola has been sucking wind lately. It is plain idiotic that the 750fx G3, which is in the iBook, supports speeds up to 1Ghz and a 200Mhz FSB bus that supports DDR and Apple will not use it to it's full potential. Altivec or not, I bet that thing would kick some serious booty running in that config.

Let's just hope that whoever supplies Apples next chip, it will be at or above 2Ghz, with a speedy FSB that supports DDR and some fast L2 & L3 cache.

-mark

 

[Rocketman]

Re: Yes!

Originally posted by avkills
This is great news. I have no doubt that if IBM delivers, Apple will deliver and the Wintel world will be in tears.

8-way superscalar, 64bit, Alitvec and speeds up to 2Ghz. Bring it IBM, and let us put the final nail in the Wintel coffin.

-mark

Technology is one thing but price is why so many crappy computers get sold to unsuspecting Wintel weenies.

If those crappy boxes are $200 less retail with a distributed distribution channel and 10% more margin for marketing, there is no hope against evil.

Rocketman

 

[DannyZR2]

What exactly does "8-way superscalar" mean????

 

[e-coli]

I know IBM as a company like the back of my hand. They've never made on single red cent on PC's. So a new low-end personal computer seems highly, and I emphasize highly, unlikely. It was just last year that IBM seriously considered selling off their personal computer division.

trust me. No threat to Apple.

 

[avkills]

Superscalar

Originally posted by DannyZR2
What exactly does "8-way superscalar" mean????

I think it affects how many operations can be performed in a single clock cycle. I could be wrong though. The current G4 is 4-way superscalar.

-mark

 

[dongmin]

Originally posted by e-coli
I know IBM as a company like the back of my hand. They've never made on single red cent on PC's. So a new low-end personal computer seems highly, and I emphasize highly, unlikely. It was just last year that IBM seriously considered selling off their personal computer division.

trust me. No threat to Apple.

I don't think anyone is suggesting that IBM is getting into the "low-end personal computer" market, but rather the high-end desktop market and low-end server[/B] market. One possible strategy might be the consumer Macs being outfitted with Motorola G4s and the pro desktops being outfitted with the Power4-variant.