www.xlr8yourmac.com

Recent Updates  | Mac Upgrades/Repairs  | CPU Upgrades  | Storage  | Video  | Audio/HT  | Apps/OS/Network  | Recent
News, Tips, Reviews or Questions to News at Xlr8YourMac.com
G5 vs MDD G4 Performance Tests:Return to News Page

Click for ThunderBolt Upgrades!
Click for ThunderBolt Upgrades!

Comparison of G4 Dual 1.25 MDD vs G5 1.6 and Dual G5 2.0GHz
By Dave Linder
Posted: 5/14/2004 (part 2)
(For features comparisons, see part 1)

Benchmarks and Real World Tests
Doing the export test on the 6 sequential iMovie2 tutorial clips to the CD-ROM preset in iMovie 3.0.3 finished in 48 seconds on the G4/1.25 MP, 45 seconds on the G5 1.6, and 35 seconds on the G5 Dual 2.0. I am not sure what the iMovie export test is really stressing, but regardless even the slowest G5 beats the fastest currently available G4 system. (FYI - The iMovie "washing the dog" tutorial export to CDROM setting test is converting DV to Quicktime (H263) format, this is an easy to run test that others can reproduce for comparison. With iMovie4 however, the "Export" menu item is gone, replaced by the "Share" output option.) -Mike)

A common operation many do on their Mac's is exporting a DV clip to another format for the web. So I started with exactly 1 minute of DV footage captured in iMovie, and experimented exporting the 411.8MB file to Apple's MPEG 4 codec at the default LAN setting. ( It took less time to convert the 1 min DV movie to MPEG4 than to convert the 46.13 sec iMovie Tutorial DV to QT H263, however the two movies were not identical in content.)

Using iMovie 3.0.3 to do the export, the G4/1.25 MP took 45 seconds, the G5 1.6 took 36 seconds and the G5 Dual 2.0 took 27 seconds. Same test but now using QuickTime player v6.3, the G4/1.25 MP took 47 seconds, the G5 1.6 took 37 seconds and the G5 Dual 2.0 took 28 seconds. For some reason, the export was slightly faster in iMovie than using QuickTime Pro 6.3, but again the G5 1.6 beats the G4/1.25 MP in real world tests.

In all these consumer level QuickTime tests, it is too bad that even the fastest machines are marginally faster than the more common G4 class hardware. I had hoped that the G5 would be more of an advantage, but apparently even Panther and QT 6 are not yet optimized for the G5, nor appear to take advantage of dual processors. Perhaps Tiger will show some promise on this front.

(As a FYI on a larger conversion (1.9GB QT movie) to MPEG4 test, here's a copy of the results from my first Dual G5 vs G4 tests I posted here last fall.-Mike)

QT to MPEG4 Export Test Results

(Using the QT6 Pro version from last fall - the export showed no real benefit from dual CPUs. High end apps like Final Cut Pro do however.-Mike)

It does seem that instead Apple is placing more G5 optimization efforts on their Pro Apps instead, and from what I saw in newer builds of the Compressor App (QT to MPEG-2/4 conversion) just one revision of the software had a large impact in speed. As an example, I did an export of the same DV clip in Compressor v1.0.1 and then in v1.1 on a G5 1.6 machine, as well as v1.0.1 only on a G5 Dual 2.0 (v1.1 was not available at the time of my testing of the G5 Dual 2.0). Below is the result of these tests and shows the true level of improvement possible when the software is optimized for the G5 hardware:

As you can clearly see, the upgrade from compressor v1.0.1 to V1.1 gave a great speed boost to the G5 1.6 system, in fact a 36% improvement with only software optimizations. Without efficient code, even the fastest computers will get bogged down.

Checking the Apple website, there is now a listing of a v1.2 of Compressor, but it does not specifically mention any performance gains in that version, but they have added MPEG-1 exporting, which will be welcomed by many cross-platform users.

Another video related test is exporting the same DV clip using the BitVice 1.3b3 Demo. This application is apparently highly optimized to run well on the G4, so lets see how it fares on the G5 CPU's. Using the same 1 minute DV clip, and using the default settings of 2 pass VBR at 4.5Mb/s in BitVice but also selecting Interlaced, DV Color Correction and DV Luminance Correction:

The G4/1.25 MP finishes in 5m29s, the G5 1.6 in 6m23s, and the G5 Dual 2.0 in 3m27s. Here the G5 1.6 is around 20% slower than the G4/1.25 MP, likely due to heavily G4 optimized code in this application. The G5 Dual 2.0 however the speed champ by far. Expressed in realtime processing frames per second:

G4/1.25 MP5.58 fps
G5 1.64.53 fps
G5 Dual 2.08.95 fps
BitVice MPEG2 Export

Here the G5 1.6 is around 20% slower than the G4 Dual 1.25, likely due to heavily G4 optimized code in this application. The G5 Dual 2.0 however is the speed champ by far, the sheer performance of Dual CPUs and the higher clock speed make it the winner. So here the extra bandwidth and CPU power of the Dual G5 really shines, almost exactly 2x the speed of the single G5 1.6 system. I wonder how this application would perform if optimized as heavily for the G5 as it is for the G4 CPU.
Already BitVice is up to version 1.5.1, and claims better performance on G5 systems (however it is not optimized for the G5), but I did not have this version for my tests.

