Other Uses for the PowerPC 970

[jamilecrire]

Originally posted by Foocha


Much as I love Mac OS X, I think it's optimistic to imagine that it could take on Linux in server space, which has the advantage of:
- being free
- running on cheaper hardware
- being a lot faster
- better documentation

In my view, OS X Server has more of a role in file serving in a mixed client environment, where it competes toe-to-toe with Windows 2000 Server, not UNIX.

Let's hope that Apple has got something up its sleeve on the hardware-front. It's pretty key to the future of the company right now - however good the OS and the Apps are, without a great hardware platform to run them on, Apple will be doomed (Be teaches us that a software-only option is no option at all...)

First off I'm a software engineer that maintains my companies and other companies servers (and develops sw for these platforms). Personally I love Linux on the server side (in fact this site runs on linux). However, I run Windows on 1/2 the servers because of a client need or desire. Personally I cannot stand MS products (server side).

I run Solaris, Linux, Windows and OS 9/X at my company and I am seriously considering adding an OS X Server to the mix. All I'm waiting on is Oracle to make it out of release candidate stage (if it ever will). With the available UNIX apps and Oracle I can justify an OS X Server because the UI Controls far exceed what's available on Linux (mind you I've been using Linux since '95). Don't get me wrong, I'll need a lot of proof before I migrate from Linux to OS X Server, but the seed is planted. All Apple needs to continue doing is making good interfaces to the available UNIX based server apps.

Oh yeah I'm waiting on 64bit as well. Oracle needs 64bit, hence Solaris on SPARC.

 

[TMay]

something new and different

http://biz.yahoo.com/iw/030214/051157.html

It looks like IBM has brought in a couple of design houses to support the PPC design process, and, IBM is Licensing the PPC core of (at least) one of the embedded processors.

Throwing down the gloves to ARM and XScale or just an evolutionary business practice? Will Steve put it into one of the Digital Lifestyle devices?

 

[chabig]

Apple's "no comment on unannounced products" is all very well for consumer products like iPod and iLife, but is entirely inappropriate if they are serious about getting into enterprise servers.

I don't know, but I'd guess that if you are a large customer and you are in the market for a lot of enterprise servers, Apple might give you a bit of inside information on their future plans along with an NDA. So although they don't make public comments, they probably aren't keeping all of their customers in the dark...just a guess.

Chris

 

[Foocha]

Big customers under NDAs - maybe, but we're not all big business customers, and we need to make a proper evaluation before making the commitment involved in becoming a big customer.

The likes of Sun and Oracle let you download software for evaluation. The also normally provide product roadmaps that enabled their customers to plan.

It would be good to see Apple adopting some of these practices. When we recently purchased a copy of Mac OS X Server for evaluation, the guys at the Apple store told us that we "may" get a discount on an Xserve if we mention that we previously purchased the server software for evaluation. The only reason we agreed to pay at all was because we're Apple junkies - I suspect regular business customers just wouldn't have considered it, and would have selected another platform instead.

 

[Tim Flynn]

Re: Altivec

Originally posted by Blackcat
Servers have no real need for Altivec, so the fact it is included means something.

On Motorola's website is a document on TCP/IP performance improvements using the Altivec unit. So ... perhaps Altivec is very important for servers!

 

[Tim Flynn]

Re: Re: Altivec

Originally posted by Tim Flynn


On Motorola's website is a document on TCP/IP performance improvements using the Altivec unit. So ... perhaps Altivec is very important for servers!

The link to info on Altivec and TCP/IP
http://e-www.motorola.com/brdata/PDFDB/docs/ALTIVECTCPIPWP.pdf

 

[praetorian_x]

Re: Re: Re: is it worth buying a G4 PowerMac this year?

Originally posted by ryan

By IBM's accounts the 970 is 2x as fast as a G4 running at the same MHz, so a 1.8GHz 970 would be like having a 3.6GHz G4, almost a full 1GHz faster than the combined speed of a dual 1.42GHz G4. Given than most apps and parts of OSX are not able to take full advantage of dual processors I bet the 970 would/will be blazingly fast even in comparison to today’s fastest G4 PowerMac.

