iMac G5 / PowerMac G5 performance

I'm pretty sure the motherboards used in the low end models do not use PCI-X, they also only support a mere 4GB of RAM instead of 8GB (though some claim that the board should support a full 16GB of RAM using the 2GB modules, its just amazingly expensive to configure that!).

See, for my purposes I'm thinking of the exact opposite. I can take the internal 160GB drive in the iMac G5 and make 4x40 GB partitions, running 10.3 Client, 10.3 Server, 10.4 Client, 10.4 Server and be able to use whatever I need, wherever I need it ASAP. I'm thinking this machine would be a great server for me, especially considering the machine I used before rated about 70-80 on xbench.com, so 155 is a huge upgrade in that sense...

I'm still holding my developer discount for the next PowerMac revision, so I can get a full graphics workstation...I just hope its not all the way at WWDC again. I think it'd be a mistake to make only yearly industrial grades updates.
 
After looking through the hardware page i am sure that the dual 1.8 PM does not have PCI-X. I was just wondering because of that customize page had the PCI-X NIC as an option.
 
When comparing the new iMac G5 to the PowerMac G5 1.6, keep in mind that the PowerMac was using DDR333 single channel compared with the iMac's DDR 400 single channel, with the latter's FSB more than adequate to handle the throughput. Considering the "U3 lite" does not need to handle cache coherency and there is no need for dual pipeline switching, the iMac's memory latency should also be faster. Again, this is in comparison to the single processor PowerMac G5.
 
Ok, I ran XBench on a new (original) Dual 1.8GHz G5 with 250GB drive and 1GB of RAM. Video is standard NVIDIA. Got a 211.43 score. So, that's pretty much double what my 1.25GHz PowerBook scored. I'll bench my iMac when it arrives, sometime!
 
The performance doesn't look too bad, actually. I would have liked it more if they had compared the 12" Powerbook as well since that is quite close in price to the iMac.
 
Looking at the PB's scores, I guess a PB G5 wouldn't currently even do us that much good, AND kill our battery lives... Kinda glad there'll be another G4 PowerBook. These G5 processors just don't seem 'ready' yet - unless you also have the surrounding technology to back it up. That's quite clearly what sets the 2.0MP in those tests apart - and we all know that the space restrictions of a PowerBook do not allow for, say, graphics cards with their own fans that are three times as thick as a PowerBook. ;)
 
Fryke -

Yes and no. Motion, as I understand it, and please correct me if I'm wrong, is built to be very intensive on graphics card performance. We see the cream of the crop of cards rising to the top in these tests easily. After that, in gaming, we see that the iMac G5 and the PowerBook G4 are about the same.

When we get to Cinebench, which always makes me want to go to the mall to Cinabun or whatever its called, we see about a large (90-ish%) increase in performance from iMac G5 1.8 to 1.5 PBG4. We see iMovie score about 20% faster and so on.

I'd be also interested in seeing a PBG4 benchmark with the 128MB card installed, since a PBG5 would probably have a quick upgrade available to 128MB as it does now. Also, it'd be interesting to see a PowerMac G5 1.6GHz outfitted with 128MB of ram perform, as that would be close to a 1.6-1.8GHz iMac G5, but with the ability to see how video performance works with the G5 architecture even further.
 
The gaming benchmarks actually seem to favour the Powerbooks. This is because the Powerbooks are equiped with the Radeon 9600, which is a much better video card compared to the iMac G5's Geforce fx5200 Ultra.

Looking at the rest of the performance benchmarks, it seems that the iMac has about 20% better performance(iMovie and Filemaker) compared to the 1.5 GHz Powerbook. This can be put down to the clock speed advantage since the 1.8GHz iMac is clocked 20% more than the powerbook. The iMac handily beats the Powerbooks at Cinebench largely because of it's much stronger FPU. This is speculation on my part, btw.

It performs very poorly in Motion, even when rendering is off. That's really dismal performance.

After examining the results more closely, I have to retract my previous statement. The performance of the G5 iMacs are quite lack luster in my eyes. A Powerbook G5 is hopefully a long way off.
 
