[supernovamac]

I just tested a powerbook G4 1.5GHZ and a Powermac G5 2x 1.8GHZ with xbench. And what did I find out: PB CPU test 175.87 and PM 183.10. And what does that mean? Well the dual 1.8GHZ is just a few points faster than a G4 1.5GHZ. Isn�t that disappointing?

Is G5 really faster��..

[Waragainstsleep]

That really is alot lower than I wold have expected.
Did you reboot them both before testing?
I would have thought the SATA drive would have made more of a difference that, let alone the extra CPU and 6x faster system bus.

[supernovamac]

IThis was just the result of the cpu test. But the total score was G5 217.66 and PB 124.91. So the faster bus and hard drive dues matter. But I was astounded at the CPU result.

[Lateralus]

Xbench really isn't the best thing to use for speed comparisons. It is more than a little quirky, and it accentuates a lot of things in the results that have very little to do with real world performance but affect the score in the test quite a bit.

That said, I do believe the performance 'increase' that the G5 offers over the G4 has been vastly over exaggerated. Both by Apple and by Mac users who would have praised any new processor that replaced the G4. In actuality, the G5 is about as fast as the G4 at the same clock speed.

[OB1]

Originally posted by PowerMacMan:
the G5 is about as fast as the G4 at the same clock speed.

Really?? How disappointing. Here's me thinking we'd made this huge leap. i was really looking forward to checking out a G5. I was expecting... dare I say it.... snappiness!

[supernovamac]

Well maybe the G5 would perform better if programs were optimised for G5. But before that happens there will be a G6 ore something.

[Catfish_Man]

I think it depends on what you're doing. If the task is memory limited, the G5 will destroy the G4. If it's floating point limited, it'll also destroy it. If it's branch heavy, or has lots of simple integer math, or fits in the L3 cache, or relies on altivec tuned for the G4 (a couple of the G4's altivec instructions do bad things on the G5), then the G5 doesn't do so well. Each processor has different strengths. The G4 is basically an altivec processor with an ok integer unit and a rather poor floating point unit. The G5 is designed for streaming lots of data from memory, through the floating point units, and back. It's integer and altivec units are decent, but not up to the level of its floating point. The G5 also hits higher clock speeds, so comparing clock for clock is kinda pointless.

[eddiecatflap]

anyone seen a 2.5ghz G4?


thought not

[supernovamac]

The G4 is dead. But that does not mean it is forgotten......

[eddiecatflap]

next feb , when tiger is released , we'll see the 64 bit-able G5 move far ahead

if you're thinking ahead , buy a G5

it's like the OS-9 / X conundrum all over again

[driven]

Originally posted by OB1:
Really?? How disappointing. Here's me thinking we'd made this huge leap. i was really looking forward to checking out a G5. I was expecting... dare I say it.... snappiness!

Jesus H. G. Christ on a popsickle stick!

It ain't gonna get more snappy. My Dual CPU G5 does things faster than I can think of them. It's certainly gobs faster than my Athlon XP 3200 with the same amount of memory.

I can't imagine a faster machine.

[Lateralus]

Originally posted by supernovamac:
The G4 is dead. But that does not mean it is forgotten......

Not quite dead. Certainly not forgotten.

Motorola is pushing ahead with G4 development. The 7447A supposedly has a little bit more growing to do. After that, the 7448 is going to take over. Supposedly it will take the G4 to 2GHz either late this year or next year.

The 7457 is no longer under development, and there is to be no 7458 to compliment the 7448. Motorola has determined that the addition of 256K more L2 cache to the 7447 and later G4s negates the performance increase brought by L3, making L3 cache, in their opinion, too expensive to continue relying on.

The L3 has been a big issue in the G4 scaling upward. It is the first thing to give out when overclocking, for example. So without it, we might see some real G4 progression.

And after the G4, Freescale (the name for Motorola's spun-off semiconductor unit), has two processors under development: The e600 and e700, both of which are G4 derived.

[Cadaver]

The G4 would have performed much better if it hadn't be trapped on such a crappy bus design. 167MHz in 2004 is ridiculous. A G4 on a fast bus would have evened the playing field between Mac and PC long ago.

While perhaps the G5 clock-for-clock isn't much faster than the G4, the bus speed on the DP2.0's is running 6 times faster than the fastest G4-based PowerMac bus. The huge boost in bus speed (including separate FSBs for each processor) is where the major speed of the PMG5 comes from.

[Loco Engr]

Originally posted by eddiecatflap:
anyone seen a 2.5ghz G4?

thought not

Outside Apple, anyone seen a 2.5ghz G5?

[supernovamac]

It is strange that apple is not making a quad G5 machine.