Two 1.42 Ghz g4's is *not* equivilent to 2.84 ghz. Dual processing imposes a cost regardless of the architecutre becuase of added complexity. Dual processing as ghetto and FSB-choked as Apple's is even worse.

This is why the 970 is going to kick so much ass, even compared to a dual g4 system clocked similarly: the memory throughput is going to be *gigantic*. And, if the 970 is released as a dual system on the high end, they will likely be using a hypertransport based system, which will make dualies scale better than the current shared bus system.

The 970 will get Apple back in the game.

Originally posted by ryan

But given all that I would agree with your overall assessment, if you need a faster machine now, buy one now. Or as I like to say, the fastest machine of tomorrow won't get the work you need done today completed any quicker.

See above. If you are buying a powermac right now, you had better *really* need one. That's why no one is buying them, especially in the pro space. Everyone can see that Apple either has to go with something new or is going to fade away in the workstation market.

I'm an optimist on this one: I think that the 970's will be announced at MWNY in July. I think that the current move to a single proc low end, dual proc mid and high end is a precursor to get the market "adjusted" to the idea of a single processor system on the low end, so that they don't have to go all duals with the 970.

I also think that they can't put the g4 too much further than the current lineup because of Mhz marketing considerations. The 970 is supposed to be released at 1.2 and 1.8 ghz. Apple is currently at 1 ghz and 1.42 ghz with the g4. They will look silly to the mhz-obsessed tech press if the g4->970 transition doesn't involve at least a small mhz bump. 1->1.2 and 1.42->1.8 would be good. They *might* be able to get away with 1.1->1.2 and 1.6->1.8. But not likely.

So, it is my prediction, based on the usual six month upgrade cycle of the pm's, that we will see the 970's announced in July, shipping in late august/september.

Just speculation,
prat

 

[nickgold]

Altivec on servers

If you are some mad-scientist dude who works for the government and wants to build a cheap cluster for folding DNA or whatnot, an Altivec-bearing 970-based Xserve grid would be pretty good. Those kinds of applications tend to take advantage of vector processing in a big way.

 

[Foocha]

There are maybe one or two customers out there who work in biotech and love Apple's use of PowerPC, but for the rest of us, the RISC chip that we used to be so proud of has become something of an embarrasment.

I hope praetorian_x is correct - it all *sounds* great

 

[Sol]

RISC pride

I am not embarassed of my two RISC chips. RISC is the more sophisticated way of doing things. If it was not for RISC chips Apple would not have produced industry leading portables and fan-less desktops.

 

[eric_n_dfw]

Originally posted by Foocha
There are maybe one or two customers out there who work in biotech and love Apple's use of PowerPC, but for the rest of us, the RISC chip that we used to be so proud of has become something of an embarrasment.

"The rest of us" don't use iMovie, iDVD, Final Cut Pro, Photoshop, iTunes, or iPhoto or much, eh?

All of these, plus OS X's GUI, get major speed boosts on th G4 due to AltiVec. (iDVD doesn't even come up on a G3)

The BEST upgrade I put into my B&W was a G4/400. Many Final Cut Pro renders are 2 to 3 times faster or more!

 

[Foocha]

Sorry - I didn't mean for my post to sound like I was presuming to speak for everyone.

Don't get me wrong - I love Apple software (when it works ) but imagine how much better it would run on a faster chip...

The Intel world overtook PowerPC a long time ago, and PowerPC failed to keep up. Sure, AltiVec is nice for this and that, and enables Apple to keep some pride by doing Intel speed comaprisons with selected Photoshop filters - but for the average Photoshop user, I think they benefit from the overall performance of Pentium.

Who knows, perhaps PowerPC 970 will change all that, but with clock speeds of just 1.8GHz for launch, I'm not holding my breath.

