PowerBook To Get Dual-Core G4?

[macsrus]

64-bit is not faster than 32-bit. It can address more memory, but that can actually cause a slight slowdown.

Don't believe in the "64-bit myth" - unless you have more than 4 GiB of RAM in your 'book, you almost certainly won't benefit from 64-bits!

WHILE This is true and can actually slow down the results because of the way registers are used.... during OS and code testing we found the the opposite results


Using 32 bit code with OS X as our baseline


64 bit code on OS X 3.1% performance gain over baseline

32 bit linux... 32 bit code... 3.8% performance gain over baseline

32 bit code on 64 bit linux 4.9% performance gain over baseline

64 bit code on 64 bit linux 7.9 % performance gain over baseline

Because of these results we are dropping OS X on our cluster and switching to LINUX

 

[macsrus]

Tell you what. If I could get a choice between a dual G4 running at 3ghz tomorrow and a dual G5 running at 2.5ghz, I'd certainly consider the G4 - especially given its better altivec performance.

I dont know where you got the idea.... That the G4 has better altivec performance than the G5... Because every bench mark and Document I have every seen has showed the G5's altivec to be from same to slightly superior to the G4

 

[adamcoop]

Actually Apple explicitly stated that the new iMac will be G5 based during last quarters earning meeting.

Yeah - I knew I'd be flamed for that one... 😛

 

[mikeyredk]

couldn't they rebrand the chip from freescale with the g5 name? it will be confusing for us but they could 😉

 

[macsrus]

couldn't they rebrand the chip from freescale with the g5 name? it will be confusing for us but they could 😉

LOL maybe they will call it the G4 squared

 

[gerrycurl]

an x86 analogy

The pentium m processor came out after the pentium 4's, but actually used the pentium 3 processor core and added things like a faster bus, more cache, simd (Single-Instruction Stream Multiple-Data Stream, SSE1,SSE2,SSE3, MMX, MMX2, think altivec) instruction sets, and they lowered the voltage requirements and form factor to 130nm which dropped the power requirements greatly. This pentium m chip although based on the pentium 3, outperformed or rivaled the pentium 4 in many tests. it's actually a fabulous processor for notebooks because it's low on energy consumption and was almost as fast as the best desktops. Of course for things like multimedia such as divx encoding, pentium 4 HT with 800MHZ FSB's blew these out of the water. But whenever the pentium m hits higher clock speeds and higher bus speeds, it will once again rival the pentium 4's.

As to the dual cores, the proximity of the cores will obviously allow for faster context switches, shared L3 cache and overall just lower cost of production. And if freescale can get the form factor down to about 90nm, this could be a really compelling chip for apple. it will be freescale's pentium m. The g4 really isn't a bad chip, it's just that the architecture doesn't allow the clock cycles to be scaled up as high as say pentium 4's
or G5's. But there are other determining factors to lead to performance gains such as on board memory controllers, larger bus, more cache, etc.

The dual core g4 may not be a bad chip afterall, lower power consumption, upgraded speeds, but I'd most certainly wait for the dual core g5's for my next purchase. ~1-2 years.

 

[Maxx Power]

Why? The 7447A is a MUCH better processor. I have had G3 based machines and they were fine, but in NO WAY do they compare to the newest G4's. Everyone complains about the 167Mhz FSB speed...um the 750 ran at 100Mhz. It did not have Alti-Vec which makes a huge difference especially on the newest applications and OS from Apple. Last, it never came close to clocking as high as the G4 has gone, 1.5Ghz. Look at benchmarks anywhere (ie. iBook G3 vs. iBook G4) and you will see the current G4's kill the old IBM G3's. The G3 was a good processor, but WAS is the key word.

Woo, hold on a second. G4 runs on a 167 Mhz bus because that's what it's spec'ed to do. The G3, according to IBM's papers can run on any speed between 100 to 200 mhz. Apple decided to cripple the processor to not outshine the G4 in any way. By the way, the G3 has a more powerful FPU unit per mhz than the G4 any day, as well as lower power consumption. The G3 running at 1.2 Ghz with at 200 mhz bus will smoke any G4 crunching float/integer non-altivec code.

 

[Maxx Power]

I think you mean integer performance, not floating point. Floating point math is what AltiVec does.

No, he means Floating Point, as well as I originally started this comment in this thread, and I mean Floating Point as well. You missed a few posts in which we discussed the FPU issue in this thread.