When I by my 2.5 GHZ G5 I am looking forward to testing it. And finding out how fast it really is. But in the end you are wondering what to use all that power for.

But one thing that the G4 is better at than G5 is heat. It does not get as hot as the G5�.

[OB1]

Originally posted by driven:
Jesus H. G. Christ on a popsickle stick!

It ain't gonna get more snappy. My Dual CPU G5 does things faster than I can think of them. It's certainly gobs faster than my Athlon XP 3200 with the same amount of memory.

I can't imagine a faster machine.

That's just what I wanted to hear. My main machine is a 867MHz G4 12" Powerbook. Audio is it's main use, so it has it's limitations. I've been eyeing up the G5 PowerMacs, but some of the posts here had me thinking it might not be worth bothering...

[Nerozwei]

Originally posted by supernovamac:
The G4 is dead. But that does not mean it is forgotten......

Arf.. Damn you Power Mac people, I just got my first G4 a month ago (iBook G4)

[supernovamac]

Originally posted by Nerozwei:
Arf.. Damn you Power Mac people, I just got my first G4 a month ago (iBook G4)

I would have bought a powerbook G4 if it weren�t for the fact that I am tired of laptops. I want a stationary computer with lots of power. I hate waiting for things to happen. And I used to have a powermac but I switched It for a Ibook 500. And I did not now then what I know now. That a small monitor sucks�..

The powerbook Is fast, but the powermac is faster�.

[flanders]

i think it has more to do with what you're doing software wise on the computer. the G4 is a great processor and just between CPUs I think the G4 and G5 are pretty close but that's where the similarities end. If you use the bus or memory at all to move large chunks of data, the G5 will crush the G4 otherwise they will appear very close.

[Catfish_Man]

Originally posted by flanders:
i think it has more to do with what you're doing software wise on the computer. the G4 is a great processor and just between CPUs I think the G4 and G5 are pretty close but that's where the similarities end. If you use the bus or memory at all to move large chunks of data, the G5 will crush the G4 otherwise they will appear very close.

Heh... try some non-vector floating point code. If you want to even it out, you can even make sure it fits in the cache so that the bus isn't used at all. The G5 will win by a lot more than you seem to think...

G4: 1 floating point unit, basically no out of order execution, 1.5GHz max
G5: 2 floating point units, extensive out of order execution, 2.5GHz max

[Peder Rice]

I really hate it when it is said that clock-for-clock comparisons do not matter, as long as the processor is ultimately faster.

However, I believe we've all noticed AMD's extensive use of internal improvements to take their 64-bit chips to the next level, thus negating the need for a high clock speed. After all, if more work can be done at a lower clock level, then less heat is produced and less energy is consumed. This not only makes systems run quieter, but it has the ability to extend battery life in a laptop scenario. Therefore, if the G4 is faster clock-for-clock, then why don't we continue to employ it in the PowerBook and iBook lines? If we can raise the bus speed on the G4, then the PowerBook line could look very tempting.

I haven't had a chance to read anything on the new Freescale chips. What's the word on the technical side of things? Any vast improvements, or are the G4 and its derivatives doomed to land themselves in the embedded market forever?

[Commodus]

The G5 is considerably faster than the G4, but it's not a knockout punch - it's mainly in certain areas that it pulls ahead.

The biggest deal to me is floating-point math (as Jobs was keen to point out at WWDC 2003). The G4 was quite weak in that regard, and that's why G4s didn't have much credibility as gaming systems. The G5 absolutely stomps all over the G4 (and the Pentium 4, for that matter) in floating-point. You can buy a PowerMac and actually expect to keep up with current AMD/Intel-based computers in general number crunching.

There's also the overall system architecture. We know there's bandwidth, but the whole design is geared to avoid conflicts. There are two PCI-X buses, for goodness' sake (one for the 133 MHz slot and one that handles the two 100 MHz slots). The performance hit taken by adding a data-heavy peripheral, such as a gigabit Ethernet card, is substantially smaller than it was.

[chris v]

Originally posted by OB1:
That's just what I wanted to hear. My main machine is a 867MHz G4 12" Powerbook. Audio is it's main use, so it has it's limitations. I've been eyeing up the G5 PowerMacs, but some of the posts here had me thinking it might not be worth bothering...

Go check around the Logic forums. The G5's are an order of magnitude better for running Logic, anyway. Folks have reported 50-60 audio tracks and 30-40 reverbs. They're supposedly just monster for Audio, especially compared to a single proc. G4. You'll see about a 10x improvement in # of tracks and plug-ins.(Depending on your app, of course--Logic is dual-processor-aware.)