 

[BenRoethig]

Originally posted by Foocha
Sorry - I didn't mean for my post to sound like I was presuming to speak for everyone.

Don't get me wrong - I love Apple software (when it works ) but imagine how much better it would run on a faster chip...

The Intel world overtook PowerPC a long time ago, and PowerPC failed to keep up. Sure, AltiVec is nice for this and that, and enables Apple to keep some pride by doing Intel speed comaprisons with selected Photoshop filters - but for the average Photoshop user, I think they benefit from the overall performance of Pentium.

Who knows, perhaps PowerPC 970 will change all that, but with clock speeds of just 1.8GHz for launch, I'm not holding my breath.

1.8ghz is if it stays at 0.13. It might go to 0.09 before production starts. IBM and Apple may want to look at a numbering system similar to that of AMD.

 

[Foocha]

IBM and Apple may want to look at a numbering system similar to that of AMD.

...agreed. And that says it all, doesn't it? That PowerPC is in the same boat as AMD.

I don't know how Intel are doing it, but they're doing something right. Sadly, problems with legacy code in Carbon apps means it's unlikely we'll see Apple switching to Intel for their desktops any time soon - Xserve, however, may be a different matter all together.

 

[shadowfax]

Originally posted by Foocha


...agreed. And that says it all, doesn't it? That PowerPC is in the same boat as AMD.

I don't know how Intel are doing it, but they're doing something right. Sadly, problems with legacy code in Carbon apps means it's unlikely we'll see Apple switching to Intel for their desktops any time soon - Xserve, however, may be a different matter all together.

why would apple be more likely to switch the Xserve to x86 without also changing the rest of its line?

 

[Foocha]

Hi Shadowfax,

I'd guess that most users don't run Carbon apps on an Xserve, so there's less of an obstacle to migrating OS X servers to Intel.

It's all the desktop apps like Photoshop & Internet Explorer that are Carbon - the OS X Server stuff is all C, C++, Java & Cocoa, quite easy to port to Intel, (and presumably already running on Maklar).

If Apple ever were to decide to migrate to a new chip, I suspect they'd need to do it in stages, and I suspect the server line is the natural place to start. Imagine the reaction of an average iMac user when they double click on an Internet Explorer icon only to get a dialogue box explaining that this app does not have the required binary.

 

[ktlx]

Re: RISC pride

Originally posted by Sol
I am not embarassed of my two RISC chips. RISC is the more sophisticated way of doing things. If it was not for RISC chips Apple would not have produced industry leading portables and fan-less desktops.

The fact that the PowerPC followed the RISC design philosophy has absolutely nothing to do with its power characteristics. In fact, nearly all of the RISC processors used in servers are just as bad for heat and power as Intel's.

The PowerPC has a high performance to heat ratio because the embedded market was so important to Motorola. That led to a set of design tradeoffs to keep the power and heat requirements low. Sun with their UltraSPARC, IBM with their Power4, DEC/Compaq/HP with their Alpha and SGI with their MIPS 1x000 did not have that focus and their RISC chips uses lots of power and generate lots of heat.

The RISC versus CISC debate is for the most part irrelevant. Neither side is very much RISC or CISC any more as the traditional RISC designers have added more CISC-like features and the CISC designers have made their cores more RISC-like and left the CISC ISA more to instruction decoders. ArsTechnica.com has some good papers talking about this and comparing what processors are really doing under their hoods.

And AltiVEC is definitely not in keeping with the RISC design philosophy. It is a wonderful example of pragmatism overriding dogma by adding a great idea from vector processors to enhance an already good design.

 

[Zaid]

On a different note, if Apple does go with the 970, presumably the first place that we would see them is in an XServe and/or a Powermac; but what do you think are the chances of seeing one in a Powerbook soon thereafter?

From what I've read (an ArsTechnica article ) the 970 at 1.2Ghz is cool enough to go into a laptop straight away, and that is if IBM is still using a .13 process when the chip debuts.

Just wondered what everyone's thoughts on the subject were.