USB 2.0 Fast vs. FireWire 400
What about USB 2.0? Marketing people are good at telling everyone how USB 2.0 is faster than FireWire 400, but in the real world the overhead in processing the data in USB makes it fall far short. I tested a generic brand of external combo FireWire 400 and USB 2.0 case, which included an Oxford chipset. Inside was an empty 40 Gig Maxtor DiamondMax Plus 8 series 7200rpm drive with 2 MB cache IDE drive.

Using the front panel USB 2.0 port and FireWire ports on the G5 1.6, here are some finder tests:

 Write FolderRead Folder
USB 2.046 sec41 sec
FireWire 40028 sec21 sec
Time to copy 579.2 MB folder in the Finder

That is terrible, is USB 2.0 really that bad? (I saw the same thing in tests with a friend's Maxtor 200GB USB2.0/FW400 drive last month - the Mac's USB 2.0 performance is about 1/2 what Firewire400 ports are with the same drive. And not just on the G5 - same thing on a PB G4/1.33GHz AL also. But the (original) G5's onboard Firwire is slower than previous G4 towers and slower than the PB G4 AL models (appx 5MB/sec slower in tests with a FW400 drive, hopefully that will be improved in the next production G5 models.) Non-Mac website tests of FW400 vs USB 2.0 show FW400 appx 16% to 70% faster depending on number of files/folders in the R/W tests due to more overhead with USB 2.0. See articles/tests at cwol, techtv, PCMag and digit-life.-Mike)
Well I didn't have access to another unit to try out, but I did check in Apple System Profiler that the drive was recognized as USB 2.0 and High Speed, and I used a new un-opened USB 2.0 cable. Here are 2 screen shots showing the data reported by ASP in each mode:


USB 2.0 connection


FireWire 400 connection

It is interesting to note that ASP can report the speed of USB 2.0 devices, as not all are high speed. However only in FireWire mode does the hard drive model get recognized. Of course anyone with a recent Mac already has FireWire, so you shouldn't be using USB for connecting hard drives even with USB 2.0. USB may come in handy when you need to transfer files to your unfortunate PC using friends that only have USB, or to take advantage of new scanners and digital camera that use USB 2.0 connections.

As a final test of USB 2.0 and FireWire 400, I used QuickTime Player 6.3 to save the 75.5 MB G5 intro movie downloaded from Apple as a self contained movie to the same drive. As shown below, the FireWire 400 connection gets us a 60% speed boost over USB 2.0.

 Time
USB 2.025 sec
FireWire 40015 sec
Time to save self contained 75.5 MB QT movie to same drive

Using Xbench v1.1.1, disk test scores on the same external Maxtor drive connected via USB 2.0 and FireWire 400 are shown below:


Xbench 1.1.1 Disk Tests, USB 2.0 Fast vs. FW400

You can see from the Xbench scores that the sequential speed on the USB 2.0 connection hover around 50% of the FireWire 400 speeds. Below are Xbench scores for the internal Seagate S-ATA drive, filled to about 50% capacity as a comparison:


Xbench 1.1.1 Disk Tests, internal SATA drive

Java Linpack
I also performed a test of Java performance using 2 different browsers on each of the 3 systems. Using the Linpack Java benchmark and Apple v1.4.1 Java, the G5 1.6 more than doubled the Java performance of the G4 system on this test, 2.5x as fast to be exact. Later in the textbenchX benchmark, there is another Linpack test, not implemented in Java. There the gain from the G4 Dual 1.25 to the G5 1.6 is only 1.8x as fast, but both benchmark tests showed an equal gain of 1.3x from the G5 1.6 to the G5 Dual 2.0.

testbenchX
I also tested the 3 systems with a utility called testbenchX v2.1, which tests basic math functions to test Integer, Floating Point, Cache and SIMD processes. It was built with gcc 3.1 using a posix thread model. The results start with the tag "Ars Technica Benchmark [Ver: 2.0]" but I find no reference to it at the Ars Technica website. Although it is not recent (the base system is a G4/400 in OS 9), it did correctly identify both the G4 and G5 systems CPU type and other details:

G4 Dual 1.25

G5 1.6

G5 Dual 2.0

Model:

PowerMac3,6

PowerMac7,2

PowerMac7,2

RAM (MB):

1024

1280

1536

CPU Type:

PowerPC 7450

PowerPC 970

PowerPC 970

CPU Count:

2

1

2

CPU (MHz):

1249

1600

2000

FSB (MHz):

166

800

1000

L1 iCache (KB):

32

64

64

L1 dCache (KB):

32

32

32

L2 Cache (KB):

256

512

512

L3 Cache (KB):

2048

0

0

The results from testbenchX are separated into Integer, Floating Point, Cache and SIMD (AltiVec) tests, and I did find some of the results surprising compared to tests on the G4 1.25 MP system.

Looking first at the Integer results, even the G5 1.6 outperforms the G4 1.25 MP system in most tests, but there are some exceptions, notably the FAT Manipulation test where even the G5 Dual 2.0 was slower than the G4. In the Encode G.723-40 Audio and TSCP 1.71 Chess Benchmarks, only the G5 Dual 2.0 could beat the G4 system. In fact, other than the Towers of Hanoi, LZ77 Data Decompression and Dhrystone 2.1 tests that the G5 systems aced, the performance delta between G4 and G5 systems was not dramatic.

Moving to the Floating Point tests, here the G5 Dual 2.0 system was no less then 2x the performance of the G4 system, and up to 5x in the single precision NTSC YIQ -> PAL RGB test. Altogether an impressive showing by both G5 systems, but there is one twist in the Double Precision version of the NTSC to PAL conversion test. You will see from the graph that the G5 1.6 outperformed the G5 Dual 2.0 system in this test, how can that be? Is there some bottleneck in Dual G5 systems that clock speed alone cannot overcome or is it because the dual system was running multiple threads of the test (by default all systems were tested with 1 thread per CPU, the default in testbenchX).

As it turns out I ran the tests with testbenchX on the G5 Dual 2.0 system with 1, 2, 4 and 8 threads, and it turns out that with 1 thread only, the Double Precision NTSC to PAL conversion improved as shown below:

In the graph, you can see that with more than 1 thread the operation becomes slower and slower until you reach 8 threads when speed improves slightly. Of all the testbenchX tests, this was the only test that behaved in this manner.

This may be due to the fact that that each G5 CPU has 2 FPU units, vs. one per G4 CPU. But Apple developer documentation does say that code needs to be tuned to take advantage of the extra FPU in the G5, while the code in this test was created for the G4 CPU and would have no G5 optimizations.

The memory testing portion of testbenchX refers to cache memory, but the cache size that is being tested is not specified. I had to double check the results of the Sieve of Eratosthenes (Prime Generator) test, as it was exactly the same on all 3 machines. As for the LinPack (500x500) test, the G5 systems are clearly faster, but with the Fast Fourier Transform 64k test the G4 was obviously faster. This could be as simple as the code in this test falling into one of the performance traps of the G5. (FYI - last November I posted a link to http://www.fftw.org/speed, which shows the G5 (2.0GHz tested) having outstanding FFT performance (but noted one test anomaly) - unless the code used in the above test is using some instructions that have performance hits on the G5 CPU.-Mike)

In the SIMD (AltiVec) portion of the tests, the G5 systems run at about twice the speed of the G4 system, but on the "Complex Bitfield Operations" test the G5 1.6 edges out the G5 Dual 2.0 as shown below:

Haxial 1.0
The Haxial benchmark is patterned to use real functions as used in their software which include generating checksums and memory operations:

The Memory Fill and Memory Clear operations were slower on any G5 than the G4, while the Memory Copy and Memory Equal tests were more than 2x faster on the G5. Again like the cache memory tests in testbenchX, code that is not well adapted to the G5 can run slower than on a G4.

CacheMark
On to another usual benchmark, CacheMark 2.0.0b4 the latest version. This test really shows the improvements made in the new architecture of the G5. Even though the G5 systems have no L3 cache, they outperform even the 2MB of L3 cache (unlike L1/L2 onchip cache, the L3 cache is running at a fraction of the CPU speed) in the G4/1.25 Dual system tested. At the 1MB mark, the G4's L3 cache does narrow the difference in speed vs. the G5 systems.