CV

[wulf]

Similarly with Final Cut Pro: the difference between the 2GHz dual G5 and a 867MHz single G4 is not four or five times the speed. The G5 is around 10 times faster for rendering. At least.

That is based on real-world test renders of work projects, rather than any benchmarks.

[Waragainstsleep]

So the G5 is faster in some areas than the G4? And the G4 was faster in some areas than the G3.

So the G5 is a 64-bit, Altivec-equipped G3 with a better FPU (or does it have two of them? Whatever)

[OB1]

Originally posted by chris v:
Go check around the Logic forums. The G5's are an order of magnitude better for running Logic, anyway. Folks have reported 50-60 audio tracks and 30-40 reverbs. They're supposedly just monster for Audio, especially compared to a single proc. G4. You'll see about a 10x improvement in # of tracks and plug-ins.(Depending on your app, of course--Logic is dual-processor-aware.)

CV

........ Wow. That is truly more than I'll ever need. Can I really have my dream computer - one that I don't actually need to upgrade?

[Catfish_Man]

Originally posted by Waragainstsleep:
So the G5 is faster in some areas than the G4? And the G4 was faster in some areas than the G3.

So the G5 is a 64-bit, Altivec-equipped G3 with a better FPU (or does it have two of them? Whatever)

Uhhhh.... The G4 is an altivec equipped G3 with a new floating point unit, a new bus protocol, and a new cache interface (along with quite a bit of other stuff in the later revisions). The G5 is totally unrelated to either of them (except in the altivec unit, which is very similar to the original G4). The G5 is a cut down POWER4 with early G4-style altivec.

[P]

Originally posted by Waragainstsleep:
So the G5 is faster in some areas than the G4? And the G4 was faster in some areas than the G3.

So the G5 is a 64-bit, Altivec-equipped G3 with a better FPU (or does it have two of them? Whatever)

Nope. There has been rumors of such a chip, but the G5 is not it. The G5 is a Power4 with one core (the Power4 has two), smaller cache and a bolt-on Altivec - bolt-on in the same way that MMX and SSE was added to Pentium Pro to make it Pentium II or III, opposite being designed in like Altivec in the 7450 or SSE2 in Pentium4. Note that the Altivec in the G5 is more of hack than the others, as it was added using automated tools rather than with a hand-designed new stepping. GPUs like Radeon or Geforce are often designed with automated tools for speed in development, but CPUs are usually hand-made for better performance.

[Waragainstsleep]

Its amazing how much you can learn from people by posting half-arsed, half-asleep comments.

Many thanks.

[Randman]

I would think how much ram is in each is also going to impact scores, as well as what tasks were the ones being studied.

[GORDYmac]

In the future, I'd buy a Dual 2GHz G4 upgrade for my Quicksilver before I'd consider a G5. Right now it works fine.

[chatam]

I quite agree with your attitude.

The G 5 is a big improvement in theory, but the version available for the time being does not seem to justify the switch from a robust dual G 4 configuration.

I would gladly accept points leading to the contrary, but this is it, unless your needs are indeed paticular or fashion is your guide.

[supernovamac]

If you already have a G4 powermac then there is no need to upgrade yet. But if you are bying now I am going with a powermac G5, and not a imac or an powerbook.

But I think that the megahertz myth is one day going to die. And se more efficient chip designs. Maybe more energy efficient. And I have heard that IBM is working at the heat and energy usage of the G5. But it maybe some time before we see something.

[hmurchison2001]

Originally posted by GORDYmac:
In the future, I'd buy a Dual 2GHz G4 upgrade for my Quicksilver before I'd consider a G5. Right now it works fine.

Hell I would too! I think people screamin' about slow FSB don't really understand why the G4 can survive on 167mhz FSB.

Anyone feel free to correct me if I'm wrong but CPUs like Intel and AMD need faster FSB because they have over twice the pipelines of the G4's 7. This means if the processor has to discard data it has to wait that much longer before filling the pipeline again. The G4 is penalized for having bubbles and stalls in the pipeline but it's not as severe because it's only contending with 7 total pipes and the original G4 had only 4.

CPU design is far more technical than my knowledge but I do know that heavily pipelined CPU require fast access to memory. Thus we have the higher pipelined higher clocked G5 which is faster than the G4 but on a clock for clock basis the increase is very small. That's the tradeoffs you make when you need to increase clockspeed. There is no magical elixir here.

[Catfish_Man]

Originally posted by hmurchison2001:
Hell I would too! I think people screamin' about slow FSB don't really understand why the G4 can survive on 167mhz FSB.

