Jaguar, Panther, and Tiger Benchmarks
Update: Results for a PowerBook G4 are available here.
We recently got a chance to play with Tiger, and with the help of Xbench, thought it might be interesting to see how Tiger compares to Jaguar and Panther in terms of performance. We ran Xbench three times under all three operating systems and took the average score of the three runs. Here is the configuration of our test machine:
- Power Mac G5 1.6GHz
- 768MB DDR333 RAM
- nVidia GeForce FX 5200 Ultra w 64MB
- Mac OS X 10.2.8 (6S90), Mac OS X 10.3.4 (7H63), or Mac OS X 10.4 (8A162)
- Xbench 1.1.3
Here are the results of the test runs. Along with the Xbench scores, we’ve included the score expressed as a percentage of the previous version’s score. As always, higher is better.
| Test | Jaguar | Panther | Tiger |
|---|---|---|---|
| Overall | 122 | 130 (107%) | 133 (102%) |
| CPU | 118 | 118 (100%) | 119 (101%) |
| Thread | 72 | 76 (106%) | 79 (103%) |
| Memory | 199 | 214 (107%) | 217 (101%) |
| Quartz | 156 | 183 (118%) | 198 (108%) |
| OpenGL | 170 | 159 (93%) | 158 (100%) |
| UI | 140 | 210 (150%) | 200 (95%) |
| Disk | 94 | 92 (98%) | 95 (104%) |
Looking at the results, it appears that Tiger is as fast (if not faster) than Panther in all areas except for UI performance. Given that Tiger is still pre-release software, it’s entirely possible that Tiger’s performance will change as development continues. However, given that Tiger is already faster than Panther, we think it’s likely that Tiger’s performance will continue to improve, and that it will be even faster and more responsive than Panther when it is released next year.
Would it be possible to run the test on a non G5 system? It would be nice to see what the speed changes would be if not for the 64 bit optimization.
Awesome!!!
I would like to know how well Tiger will run on my G3 500 mhz Classic iMac. Panther runs great on it now, how will Tiger do?
Great job guys briliant subject for review. I would be very interested in seeing dual processor performance comparison. I am awed by the diference in system responsivnes in real world situations with my dual 1 gig compared to faster single cpu machines. If I have many apps open as I always do, everything is faster on my dual. Multitasking is in a class by itself compared to any other OS I have experienced when used in a dual processor invironment. If you have the chance to test with a dual G5, I’d love it!
thanks for the great work.
in the overall performance score, one of the numbers is wrong. if tiger’s score is higher than panther’s, then the percentage should be too.
I too believe your percentages are calculated wrong. Ex. Overall, 133/122=1.09 or a 9% improvement, while the Open GL (93%) number is MUCH lower and the UI number is higher (143%)…. Changing the the conclusion of your article. Sorry to say sloppy math skills :’(
I think Panther percentages show how much faster it is than Jaguar. The Tiger percentages are in relation to Panther.
Math skills … People, take a second look before you critisize. The Panther percentages are % increase from Jaguar (release-to-release). The Tiger percentages are % increase from Panther (release-to-release).
The conclusion of the article was that Tiger was an improvement OVER Panther in all areas besides UI. Yes, Tiger is also an improvement over Jaguar in UI, but that’s beside the point (you can get more of a UI speedup with the currently-released Panther).
Please, learn critical reading skills before criticizing someone else’s numbers!
Jet—
It’s not a percent increase, it’s absolute percentage when compared to the previous release. Looking at the overall numbers, Panther had a 7% increase in Xbench score compared to Jaguar. That is, Panther’s score is 107% of Jaguar’s score.
Similarly, Tiger’s score represents a 2% increase over Panther’s score (not Jaguar’s score, as you correctly pointed out). Again, in other words, that means that Tiger’s score is 102% of Panther’s score.
In pure math, simply multiply the previous score by the given percentage to get the new score (i.e.: 122 * 107% = 130, 130 * 102% = 133).
So it’s not percent increase, but the math isn’t sloppy, and it does make sense.
Actually I think the math would be better done like this… in terms of increase/decrease from Jaguar..
Panther Tiger (early developer preview)
Overall 6.5% 9.0%
CPU 0.0% 0.8%
Thread 5.6% 9.7%
Memory 7.5% 9.1%
Quartz 17.3% 21.2%
OpenGL -6.5% -7.1%
UI 50% 42.9%
Disk -2.1% 1.1%
Tiger relative to Panther…
Overall 2.3%
CPU 0.8%
Thread 3.9%
Memory 1.4%
Quartz 8.2%
OpenGL -0.6%
UI -4.8%
Disk 3.3%
Also note that developer preview of Tiger doesn’t have some rather nice performance enhancements turned on by default.
Shawn: “Also note that developer preview of Tiger doesn’t have some rather nice performance enhancements turned on by default.”—> Do you or anyone else know how to turn them on?
Yes I do… but that is under NDA and the above information should be consider under NDA as well… it really shouldn’t be posted.
Sweet deal… I cannot wait for my copy! Come on kitty!
That’s great, but wouldn’t Xbench need to be recompiled to take advantage of Tiger?
Good Test!
No it doesn’t need a recompile.