To keep you updated i quote from earlier in this thread "that the fpu on the G4 is per unit weaker than that of the P4 from www.arstechnica.com the link is:
http://arstechnica.com/cpu/01q4/p4a...4andg4e2-3.html

and i quote "The P4's FPU hardware is a little beefier than that of the G4e, but a little less beefy than that of the PIII or the Athlon. The P4's designers slimmed down the FPU and weakened it a bit, banking on increasing clock rates and the widespread replacement of x87 code for SSE2-optimized code to take up the resulting performance slack. A closer look at the P4's FPU will show just how it has been weakened.""

 

[Maxx Power]

i read that article too. i just can't help but feel that on a clock-for-clock basis, the ppc 74xx's fpu is better than the p4's. perhaps it's the shorter pipeline that gives it that advantage.

One feels that way because the perceived Floating Point Arithmetic is good on the G4 since the only way we use that processor is with Apple's pre-compiled/optimized code. Apple optimizes very heavily in AltiVec since they spend a lot of marketing dollars on that technology. G4 is designed with the vector unit in mind since number one bulk customer of the processor is Cisco, who needs the vector units for lots of network processing that would benefit from this. Apple leverages this as well, but the traditional FPU is the way to go for most code, it is used regardless of VMX optimizations. I also feel the P4's design is inefficient like north american vehicles.

 

[MacQuest]

This might be kinda true, except for the part where microsoft gives any type of notice whatsoever to Apple. If they cared, they would have bought Apple long ago.

Never could/would happen. Shareholders would never allow it.

That's what makes Apple such a great company and a pain in the a$$ to monopolysoft. 😀 😉

 

[AidenShaw]

during OS and code testing we found the the opposite results...

Because of these results we are dropping OS X on our cluster and switching to LINUX

Were there any other variables? Same compiler used for all tests?

Did you separate 64-bit addressing from G5 scheduling optimizations (the former might hurt a bit, the latter might help a lot)?

Did you separate 64-bit addressing from 64-bit integer arithmetic?

While all your data seems reasonable, it seems that there's a good chance that the only variable isn't 32 vs. 64 bit virtual addressing.... But I'm glad to hear that you're ditching the proprietary OS - it will give you lots of options in the future.

 

[Sol]

In the context of dual-core G5s

It sounds like these dual-core G4s will be ready for next year's OS X 10.4 release. By that time we could even have dual-core G5s or G6s in the PowerMacs, if IBM can get on top of their manufacturing problems. Things get murky when it comes to the iMac and eMac; a G4 processor which can control Ethernet, USB, etc from the CPU would be ideal for the all-in one computers but Apple has allready gone on record about the G5 iMac. Does this mean that eMacs will also use G5s and that dual-core G4s will be exclusively in portables? Either way, it looks like a win-win situation for Mac users.

 

[paxtonandrew]

Is there any news weather the new chip will be able to fit inside existing computers? If possible, I will be ebaying my PB, and putting one into my PowerMac. I believe it has been said that the new Freescale chips to be pin for pin compatible with the current G4's, so maybe.

Has anybody heard anything?

 

[.a]

is it a bad idea to get a powerbook 17" now? when can we expect new powerbooks?
my powerbook is due to be shiped 9th of septembre (what i think is really late).
.a

 

[Windowlicker]

Apple works with Motorola, and IBM starts doing neat stuff whilst Motorola spins in the toilet. Apple jumps ship. IBM goes into a tailspin. Motorola starts doing neat stuff. 😱 Murphy's law. You switch lanes in traffic, the faster lane you switched to will grind to a halt and the lane you just got out of starts moving.

that's exactly what I thought when reading this.. I believe this could be a good thing. IBM's having problems with producing the current chips, and they surely will have problems producing G5's for laptops.

At the same time the iBooks start needing an update soon, but they really can't be updated to be as fast as the powerbooks. So if the powerbooks don't get a speed bump soon, both computers will be in a halt.

We don't wanna wait a year for new laptops -- we didn't wanna wait for rev B G5's for a year.

 

[JFreak]

You guys don't give the g4 enough credit. Its a great chip!!! all its missing is the massive bus speeds

you got that right 🙂 clock-for-clock the G4 outperforms the G5, so it all comes down to overall throughput of the system. in some cases a 1.5GHz powerbook G4 performs equally to a single G5 1.6 powermac, so the G4 is not so bad... it's just crippled with a slow bus.

 

[MacFan26]

I think apple could easily call this a g5 by name. G5 mobile, or "mini" or something. This wouldn't just be a minor improvement to the g4, they mine as well call it a g5.

Yeah, even if it isn't quite as good as the powermac G5 chips, they shouldn't call it the G4, so people can stop holding out and waiting for a G5 notebook. As for me, I don't really care if it's a G5 or not, if it's progress, it sounds good to me 🙂

 

[Trowaman]

is it a bad idea to get a powerbook 17" now? when can we expect new powerbooks?
my powerbook is due to be shiped 9th of septembre (what i think is really late).
.a

This is a check the buyers guide and make you own best guess.