Anyone feel free to correct me if I'm wrong but CPUs like Intel and AMD need faster FSB because they have over twice the pipelines of the G4's 7. This means if the processor has to discard data it has to wait that much longer before filling the pipeline again. The G4 is penalized for having bubbles and stalls in the pipeline but it's not as severe because it's only contending with 7 total pipes and the original G4 had only 4.

CPU design is far more technical than my knowledge but I do know that heavily pipelined CPU require fast access to memory. Thus we have the higher pipelined higher clocked G5 which is faster than the G4 but on a clock for clock basis the increase is very small. That's the tradeoffs you make when you need to increase clockspeed. There is no magical elixir here.

This is somewhat true, but it's mostly true for integer code. Floating point code tends to look something like this:

stream data into processor -> do something to it -> stream data out of processor

Big caches and short pipelines are useless for this. The fast bus long pipeline model of the P4 actually kinda makes sense for this kind of code, aside from the heat/power problems. OTOH, integer code typically looks like this:

fetch data into cache -> work on data for a while, branching and reusing data a lot -> write some results back to memory

Here the bus matters much less, and the pipeline length and cache size/latency matters more. The first because of branches, and the second because of data reuse.

AFAIK the main reasons the G4 doesn't totally get its ass handed to it are the huge cache, short pipelines, and great vector unit. The first two help with integer code, which is probably more common (I'm not sure about that though), and the last allows it to somewhat save its dismal floating point performance by vectorizing particularly critical bits (it's still bus limited, but at least it's quick once it gets on the chip).

Corrections welcome, of course.

[P]

Originally posted by hmurchison2001:
Hell I would too! I think people screamin' about slow FSB don't really understand why the G4 can survive on 167mhz FSB.

Anyone feel free to correct me if I'm wrong but CPUs like Intel and AMD need faster FSB because they have over twice the pipelines of the G4's 7. This means if the processor has to discard data it has to wait that much longer before filling the pipeline again. The G4 is penalized for having bubbles and stalls in the pipeline but it's not as severe because it's only contending with 7 total pipes and the original G4 had only 4.

CPU design is far more technical than my knowledge but I do know that heavily pipelined CPU require fast access to memory. Thus we have the higher pipelined higher clocked G5 which is faster than the G4 but on a clock for clock basis the increase is very small. That's the tradeoffs you make when you need to increase clockspeed. There is no magical elixir here.

It depends on the task. For a simple integer task, you're mostly right. Add the bigger caches, and the bad FSB is almost totally hidden. For something like streaming data from RAM to the CPU and back when encoding video, there is no substitue for a fast bus. The point is, any modern CPU is "fast enough" for simple integer tasks.

[starman]

For work I had to build a library for our software. The problem was that it wasn't that much faster on the G5 than the G4. Doing a little research, I found that I could use a compiler directive to optimize the code for the G5. Doing that, it was SIGNIFICANTLY faster on the G5 - one test took 4 minutes with the old lib, 30 seconds with the new one (literally).

FYI.

Mike

[olePigeon]

Clock for clock, the G5 and G4 are about the same. Where the G5 really shines is 64-bit memory addressing. The G5 will be significatnly faster than a G4 in high bandwith memory applications.

[york28]

Won't the much faster bus, memory, etc. help the G5 in the long run?

[supernovamac]

Originally posted by york28:
Won't the much faster bus, memory, etc. help the G5 in the long run?

I wonder why apple is not using faster ram. There must be faster ram out there.....

[Mac Hammer Fan]

Originally posted by supernovamac:
I just tested a powerbook G4 1.5GHZ and a Powermac G5 2x 1.8GHZ with xbench. And what did I find out: PB CPU test 175.87 and PM 183.10. And what does that mean? Well the dual 1.8GHZ is just a few points faster than a G4 1.5GHZ. Isn�t that disappointing?

Is G5 really faster��..

Bench tests don't tell everything.
According to XBench the NVIDIA Geforce Ultra 5200 64 MB card should be faster than the ATI Radeon 9600XT 128 MB but every gamer will confirm that the latter is a better card.
It's better to take real world tests.
And in this case, even a G5 Dual 1.8 will be faster than any G4.
Try mp4 conversion and you will know enough.
Still not convinced? Check out these links.

http://www.xlr8yourmac.com/G5/G5_vsG....html#storytop

http://macspeedzone.com/html/hardwar...ml#Application

[Catfish_Man]

Originally posted by supernovamac:
I wonder why apple is not using faster ram. There must be faster ram out there.....

Not officially. The fastest ram that's a JEDEC standard is PC3200. After that it's all DDRII, which is expensive and a bit slow right now.