Of course a recompile with the experimental compiler available with Tiger (GCC 3.5) could yield improvements in the compiled code assuming proper compiler settings got used, etc. Of course that would skew the results since the benchmarking tool bits would change relative to existing measurements.
guys,
talking about math… if you found that your tests on one G5 machine showed an average of 102 percent of the panther results, that doesn’t really mean that tiger is faster than (or as fast as) panther. i’m simply not sure if your results are significant—given that you might have quite a bit of variance in speed on several machines. But I trust Apple is running these regression tests anyways—possibly nightly (does anybody know?)
Apart from that: as a user, you don’t really feel a speed increase that’s below about 20-30 percent. And I don’t know what that UI test measures, but I guess it’s rather graphics performance rather than snappiness…
XBench 1.1.3 results for a mostly idle 1.25Ghz running 10.3.4:
Overall: 131.11
CPU: 151.21
Thread: 109.69
Memory: 129.91
Quartz: 131.42
OpenGL: 108.93
UI: 210.38
Disk: 118.75
Hmm.
So a G5 1.6 scores only 118 in CPU and a 12” PowerBook scores 95? Hmmmmmmm indeed!
http://www.barefeats.com/pb17.html
Yeah. Somethings odd about the xBench scores. My Sawtooth upgraded with dual 1.3GHz processors. got 131.1.
Check out the Sawtooth category (include upgraded ones):
http://ladd.dyndns.org/xbench/comparesubindex.xhtml?machineTypeID=4&sort=score#tableTop&minVersion=1.1.3
And the G5 category:
http://ladd.dyndns.org/xbench/comparesubindex.xhtml?machineTypeID=22&sort=score#tableTop&minVersion=1.1.3
Does xBench run properly on a G5?? I know that it is NOT dual processor aware—not that that applies in this instance—but it did with my test:
http://ladd.dyndns.org/xbench/merge.xhtml?doc1=55267
Apple is writing the codes to make performance faster on every new version of OS X.
Zeta works fine on my PII 450 Mhz … and it’s now all in SVG too … but if you have a lot of money to spend, a G5 is good
Tiger runs pretty well on the 500mHz iMac Indigo. I ran it for a little while and everything seemed fine to me. No major or minor crashes.
Through all of this you have to remember that X-Bench is extremely variable from run to run. Instead of single scores, I’d like to see average scores over the period of several runs.
randomnessguy, you should read the first paragraph of the review:
We ran Xbench three times under all three operating systems and took the average score of the three runs.
Maybe the Developer Preview is compiled with full debugging symbols, decreasing the real performances..
Maybe it’s a G5 thing (though I believe I’ve seen a benchmark somewhere to the contrary), but I have found that Tiger underperforms in almost all areas of XBench, when compared to Panther.
I don’t take to much from this tho, as has already been mentioned, it’s early days yet
Well, I finally got a 160GB external Firewire drive to back up my entire life. THEN, I installed Tiger on another firewire drive and tested it using my two internal ATA66 2MB cache’d drives (other specs: “Sawtooth” dual 1.3GHz, 2GB RAM, ATI Radeon 9000 Pro 64MB—but dual monitors, so it’s 32MB on the test):
PANTHER: TIGER:
IBM Deskstar
Score: 124.43 130.69
CPU: 159.66 160.11
Thread: 154.99 199.98
Memory: 99.61 104.87
Quartz: 154.07 167.28
OpenGL: 107.88 102.12
UI: 213.40 246.10
Disk: 76.60 77.20
Seagate Barracuda:
Score: 130.35 127.68
CPU: 159.83 161.94
Thread: 160.02 200.33
Memory: 99.44 106.07
Quartz: 154.41 160.42
OpenGL: 107.29 102.09
UI: 206.58 247.64
Disk: 95.24 70.61
Big jumps in thread and UI scores. Decent improvements in memory and Quartz. Slight drop in OpenGL scores. Disk scores are all over the map.
Remember that xBench is NOT dual processor aware!
As for running Tiger, the Setup assistant never completed the “Calculating Size” of my Panther Applications and User folders. I let it try all night and came downstairs in the morning to find Tiger sleeping on the job! So, I skipped that whole thing. the Finder feels faster. Window redraws in Brushed Metal are virtually realtime, now
Spotlight is very nice in the Finder, although it would take a little time to get the new functions really under your belt to know what you’re doing with the results (or modifying their display). Spotlight in System Preferences are absolutely WOW!! It’ll make a lazy man out of me!
I connect wirelessly to the internet, and after installing the Belkin drivers, I was left with kernel panics during startup. DRAG!
So, i booted off my Panther install CD and changed startup disks back to my main panther install.
All in all, I think the speed improvements will be nice yet again, and spotlight will be handy. Dashboard? It’s too early to tell. As we all know, it’s a lot like Konfabulator. But, there is A LOT more technology that Dashboard has at its disposal than Konfabulator that its future looks bright once we start seeing the third parties take advantage of it.
get a life!
You should try Cinebench instead. There is a G5 optimized beta version BTW…
what is with those vw geek patrol cars?
Test comment
Excellent tests!
On a similar vein, we’re weighing the possibility of switching to the G5 dual X Serve platform to deliver our media rich products. Can anyone point me to a comparitive study of Xeon and X Servers that contains the type of data shown here? We’re presently running Pro2000 IIS, JRun on dual Xeon Compaqs. I hear alot of bragging by MS about Pro3000, but no xserve comparisons to be found.
Something to think about