Note that at 512KB, the G5 1.6 tested slightly faster than the G5 2.0 Dual system. The PowerPC 970 CPU has an onboard 512KB of cache per processor, so it is interesting that this occurs at only that memory size. (since the on-chip L2 cache is running full CPU speed, the 2GHz G5 should be faster) You can also see that even without a L3 cache, the G5 systems faster memory architecture

Overall the G5 Dual 2.0 averaged 3x the memory performance of the G4 Dual 1.25 system, a much needed and impressive improvement in one generation of hardware. As mentioned in Part 1 of this article, the G5 systems employ Processor and Bus Slewing to keep cool, quiet and lower power consumption. So I tested the G5 1.6 at all 3 levels of procesor performance with the CacheMark becnhmark to see it's influence:

While the Reduced setting does have an impact, the Auto and Highest settings are nearly identical in this test. If you are running Pro Apps, Apple does recommend using the Highest setting.

Which brings us near to the next generation G5 desktop systems that will come out in 2004, based on the PPC 970FX chip, as is used in the recently shipping Xserve G5 systems. I look forward to comparing these new models with the first generation G5 systems.

Conclusions
Interestingly, none of the testing showed any significant performance variations between Jaguar and Panther on the test systems. That said, I do think Panther has been more stable than Jaguar, so I am generally happy with it, along with the new features. As for the hardware, the G5 systems are far ahead, but many software applications are not yet taking full advantage of the improved hardware.

With the price of a comparably equipped G5 1.6 now equal to a G4 Dual 1.25, even the base G5 systems are getting more attractive. In the majority of testing, both G5 systems were faster, and looking to the future they will only improve further. This is because more and more applications are being optimized to run faster on the G5 CPU. Most of Apple's Pro Apps have already been updated with G5 optimized versions that make the most of the advances in the G5 systems. If you use Final Cut Pro or DVD Studio Pro, the answer is the G5 without question, ideally one of the G5 Dual models. A look at the next generation of Apple Pro Apps show some very high requirements for a "Recommended System", for example Motion lists a G5 Dual 2.0 with 2GB of RAM or more.

The tests also show that for the more consumer-oriented tasks like the iLife Apps, the benefit of the G5 systems are still there although not as large as one might expect. Low level benchmark tests show that there is indeed great potential in the G5 systems, once software is further optimized to exploit the new hardware. The initial intro of the G4 systems was similar, new faster systems with few applications that were optimized for the G4. Almost a year into the G5, things are looking up at least in professional applications.
The faster and larger memory of the G5 simply cannot be reached in a G4 system. The G5 also adds USB 2.0 to the tower Mac's but with rather poor performance versus FireWire 400 in my testing. (This is true of other current Macs like the PB G5 AL current models also, but the G4s have better Firewire performance than the first gen. G5 towers.) And one thing I miss in the G4 MDD models is the front USB/FireWire port of the G5, to plug in cell phones, digital cameras and other everyday items. (One thing I miss in the G5 is the dual CD bays of the MDD though and ability to have 4 hard drives inside the case without buying any 3rd party brackets and controllers.-Mike)

Today I would have to say that the current G5 Dual 1.8 would be the best value, and the G4 PowerMac towers are the last machines available for those that still would need to boot OS 9 in addition to OS X. I am hoping for great things in the second generation G5 systems, perhaps an extra removable drive bay and/or more room for more than 2 internal SATA drives without modification would be nice. Oh and of course I want it to be faster too!
-Dave


Other G5 related articles Here (from the Systems page G5 section.)



OWC XYM SPECIALS!
(Ad/Sale items)

= UPGRADES by Mac =
Upgrades just for
YOUR Mac!

= Refurb Mac Pros =
(Click for List)

SSDs from under $50!
Fast SSDs for Most Macs/PCs

= ThunderBolt =
Drives, Docks & More

= HARD DRIVES =
Up to 10TB HDDs
HGST, WD, Seagate, Toshiba

= 2.5in HDs & SSDs =
Notebook Hard Drives and DIY drive/case kit bundles

= MEMORY =
Lifetime warranty RAM Upgrades!

= OPTICAL DRIVES =
Internal and External Superdrives/Blu-Ray drives

= VIDEO/LCDs/TV =
Graphics cards, Displays, Adapters, Cables & more

= SOFTWARE =
Apps, Utilities, OS, VM, Games and more

= WIRELESS =
WiFi and Bluetooth Devices/Adapters/More

= Repair Service =
for iPhone, iPad, Macs

= iPad/iPhone/iPod =
Accessories, Cases, Repairs & More

NuGuard iPhone Case *Extreme* Drop Tested!


XLR8YourMac T-Shirts


 
= back to www.XLR8YOURMAC.com =


= Other Site Topic Areas =
Mac Mods/Upgrades | CPU Upgrades | Storage | Video | Audio/HT | OS Updates/Network | Recent


Copyright © 1997-2017. All Rights Reserved.
All brand or product names mentioned are properties of their respective companies.

Legal: Site Privacy and terms/conditions of use.