64-Bit Jaguar and 970 PowerMacs

[soggywulf]

Originally posted by Shadowfax
sorry, that's pathetic. you should learn to write G4-optimized code before you get all disappointed. it still is a pretty competitive chip with the right software.

How do you know? Are you a developer?

Originally posted by Shadowfax
to be disappointed that a chip won't outperform another chip with an OS that is specifically written for that chip and not for the new one is kinda stupid, too, IMO.

This isn't like 68K->PPC. This is a new chip with almost the same (or is it exactly the same?) instruction set as its 3 year old predecessor. It had damn well better be a lot faster, with the same code. Fine, optimizations help...but they don't excuse. Expect better from Apple.

 

[nixd2001]

Originally posted by jettredmont
Your G4 software will not work on a 970 if your software happens to be an OS kernel.

The gcc modifications are kernel-level. OS X apps will not require G4-970 modification from all indications.

Do note that IBM is on record saying that there are "minor" changes required to the OS for the 970 relative to the G3/G4 lines. "Minor" to the hardware vendor rarely means "minor" to the software provider ...

I think we're losing track here a bit.

1) Most peoples' code isn't kernel stuff - just the way it is.

2) Instruction scheduling applies to anything - not just kernel code. No modifications are required for correct operation, but it's highly likely that modifications are required for optimal performance.

3) If a system has been sensibly designed - and Mac OS X seems to score on this basis - then the vast majority of time is spent in user code. This is where GCC optimisations will provide performance benefit - but that ain't the same as it will be incorrect (or "not work") without such changes.

 

[ddtlm]

Shadowfax:

sorry, that's pathetic. you should learn to write G4-optimized code before you get all disappointed. it still is a pretty competitive chip with the right software.

I'm pathetic because I don't waste my time hand-coding for a obsolete processor? Kiss my rear end! What drives you to defend such a hopelessly loosing position to the point where you attack people who won't waste their time keeping some illusion of compeditive performance alive?

andyduncan:

Bah.

d00d I'm slippery!

And yet there are still compiler optimizations that can bring out more speed for athlon systems, just look at some of the good linux software.

And this would be fine by me on a PPC970 too... I just wanted more Athlonism and less Pentiumfourism.

Sure the 970 is targeting the same market as the g4, but why redesign the chip just so someone doesn't have to recompile on the next point release

Of course they shouldn't redesign it now, but up to this point there was little reason to expect such bad IPC on legacy code. I have always thought that it didn't suffer from this legacy code IPC problem at all, and perhaps in the end that will be the case. But regardless of that discussion, my opinion remains that G4 IPC on legacy code is not very good at all.

Edit, added: It may be that a PPC-970 at 1.8ghz is more-or-less as fast as a Athlon barton at 2.2ghz (the 3200+ perhaps). The Athlon does good on old code, why can't we expect (in an uneducated way) the same from a PPC970? I wasn't looking for a technical response to that, its just one of those philosophical questions.

 

[RandomDeadHead]

All I know is that we need them fast, real fast!

 

[jettredmont]

Originally posted by nixd2001
2) Instruction scheduling applies to anything - not just kernel code. No modifications are required for correct operation, but it's highly likely that modifications are required for optimal performance.

I must have misread the article. It said "scheduling"; I thought "process scheduling" (which is a kernel-level concept that ordinary processes would all benefit from performance-wise but need not be recompiled).

You could be correct in that the gcc optimizations eWeek is talking about are instruction scheduling optimizations (which I took as a given for any processor out there ... every compiler can benefit from processor-specific instruction scheduling...) That would mean you'd have to recompile your apps to get the benefit (obviously) except where you're calling into system libraries that already take advantage of the latest gcc.

 

[bertagert]

Re: I can explain the Smeagol name...

Originally posted by Stike
SJ can be seen in a completely dark room, only illuminated by the pale screen light.

SJ: "Get out!! Leave the lights out! I need more testing! MorrE! moOOorrree!"
(Developer2 instantly closes door)

SJ: "My precioussss...!!"

That was great!!!!

For everyone else: We only have two weeks to find out whats going on. No one here can say how well things will perform on the 970. Lets wait a little and find out.

 

[bertagert]

Originally posted by RandomDeadHead
All I know is that we need them fast, real fast!

I second that!

 

[copperpipe]

Oh joy of Joys

So it's coming. In a couple of weeks, it, they, will be here. It will be a big leap for the speed of our beloved Macs. It might, it just might, be a HUGE leap. But we will have to wait an exhausting T-minus 14 days. Some of us might not make it. Blown fuses, overheating systems, and insufficient brakes will probably take a few out before the event. Poor bastards will be laying in a puddle of their own vomit, while muttering "powa-mah nine-seveny". I don't know about you guys, but I'm gonna drink solidly over the next two weeks in order to keep the pressure down.