I agree and disagree. With Tiger, the code is being even more geared towards the strengths of the G5. If we're looking at Tiger shipping by mid 2005 and the PowerBook G5 at WWDC, for arguments sake, we should see those numbers change greatly. I'd expect the iMac G5's performance to be increased as well, if Apple gears Core Video towards G5 and has Motion take advantage of it, which I'm sure they would.

To say that the performance of the G5 iMac is lackluster may be slightly shortsighted, no offense please. We all expect G5 to be faster and more capable and in some tests, it was, but we also have to consider that a 1.8GHz iMac G5 is about $1500 compared to a $2500 laptop. Its good to see that the portable line is still packing a punch, but I would think that Tiger, especially with Apple's Pro-Apps will show improvements. I haven't used Tiger on G5 architecture yet, but that's not a really fair evaluation either being that Tiger is still very far from complete. On the other hand, on my Athlon64 system, moving from XP Pro to XP Pro 64 Beta showed large speed increases in the OS. It wasn't a slow machine to start with, but you could see a difference.

I think a lot of what we're seeing is the graphics card, which is why I'd be interested in seeing a 1.6GHz PowerMac with 128MB graphics installed. If they were to have a 1.8GHz PowerBook G5 today with 128MB pro graphics card installed, we may look at this differently.
 
hulkaros said:
iMac G5 vs the rest of the "family":
http://www.barefeats.com/imacg5.html

Nice? :)
Click to expand...

I don't get it. Dual G4 1.42 stomps all but the 2.0Ghz G5. Is the 2.0 much better than the 1.8, other than the PCI-X and the extra memory slots what is different?

Edit: Never mind, the ones the the 1.8 G5 does bad in I am assuming are dependant on video card. Which is why I am planning on getting a 9600XT in my G5. 50 bucks is so worth it.
 
Go3iverson said:
I haven't used Tiger on G5 architecture yet, but that's not a really fair evaluation either being that Tiger is still very far from complete. On the other hand, on my Athlon64 system, moving from XP Pro to XP Pro 64 Beta showed large speed increases in the OS. It wasn't a slow machine to start with, but you could see a difference.
Click to expand...

That's a very different move :). The x86 architecture has always been hampered by a design that dates back to the 70s. One of the biggest drawbacks of that design is the lack of registers[1]. Think of registers as a high speed memory on the CPU for it to store data that it is currently working on. Data and instructions get fetched from RAM and is placed in registers for the CPU to work on[2].

The x86 has 8 general purpose registers (GPR). While this was sufficient in the 70s, it isn't sufficient today. This has been one of the criticisms of x86 and is addressed to a certain extent by x86-64(or AMD64 or EMT64 which is basically the same thing). The x86-64 spec extends the number of GPRs to 16, basically doubling it. It is estimated that a simple recompilation of standard x86 code to x86-64 code results in an approximate 20 - 30% speed improvement[3]. Due to the increase in GPRs which is expected on register starved processor architectures. The move to 64 bits has little to no impact on the performance of a processor. In some cases, you actually see a drop in performance[4].

This is not the case with the PPC architecture. Right from the very start, the PowerPC architecture was designed with 32 GPRs. The main advantage of moving to PPC64 is the increased amount of addressable RAM which is a good thing if you're working with very large documents like video files, etc.

It's a sunday afternoon where I'm at, and I'm a little too lazy to dig up references to what I've said. A quick search google on the terms that I've posted should turn up the references needed :)

references:
[1] The AMD64 ISA value proposition, page 9 http://www.amd.com/us-en/assets/content_type/DownloadableAssets/dwamd_Value_of_AMD64_White_Paper.pdf
[2]FOLDOC, http://wombat.doc.ic.ac.uk/foldoc/foldoc.cgi?query=register&action=Search
[3]Anandtech, Linux desktop CPU rounduphttp://www.anandtech.com/linux/showdoc.aspx?i=2213&p=2
[4]OSNews, Are 64bit binaries slower than 32 bit binaries, http://www.osnews.com/story.php?news_id=5768&page=2
 
Viro,

Now *that's* a reply! I knew drips and drabs of that info, but that was a very well thought out, comprehensive comparison and is greatly appreciated! Especially conisdering its a Sunday! :)

I have heard questions of if 64 is slower than 32 and I've heard some compelling arguments. My argument back to them was, for my case, being an argument over the huge increase Apple users gained in the G5, was "I don't care if its technically slower or not, in the realm of OS X, plunk down in front of a dual 1.42GHz G4 and a dual 2.0GHz (high end at the time) G5 and tell me you don't feel a performance gain!" ;)

Scientific, no, but true! But I do understand the arguments on either end, for the actual 64-bit part of this argument, not the ghastly leap forward to having dedicated buses that run at 1GHz instead of a shared bus of 167MHz and such. ;)
 
I think preliminary tests of 64-bit speed vs. 32-bit speed would be incomplete (or possibly inaccurate), since we really don't have a true 64-bit OS to test it with.