If I were to GUESS, and I am GUESSING, with an announcement of dual core Jobs woudl save it for MWSF and not jsut middle of the Fall, if not then, it'd be at WWDC '05, assuming the G5 is still a no go.

*money still places G5 at WWDC '05.

 

[CmdrLaForge]

And no one has even considered that this, instead of a G5 processor, will be going into the next iMac? I mean, Apple didn't actually say that the iMac will be a G5. And if the components are to be mounted behind the screen (as per previous rumours), you'd think some sort of low heat solution would be needed.

Sorry, you are wrong about that. Apple said that the new iMac will have the G5.

 

[g3ski]

don't hold your breath

That says it all.

I am guessing they have more PR people working on Freescale than engineers. tehe.

 

[g3ski]

...........call it Cr@ppy Chip®

Yeah, even if it isn't quite as good as the powermac G5 chips, they shouldn't call it the G4, so people can stop holding out and waiting for a G5 notebook. As for me, I don't really care if it's a G5 or not, if it's progress, it sounds good to me 🙂

Yeah the G4.5 is what the world has been waiting for. Or, hey, how about a quadra and a half on your desktop ; )

Seriously though, you make a good point, as Intel has been dealing with this type of issue (namely naming and consumer perception of what's inside™) for a couple of years.

If it's a blazing fast dual-core at 2Ghz that fits inside a Powerbook without heat issues, then for all we care, call it Cr@ppy Chip®. 😛

 

[sebisworld]

Confused

So, help me to clear this up:

If a 1 THz Chip takes an hour to render a movie with a "really good" (tm) application, the dual 1 THz Chip would ideally only take 30 minutes. How long would a Dual Core take? Is it a second processor in the same enclosure or a trick to benefit more from what's alreay there?

 

[ac_1000]

It's possible but a recent discovery has indicated that the imac will have a G5. Why? Because there's a bit of code in 10.3.5 that works with the 970fx's PowerTune feature.

<dict>
<key>PowerBook7,1</key>
<string>SMU_Neo2_PlatformPlugin</string>

<key>PowerBook7,2</key>
<string>SMU_Neo2_PlatformPlugin</string>

<key>PowerMac8,1</key>
<string>SMU_Neo2_PlatformPlugin</string>

<key>PowerMac9,1</key>
<string>SMU_Neo2_PlatformPlugin</string>

The PowerMac8,1 refers to a consumer-grade G5 Power Mac, i.e. the new iMac. Using this same logic, the next PowerBooks will probably be G5 based.

I think these PlatformPlugins can only be seen as a hint to the chipset (or parts of it) on the mainboard. It's not a clear sign, that the new iMac and a new PB will have the same processor. One could also split the mainboard chipset in a "common" part and one part specific for a G5 or a new G4 CPU.

 

[Sol]

Dual core is faster compared to dual processor.

So, help me to clear this up:

If a 1 THz Chip takes an hour to render a movie with a "really good" (tm) application, the dual 1 THz Chip would ideally only take 30 minutes. How long would a Dual Core take? Is it a second processor in the same enclosure or a trick to benefit more from what's alreay there?

It is not as simple as that. A dual processor computer would be closer to 1.5 times faster than the single processor computer. Dual-core might be closer to 2 times faster than the single processor. That is, assuming that the application doing the rendering takes advantage of a second core or processor. Roxio Toast 5's VCD rendering plug-in is a good test for this. It is dual-processor aware so you can use it for benchmarks if you have the time and patience.

 

[macsrus]

Were there any other variables? Same compiler used for all tests?

Did you separate 64-bit addressing from G5 scheduling optimizations (the former might hurt a bit, the latter might help a lot)?

Did you separate 64-bit addressing from 64-bit integer arithmetic?

While all your data seems reasonable, it seems that there's a good chance that the only variable isn't 32 vs. 64 bit virtual addressing.... But I'm glad to hear that you're ditching the proprietary OS - it will give you lots of options in the future.

IBM XLC and XLF were used for all tests....
The linux used was SUSE Enterprise 9 RC5
The OS X was 10.3.5 Server
Our code is all fortran 90 ... utilizing double precision...
We didnt have any single threads that addressed more than 2 GB memory per processor

Buy the way... The reason OS X was out preformed is...
OS X has very poor/slow memory management.....
We had talks with Apple and they insured us that Tiger would fix the memory management bottleneck....
While we plan to dump OS X.... we will re-evaluate it when Tiger ships....