 

[Sun Baked]

Originally posted by Shadowfax
LOL
LOL
LOL

sorry, that's pathetic. you should learn to write G4-optimized code before you get all disappointed. it still is a pretty competitive chip with the right software.

to be disappointed that a chip won't outperform another chip with an OS that is specifically written for that chip and not for the new one is kinda stupid, too, IMO.

If you follow some of the discussions of the compilers you'll see that they whine about the sad state of the actual "optimization" for the G4 that the compiler generates.

Since Apple is using the same tools as they offer to the developers, so there is an incentive to improve the tools for everyone.

 

[shadowfax]

Originally posted by soggywulf
How do you know? Are you a developer?

whatever the validity of your point. this is a stupid question. you don't have to work at a mint to know a little something about the value of money. don't bait me.

 

[ddtlm]

Shadowfax:

you don't have to work at a mint to know a little something about the value of money

Nor do you have to be a programmer to know the usefullness of software. Neither statment has anything to do with your claims about the merit of hand-optimizing for G4's, which is something a programmer is much more likely to know than most, hence the question soggywulf posed.

 

[shadowfax]

Originally posted by ddtlm
Shadowfax:


I'm pathetic because I don't waste my time hand-coding for a obsolete processor? Kiss my rear end! What drives you to defend such a hopelessly loosing position to the point where you attack people who won't waste their time keeping some illusion of compeditive performance alive?

i never said you were pathetic, i said that your comment that the code you write works better on x86 than on the G4 was. i am sorry. i don't honestly know that much about what i am talkinf about, but i do know that the mac platform gets a lot of pisspoor performing games and other apps because of poor porting skills. but whatever. sorry.

 

[shadowfax]

Originally posted by ddtlm
Shadowfax:


Nor do you have to be a programmer to know the usefullness of software. Neither statment has anything to do with your claims about the merit of hand-optimizing for G4's, which is something a programmer is much more likely to know than most, hence the question soggywulf posed.

the analogy wasn't intended to be that perfect. i can also appreciate the merits of optimizing code, not being a programmer (yet). you stop baiting me too.

 

[ddtlm]

Shadowfax:

Ah, I see. No hard feelings here. Well, anyway the best porting jobs are the ones where all you need know is how to recompile. Honestly I don't have time to hand-code for any chip, I much prefer to try to make things easy for compilers to speed up where I can. You'll find that even the really good programmers like Carmack are heading that way.

Sun Baked:

Yeah with better compilers I bet the G4 would fare better, but they claimed GCC 3.x was a lot better than 2.9.x anyway. I never really noticed anything major.

 

[Silencio]

Q37

Q37.... Q37...

If the first PowerMacs were released in March 1994, then PowerPCs have been in Apple machines for just 9-1/4 years, or 37 fiscal quarters. Q37.

Well, that's my best guess.

 

[Frobozz]

Originally posted by rjstanford

Of course, there's only one problem. Can anyone confirm the gcc issues? Is this really so significantly different? If so, that would mean that software built for the G4 series wouldn't run on the 970 without modification, which is something that I have a hard time believing (from previous statements and from IBM history).

I'm late to this thread, so someone may have addressed this, but here's my take.

The above issue really isn't as dire as it sounds. Essentially, everything will work just fine on a 970 based Mac with the modified GCC 3.1 compiled Jaguar branch. The only real issue is when can Apple sync GCC 3.3 with Panther 10.3 for a release. Obviously, when they can do this, they will have all the advantages of the 970 CPU's taken into account.

Right now, it seems that the 970 _chips_ with the 10.2 branch will be on speed par with a G4 of the same speed (if you were to put it in the same hardware and run the same OS.) This does _not_ take into account architectural speed differences in the computers, which will make the machines seem much faster than if you just plopped a 1.8 G4 in an existing machine.

Overall, a good article to fill in some gaps.

 

[Frobozz]

Originally posted by Gymnut
Is the "precious" the 970?

boo-yah!

Smeagol. Such an interesting name. Has so many interpretations. Now, I hope they didn't pick it just because they liked the movie.

 

[Frobozz]

Originally posted by ac2102
I thought that Panther was going to be ready by the WWDC? Surely the OS will be ready before the hardware comes into the shops. Or does it take a while after the WWDC for the new OS to be released? Can anyone explain this to me.....