[Groovy]

Originally posted by PowerMacMan:
That said, I do believe the performance 'increase' that the G5 offers over the G4 has been vastly over exaggerated. Both by Apple and by Mac users who would have praised any new processor that replaced the G4. In actuality, the G5 is about as fast as the G4 at the same clock speed.

now you are over exaggerating

I'm not going to get into the details but i will say if you were are the WWDC
then you would have seen first hand that the compilers are not up to snuff for G5's.
They will be in about 6+ months. Expect a good 20% to 30% increase in speed just
from a recompile and that is just getting to where the base speed should be. Doing
opts for just G5 will get a lot more of course.

Also note G4 code actually takes a speed hit when running on a G5 if at least a basic
recompile is not done and most OS and third party apps have not done so. A good
example is RC5-72. It would see a 30%+ speed increase just getting rid of all the
stalls and a few other minor tweaks the G5 chokes on because it is G4 altivec opt
and not G5 opt. Tiger will be the first release for the G5 that is equivalent to panther
on a G4. Third party apps are stuck having to wait too because the compilers are not
there yet as well.

it always takes a few years for the software to catch up to the hardware in this regard.
BTW - There is a reason Xcode 2.0 and gcc 3.5 are such big news when talking tiger.

[mikellanes]

We have a Dual 1.4 G4 and a Dual 1.8 G5

The G5 easily beats the G4 in rendering, we do a broadcast TV show and using the exact same project files run off the same x-raid (not at the same time)

Test is this: SD footage around 10 effects rendered out

G4: 8:04 minutes
G5: 4:18 minutes

So the G5 was almost 4 full minutes faster, or twice as fast.

[Catfish_Man]

Originally posted by mikellanes:
We have a Dual 1.4 G4 and a Dual 1.8 G5

The G5 easily beats the G4 in rendering, we do a broadcast TV show and using the exact same project files run off the same x-raid (not at the same time)

Test is this: SD footage around 10 effects rendered out

G4: 8:04 minutes
G5: 4:18 minutes

So the G5 was almost 4 full minutes faster, or twice as fast.

By my calculations (which assume that the G4 scales linearly with clock frequency, so they're biased towards the G4), that makes the G5 a little over 30% faster per clock at that task. Not too shabby, especially given that it clocks higher.

[Back-to-Mac]

This is something I have been pondering over for a couple of weeks. I'm having a terrible time deciding between a G4 or a G5 tower.
If I choose a dual 1.25G4 tower I can save a fair bit and put the money towards a nice LCD Apple display but could there could be a hidden hit later as the G4 will not last as long as the G5 tower. Plus the last of the MDD towers were supposed to be still pretty loud and it only has a 4xAGP

[Lateralus]

Originally posted by Back-to-Mac:
...Plus the last of the MDD towers were supposed to be still pretty loud and it only has a 4xAGP

By most accounts, the current MDDs are significantly quieter than the first revision of wind tunnels.

I had a Dual 867, first revision, and I found it to be rather quiet, quieter than my current QuickSilver.

As far as AGP: There is little to no speed difference between AGP 4x and AGP 8x. And AGP, 2x 4x and 8x, will die at the same time: when PCIe video cards become the standard.

So, I wouldn't worry about that.

[SafariX]

Originally posted by Back-to-Mac:
This is something I have been pondering over for a couple of weeks. I'm having a terrible time deciding between a G4 or a G5 tower.
If I choose a dual 1.25G4 tower I can save a fair bit and put the money towards a nice LCD Apple display but could there could be a hidden hit later as the G4 will not last as long as the G5 tower. Plus the last of the MDD towers were supposed to be still pretty loud and it only has a 4xAGP

Push for the G5.

[another_steve]

Let's compare the 1.42GHz dual G4 to the 2.0 GHz dual G5. By clock speed alone, there's a 40% difference. Performance does scale linearly with clock speed in most cases. That is, if there were a 2 Ghz G4, the best you could hope for would be a 33% gain over the 1.42 Ghz G4. Now, that said, let's look at some actual benchmarks of real applications, not meaningless synthetic benchmarks.

Barefeats benchmarks

The worse case scenario was PS 7 which only showed 38% and 25% improvements for the non-MP and MP aware tests respectively.

However, there were tests like Cinebench which showed a whopping 110% improvement. That's all CPU in that test.

Most other tests fell somewhere in between. So, the bottom line is that in the absolute worse case scenario, the G5 scales similarly to the G4 if the G4 were actually capable of scaling to the same clock speed, which it isn't due to architecture limitations. In most other cases it's MUCH faster than the equivalent G4 would be at the same clock speed.

Steve

[haunebu]

This discussion is like the "short bus" edition of an Ars Technica thread.