Hands on with the Mac Pro: Testing the limits Page 2 of 3

Frame rates and transfer speeds are put to the test

Testing the Mac Pro

You can see the results of Macworld Lab’s Mac Pro tests in our full review.

For this report, I didn’t duplicate those tests, but instead, chose to update the tests I ran for my report on the Intel-powered Mac mini.

GeekBench benchmark: In the period of time since I used GeekBench for the mini review, it’s been updated a few times. So I re-ran the benchmark on all my machines: the PowerBook G4 (1.25GHz, 768MB RAM), the Intel mini (1.67GHz Core Duo, 2GB RAM), the MacBook (2.0GHz Core Duo, 2GB RAM), the Dual G5 (2.0GHz G5, 2.5GB RAM), and the new Mac Pro (Dual 2.66GHz Xeon, 2GB RAM).

After all the crunching was done, here’s how everything came out:

ROB’S GEEKBENCH TESTS

	Threads	PowerBook	Dual G5	Intel mini Universal	MacBook Universal	Mac Pro Rosetta	Mac Pro Universal	Rosetta as % of Universal
Emulate 6502	1	65	120	76	92	153	162	94%
Emulate 6502	Multi	65	240	153	183	607	645	94%
Blowfish	1	119	128	136	163	127	233	55%
Blowfish	Multi	118	254	271	324	511	929	55%
bzip2 Compress	1	91	125	123	147	137	224	61%
bzip2 Compress	Multi	92	253	231	293	531	858	62%
bzip2 Decompress	1	65	118	113	135	143	249	57%
bzip2 Decompress	Multi	67	248	232	278	597	1048	57%
Mandelbrot	1	61	125	118	141	143	180	79%
Mandelbrot	Multi	61	250	235	282	568	718	79%
JPEG Compress	1	100	119	92	110	123	159	77%
JPEG Compress	Multi	102	237	184	220	494	639	77%
JPEG Decompress	1	81	127	84	101	125	150	83%
JPEG Decompress	Multi	80	228	164	196	444	529	84%
Memory read sequential	1	28	134	198	211	200	341	59%
Memory read sequential	Multi	29	113	169	177	21	168	12%
Memory write sequential	1	80	166	235	227	339	487	70%
Memory write sequential	Multi	80	220	226	229	219	243	90%
OVERALL SCORE	—	64.4	162.0	163.5	176.3	242.9	341.4	71%

Testing was done with no other GUI applications running. The final column compares the Universal and Rosetta results for the Mac Pro, stating Rosetta’s figures as a percentage of the Universal figures. The higher the percentage, the better the performance under Rosetta as compared to native performance.

There are some interesting tidbits hiding in that table:

If you compare the “Rosetta % of Universal” columns between this test and the mini test back in March, it’s pretty obvious that Rosetta has improved (as we discuss in this Rosetta first look ). To make sure that this wasn’t just due to the Mac Pro, I also re-ran the Rosetta vs. Universal version of GeekBench on the Mac mini, and saw similar results—no longer is the average at or below 50 percent; it’s now at 62 percent on overall score, and many of the tests are at 90 percent or higher.

Although the PowerBook is well-suited to handling e-mail and Web surfing, it’s clearly showing its age in any computational tests. It’s overall score is only slight better than one-third that of its replacement in our household—the MacBook

My Dual PowerMac G5 is also showing its computational age. Although this test is far from a measure of real-world usability, there’s no doubt that the G5 CPUs are being outperformed by the Intel chips—even the lowly mini, with its Core Duo running at 1.67GHz, outscores the G5. Now, in real world usage, the G5 is still a much faster machine, thanks to its faster hard drive, higher memory bandwidth, and high-end graphics card. But the Intel Mac mini puts up a heck of a showing for a $799 machine.

Even when running in Rosetta mode, the Mac Pro is clearly the class of this benchmark suite. In native mode, it’s score is nearly twice that of its nearest competitor—the MacBook.

As I stated, this is definitely not a benchmark for measuring real-world usability—if it were, the G5 would have scored much higher than it did. What it does accurately reflect, however, is the power of the underlying processor(s) in each machine. In that sense, I find the results to be quite accurate. In my use of iMovie, for instance, I found that rendering transitions and effects took basically the same amount of time on the MacBook as it did on the Dual G5.

Cinebench benchmark: The free Cinebench benchmark uses the Cinema 4D engine to test the graphics performance of your Mac. It’s also available for Macs and PCs, so I tested everything I could. Back in March, that included my now-sold Athlon machine, so that’s still in the results table (as the benchmark hasn’t been updated, I felt it fair to leave in). I added my MacBook and the Mac Pro to the results, and with the Mac Pro, I included the results when running in Windows XP via Boot Camp—this to see if there were any platform-specific differences in the benchmark’s performance. When the dust settled, here’s how things came out (click the image for a much larger, easier-to-read version):

When I first published this chart, my old Athlon-powered PC was the champion, handily trouncing the G5 in the OpenGL tests, and basically tying it in the other areas. The Mac Pro put an end to that, and as you can see, by a wide margin. (I have no doubt that there are probably faster PCs out there—but I don’t have one available for testing). The PowerBook, as usual, is solidly in last place, as both the no-graphics-card MacBook and mini beat it out. I was pleasantly surprised to see no significant differences in the Mac Pro’s results regardless of whether it was running Windows XP (via Boot Camp) or Mac OS X—the XP box was about 2.5 percent quicker in the Open GL hardware test, and the Mac Pro was marginally quicker in all the other tests.