Oddly, I'd bet the hardware is done. Since so many last minute fixes, enhancements, etc. factor into the announcment Jobs will make, I have no doubt that the release date can vary. In fact, I don't think any rumor site can predict with 100% certainty until 24 hours before the keynote. There's simply too many factors. I'm happy if they are released ASAP in June, but I'd be just as happy with a July to August timeframe. I won't be buying until revision 2 anyway, since I have a dual 1 GHz machine now. In many ways, this machine has held up much better, performance wise, than my previous computers.

 

[AidenShaw]

Make Panther slower on G4s.... Hmmm....

Using a new compiler to optimize for the 970 would more than likely imply that it will slow down the G4 (unless the gcc today is really that poor on PPC).

The other side of the coin is that massively o-o-o (out-of-order execution) chips like the 970 should be less sensitive to instruction scheduling. The chip itself is able to dynamically re-order the instructions to fit the memory model and execution units.

An improved compiler will seldom give you more than a few percent faster. (This is assuming that the compiler was already nearly optimum for the previous chip implementation.)

I would be extremely surprised if the new gcc gives more than 5% speedup to the 970 - unless the old version was criminally bad at PPC optimizations. (If that's the case, G4 systems should get radically faster as well....)

 

[Frobozz]

Originally posted by nagromme
The article says that without Panther, the 970 will run with the performance of a G4. (But at slightly higher MHz than we have now, and with a faster system bus.)

I have to question that. I can't see Apple releasing 970 Macs, yet dilluting their impact by having only minor speed increases until a sofware update months later. If the impact of Panther were THAT important, I'd expect Apple to wait for it. But every other report seems to suggest that Panther is NOT needed for big speed gains on the 970.

Well, I think the article has some ambiguity in it. I interpret the statement you are referencing to mean that if you were to NOT take into account motherboard changes, this would be try. However, my interpretation could be wrong.

 

[ddtlm]

Frobozz:

I interpret the statement you are referencing to mean that if you were to NOT take into account motherboard changes, this would be [true]. However, my interpretation could be wrong.

I wonder how they would be able to separate motherboard performance from processsor performance and declare which is responsible for what. Hopefully this launch is more dramatic than the arrival of the G4-733.

 

[Sun Baked]

Here is a discussion on Apple's Compiler Future that was broken off the Future Processor Thread.

 

[soggywulf]

Originally posted by Shadowfax
i never said you were pathetic

Well, yeah you kinda did. Anyway, water under bridge etc.

Originally posted by Shadowfax
but i do know that the mac platform gets a lot of pisspoor performing games and other apps because of poor porting skills.

I don't think it is poor porting. I recall that a number of seemingly highly talented developers are porting to the Mac (id, Omni, the guy who did Unreal--forget his name, etc). OTOH, it is for sure that our hardware is slow at the moment. That is more likely the cause of poor game performance.

 

[soggywulf]

Re: Make Panther slower on G4s.... Hmmm....

Originally posted by AidenShaw
An improved compiler will seldom give you more than a few percent faster. (This is assuming that the compiler was already nearly optimum for the previous chip implementation.)

I would be extremely surprised if the new gcc gives more than 5% speedup to the 970 - unless the old version was criminally bad at PPC optimizations. (If that's the case, G4 systems should get radically faster as well....)

This makes the most sense to me. I think the new systems are going to be hella fast right out of the box, even with Denethor or Bombadil or whoever.

As long as they include a 9800 or 5900, of course.

 

[shadowfax]

Originally posted by soggywulf
Well, yeah you kinda did. Anyway, water under bridge etc.

there's a difference between telling someone "you are pathetic" and telling someone "what you just said was pathetic. if you care to reread my post. you'll note that it falls on the aft side of that particular spectrum. but hey, i wasn't even talking to you; we've resolved our differences, more or less. you should have left it be.

I don't think it is poor porting. I recall that a number of seemingly highly talented developers are porting to the Mac (id, Omni, the guy who did Unreal--forget his name, etc). OTOH, it is for sure that our hardware is slow at the moment. That is more likely the cause of poor game performance.

i haven't messed with UT, it's not my style at all. i had BG2 in mind, which was "updated" to work in OS X. it has a "built for mac OS X" sticker on it--bah. also, more recently, a game called "BloodRayne" ported by Aspyr. it's a terribly dirty game in terms of performance. and it's not my hardware's fault.

in the case of BG 2, it runs better on my OLD PC--933 MHz P3, 512 MB of RAM and a GeForce 3, all of which are slower than the specs in my TiBook, barring the hard drive speed.