* first post btw*

 

[BenRoethig]

I don't think we'll see a 970 in a Powerbook until around 6 months after they reach the Powermacs and the Xserve.

 

[jettredmont]

Originally posted by Foocha
If Apple ever were to decide to migrate to a new chip, I suspect they'd need to do it in stages, and I suspect the server line is the natural place to start. Imagine the reaction of an average iMac user when they double click on an Internet Explorer icon only to get a dialogue box explaining that this app does not have the required binary.

Hmm. I don't think they'd be likely to do it in more than two stages. If Apple moved the PowerMac line to a different CPU architecture, what do you think would happen to their consumer and portable sales? Who would want to buy a computer based on a chip that would soon (within a year or so) be unsupported by new software?

I can maybe see XServe being moved over, then the rest of the lines, but there only because the "servers are different" argument is a whole lot easier to make than the "expandable desktops are different from consumer desktops and portables" argument.

That having been said, for many reasons, I seriously doubt Apple is going to move from PPC to any of the Intel/AMD architectures any time soon (def. not x86; Itanium and Hammer would be the only possibilities, but which would Apple pick? Intel hasn't even made up its mind completely yet!)

 

[GeneR]

Another use for the 970...

Nobody's mentioned the fact that if you crumble the 970 into little pieces and sprinkle it on top of a chocolate sunday, it makes for a nice topping...



Good grief! Just bring those hush-puppies on, already! My knees are killing me from all the begging and waiting...Just kidding!

 

[Foocha]

The neat thing about OS X's application bundles is that you have have multiple binaries withing a single application icon.

If an OS X app can be compiled for Intel, then it should be relatively simple for it to be compiled for PowerPC as well - the obstacle to application developers is migrating from Carbon to Cocoa - once this is done, developing cross platform apps should be relatively straightforward.

My point is that whilst Internet Explorer is unlikely to ever run on OS X for Intel since it is a Carbon app ported from OS 9, your iMac user shouldn't have problems running Cocoa apps developed with an Intel-based PowerMac in mind, since the developer will probably have included binaries for both Intel & PowerPC (remember the 68k / PowerPC transition, it wasn't such a biggie!)

 

[Rocketman]

Re: Re: Re: Altivec

Originally posted by jettredmont
if IBM's primary use for the processor was Apple, and only if Apple was only going to ship a limitted amount of this processor immediately (say, in the XServe line).

This is what I have been saying since the rumour debuted.

A X-serve debut then migrate down to high end powermacs, possibly even positioned as workstations.

Rocketman

 

[eric_n_dfw]

Originally posted by Foocha
The neat thing about OS X's application bundles is that you have have multiple binaries withing a single application icon.

If an OS X app can be compiled for Intel, then it should be relatively simple for it to be compiled for PowerPC as well - the obstacle to application developers is migrating from Carbon to Cocoa - once this is done, developing cross platform apps should be relatively straightforward.

My point is that whilst Internet Explorer is unlikely to ever run on OS X for Intel since it is a Carbon app ported from OS 9, your iMac user shouldn't have problems running Cocoa apps developed with an Intel-based PowerMac in mind, since the developer will probably have included binaries for both Intel & PowerPC (remember the 68k / PowerPC transition, it wasn't such a biggie!)

While you are right, in general, about the cross-platform development possibilities that Cocoa has (and OpenStep had) - there are still going to be parts of the major applications that will need different code for the different platforms. (AltiVec vs. MMX/SSE for instance)
How big of a deal it would be is still an unknown though. (at least to us outsiders)

And as for the 68k / PowerPC transistion: I was not a Mac user then, but from what I've read it sounds like there was a year or so where the PPC machines really sucked because of the amount of 68k emulation going on under the hood.

 

[beatle888]

Originally posted by Foocha
Don't get me wrong - I love Apple software (when it works )

you MUST be a troll, if not, something must
be wrong with your set up. are you running
minimum system requirements for the apple
software? oh and by the way....what apple
software are you reffering to?




.