What that chart can’t show you is the visual difference between running the test on my Dual G5 and the Mac Pro. Though the G5 isn’t a slowpoke, the Mac Pro is over twice as fast when drawing the Open GL hardware-accelerated scene, and that kind of difference is obvious to the eye. As I’ve said many times before, the Mac Pro with the ATI X 1900XT video card is one fast machine.

Three takeaway points

There’s enough processing power in even a low-end Intel Mac to rival the Dual G5 in many CPU-intensive tasks.

The combination of the Mac Pro’s CPU power and ATI video card make for a potent OpenGL machine.

Unlike the situation with Quake4, there wasn’t a significant difference between Mac OS X and Windows when running the Cinebench benchmark.

Xbench: Although there’s a new version (1.3) of Xbench available, I was having some troubles getting what I felt to be believable figures from the program. One problem is that it doesn’t properly test for accelerated OpenGL graphics on machines (like my MacBook and mini) with a graphics chipset instead of a card. Also, I was getting very strange results on the “User Interface” portion of the test—the mini and the MacBook were both scoring about twice what the G5 was recording. Finally, the score for the Disk portion of the test on the G5 just seemed completely out of line—even testing on a newly-formatted drive, my score was roughly one-fifth of what it should have been for a stock G5.

Based on these odd results, I’ve chosen to hold off on any Xbench results until I can do some more testing with my machines (and hopefully, a new version is released that addresses some of the issues I experienced).

FireWire testing: As a new test this time, I thought I’d compare the time required to copy three sets of data (a single large file, a folder with tons of smaller files, and a small assortment of mid-sized files) to and from both FireWire 400 and FireWire 800 drives. I only tested the Mac Pro and the Dual G5, and I placed the three data sets to be copied onto a relatively clean (recently formatted) partition on each machine, then copied them from there to the FireWire drive. When copying the files back, I copied them back to a single partition on the Mac Pro (not the RAID 1 boot drive). Here’s what I found—the top figure in each cell is the number of seconds to complete the task, and the number below is the transfer rate in megabytes per second:

FIREWIRE PERFORMANCE TESTS

		Copy to FW Drive		Copy from FW Drive
		Dual G5	Mac Pro	Dual G5	Mac Pro
FW 400	High def video clips	196 30.3	207 28.6	172 34.5	170 34.9
	Parallels disk image	234 26.6	251 24.8	214 29.1	213 29.2
	iChat archives	61 12.3	38 19.8	48 15.6	32 23.5

FW 800	High def video clips	115 51.6	125 47.4	139 42.7	113 52.5
	Parallels disk image	168 37.0	156 39.9	145 42.9	141 44.1
	iChat archives	53 14.2	32 23.5	40 18.8	29 25.9

In each cell, the top figure is seconds and the bottom is megabytes per second. Best results in red

High-def video clips: 37 HD movie trailers organized into five folders, totaling 5.93GB. Parallels disk image: The Windows XP virtual drive for Parallels, 6.22GB in size. iChat archives: 10,026 files organized into 125 folders, plus 668 unfiled items, totaling 750.7MB in size.

This is obviously a far-from-complete test, and I didn’t do everything necessary to control all the variables as would be necessary in a real performance evaluation. Still, there are some interesting results to consider:

Whether copying to or from the FireWire 400 or FireWire 800 drives, the Dual G5 had serious issues with my iChat folder—the one with over 10,000 files and folders in it. I tried this test multiple times, from multiple different internal drives, and the results were basically the same. For whatever reason, it seems the Mac Pro doesn’t choke as badly on the huge folder as does the Dual G5. I welcome explanations here, because I really can’t figure out what’s going on. The fact that I saw similar results on both FireWire interfaces rules out any sort of drive-specific cause, too.

When copying to the FireWire drive (400 or 800), there wasn’t a huge amount of difference between the two machines. The Dual G5 was faster in three of the four remaining tests, but the difference was roughly 5 percent, which isn’t all that significant given my hand-timing work.

When copying from the FireWire drive (400 or 800), the Mac Pro was the quicker machine in the four non-huge-folder tests—though again, the differences weren’t all that significant (other than the high-def video clips on the FireWire 800 drive, which took 26 fewer seconds on the Mac Pro).

Overall, FireWire seems to work about as well on the Mac Pro as it does on the Dual G5—at least when copying to or from internal drives. I’ve heard a report that copies between FireWire drives on the various buses are slower on the Mac Pro, and I plan on doing some testing in that area for a future report.

Three takeaway points

The Mac Pro appears to handle large numbers of files in copies to/from FireWire drives much better than does the Dual G5. I can’t say that this is a global issue, but it was certainly repeatable in my testing, regardless of the source and destination disks.

FireWire 800 drives are faster than the first-generation FireWire 400 drives, but not twice as fast. I was seeing improvements of anywhere from 10 percent to 30 percent.

When working with FireWire 800, the Mac Pro proved faster in five of the six test cases.