I do agree with Viro that the more GPRs you have, the bigger speed increase you'll see. As an Assembly programmer, where you work directly with the registers available, I can say from experience that having more registers to work with produces shorter and more compact (and therefore, faster-running) code. The number of GPRs available on the processor is probably the main influence behind the old "PPC is faster than x86 at the same clock speed" debate.
 
The G5 produces much better performance than the G4 on certain applications because it has a much more powerful FPU. Just look at some of the XBench CPU scores and you'll see that the floating-point performance is spectacular. I even venture to say that the FPU of the G5 is better than that found on the Athlon, which is no small compliment! Other enhancements to the G5 include like you've said, a much faster system bus.

However, all these things are there even today in the absence of a true 64 bit operating system and users already reap the benefits. So the question is will moving to 64 bits improve anything? Unless you need to address more memory, the answer is a resounding no.

Briefly, a lot of this has to do with fetching data from RAM. Moving from 32 bits to 64 bits effectively doubles the cache lines. This is a bad thing. Data in the CPU cache is stored in lines of a fixed size. For example, if your CPU had a cache line of 64 bytes, if you needed just 1 byte from RAM, 64 bytes would be read. Going to 64 bits on this theoretical CPU would normally mean that the cache lines double in size. So if you needed 1 bytes from RAM, you'd end up reading 128 bytes from RAM. You can imagine that your cache is now less efficiently utilized.

On a side note, this effect of can be reduced by some clever programming, sometimes called locatlity optimization. This basically ensures that the data that is normally operated on together is stored close together in RAM so that when data is fetched from RAM to cache, the chance of wasting cache becomes less. Even with perfect locality optimization, all that happens is that the effect of doubling the cache lines gets negated, and you're pretty much back to square one (i.e. 32 bit processor).

A 64 bit processor isn't strictly going to improve performance. It is however a nice marketing number :). This is not to say that Tiger won't sport any performance enhancements. Look at the move from Jaguar to Panther. I noticed the performance gain. Same for the Puma to Jaguar move. The only question that remains, is how much of the performance gain in Tiger was due to the move to 64 bits? The answer is most probably none.
 
@Eldiabloconcaca

PPC is faster than x86 at the same clockspeed. It just depends on what apps you run ;)
 
Yep, and that's something I've thoroughly "preached" to those that will listen... :D

...but the question is, "Why?" Is it CISC vs. RISC or some variant thereof? Is it because we have more registers to work with, and, if so, is that still true of the G5 and the AMD/Intel processors?
 
Go3iverson said:
"I don't care if its technically slower or not, in the realm of OS X, plunk down in front of a dual 1.42GHz G4 and a dual 2.0GHz (high end at the time) G5 and tell me you don't feel a performance gain!" ;)

Scientific, no, but true!
Click to expand...
Actually it's very much scientific. Those 2 machines you mention are both running in 32 bit mode. The G5 system is ~600MHz faster (per CPU) than the G4 system, and has a faster system bus. So it's a no brainer that the G5 is better in your comparison, you could have just stopped at "it has more MHz" and call it a day.

ElDiabloConCaca said:
...but the question is, "Why?" Is it CISC vs. RISC or some variant thereof? Is it because we have more registers to work with, and, if so, is that still true of the G5 and the AMD/Intel processors?
Click to expand...
Having more registers is one thing, but being able to move data quickly is another. Modern AMD/Intel CPUs compensate for lack of registers by having very fast memory buses. The G5 comes along and gives us best of both worlds.

As great as the PPC FPU is, their ALU is very lousy. If you do any MacOS programming, especially in OSX with Cocoa, why do you think Apple chose to use floating-point data types in their API even for whole numbers?
 
You must log in or register to reply here.
Back
Top