Tiger performance

[danman]

It's early days yet, but judging by the Developer preview, we should not (as I kind of expected) hold out for any more major speed increases in the step from Panther to Tiger.

This appears to be a shame, as advances in compiler technology from IBM that clearly show a great improvement in code generation over GCC 3 seem to be being ignored by Apple.

Herewith some current benchmarks from XBench, clearly showing slowdowns/ or flat performance between the two OS Xs...



Results 133.39
System Info
Xbench Version 1.1.3
System Version 10.4 Pre-release (8A162)
Physical RAM 768 MB
Model PowerBook5,2
Processor PowerPC G4 @ 1.25 GHz
L1 Cache 32K (instruction), 32K (data)
L2 Cache 512K @ 1.25 GHz
Bus Frequency 167 MHz
Video Card ATY,RV350M10
Drive Type FUJITSU MHT2080AT
CPU Test 151.77
GCD Loop 146.64 5.73 Mops/sec
Floating Point Basic 152.34 550.92 Mflop/sec
AltiVec Basic 154.00 4.47 Gflop/sec
vecLib FFT 153.75 2.39 Gflop/sec
Floating Point Library 152.34 6.10 Mops/sec
Thread Test 110.61
Computation 80.35 1.08 Mops/sec, 4 threads
Lock Contention 177.40 2.23 Mlocks/sec, 4 threads
Memory Test 131.56
System 145.77
Allocate 585.07 381.64 Kalloc/sec
Fill 159.18 1267.09 MB/sec
Copy 79.44 397.19 MB/sec
Stream 119.88
Copy 124.49 910.02 MB/sec [altivec]
Scale 126.01 929.93 MB/sec [altivec]
Add 118.55 758.74 MB/sec [altivec]
Triad 111.57 681.70 MB/sec [altivec]
Quartz Graphics Test 169.78
Line 129.41 3.29 Klines/sec [50% alpha]
Rectangle 156.87 11.04 Krects/sec [50% alpha]
Circle 159.57 3.68 Kcircles/sec [50% alpha]
Bezier 150.49 1.64 Kbeziers/sec [50% alpha]
Text 410.44 6.69 Kchars/sec
OpenGL Graphics Test 100.24
Spinning Squares 100.24 70.15 frames/sec
User Interface Test 229.27
Elements 229.27 73.75 refresh/sec
Disk Test 48.81
Sequential 48.36
Uncached Write 44.06 18.37 MB/sec [4K blocks]
Uncached Write 38.06 15.59 MB/sec [256K blocks]
Uncached Read 105.39 16.68 MB/sec [4K blocks]
Uncached Read 41.23 16.66 MB/sec [256K blocks]
Random 49.26
Uncached Write 56.43 0.85 MB/sec [4K blocks]
Uncached Write 41.32 9.32 MB/sec [256K blocks]
Uncached Read 53.66 0.35 MB/sec [4K blocks]
Uncached Read 48.44 9.97 MB/sec [256K blocks]





Results 135.71
System Info
Xbench Version 1.1.3
System Version 10.3.4 (7H63)
Physical RAM 768 MB
Model PowerBook5,2
Processor PowerPC G4 @ 1.25 GHz
L1 Cache 32K (instruction), 32K (data)
L2 Cache 512K @ 1.25 GHz
Bus Frequency 167 MHz
Video Card ATY,RV350M10
Drive Type FUJITSU MHT2080AT
CPU Test 151.86
GCD Loop 146.59 5.72 Mops/sec
Floating Point Basic 153.17 553.91 Mflop/sec
AltiVec Basic 153.48 4.46 Gflop/sec
vecLib FFT 157.97 2.45 Gflop/sec
Floating Point Library 148.60 5.95 Mops/sec
Thread Test 109.95
Computation 79.99 1.08 Mops/sec, 4 threads
Lock Contention 175.78 2.21 Mlocks/sec, 4 threads
Memory Test 140.12
System 155.63
Allocate 786.50 513.04 Kalloc/sec
Fill 163.43 1300.91 MB/sec
Copy 84.13 420.67 MB/sec
Stream 127.41
Copy 128.81 941.59 MB/sec [altivec]
Scale 130.21 960.96 MB/sec [altivec]
Add 128.28 820.96 MB/sec [altivec]
Triad 122.63 749.27 MB/sec [altivec]
Quartz Graphics Test 154.54
Line 136.05 3.46 Klines/sec [50% alpha]
Rectangle 128.48 9.04 Krects/sec [50% alpha]
Circle 156.77 3.61 Kcircles/sec [50% alpha]
Bezier 152.73 1.66 Kbeziers/sec [50% alpha]
Text 232.83 3.80 Kchars/sec
OpenGL Graphics Test 99.02
Spinning Squares 99.02 69.29 frames/sec
User Interface Test 217.21
Elements 217.21 69.87 refresh/sec
Disk Test 61.72
Sequential 65.00
Uncached Write 60.71 25.31 MB/sec [4K blocks]
Uncached Write 54.43 22.29 MB/sec [256K blocks]
Uncached Read 126.76 20.07 MB/sec [4K blocks]
Uncached Read 53.17 21.48 MB/sec [256K blocks]
Random 58.76
Uncached Write 67.88 1.02 MB/sec [4K blocks]
Uncached Write 52.18 11.77 MB/sec [256K blocks]
Uncached Read 57.41 0.38 MB/sec [4K blocks]
Uncached Read 59.68 12.28 MB/sec [256K blocks]

[bitfactory]

that's precisely your problem - you're "judging by the Developer Preview"

folks, this release is AT LEAST 7 months away. keep that in perspective.

[MisterMe]

It's early days yet, but judging by the Developer preview, we should not (as I kind of expected) hold out for any more major speed increases in the step from Panther to Tiger.

This appears to be a shame, as advances in compiler technology from IBM that clearly show a great improvement in code generation over GCC 3 seem to be being ignored by Apple.

Herewith some current benchmarks from XBench, clearly showing slowdowns/ or flat performance between the two OS Xs...



Results 133.39
System Info
Xbench Version 1.1.3
System Version 10.4 Pre-release (8A162)
Physical RAM 768 MB
Model PowerBook5,2
Processor PowerPC G4 @ 1.25 GHz
L1 Cache 32K (instruction), 32K (data)
L2 Cache 512K @ 1.25 GHz
Bus Frequency 167 MHz
Video Card ATY,RV350M10
Drive Type FUJITSU MHT2080AT
CPU Test 151.77
GCD Loop 146.64 5.73 Mops/sec
Floating Point Basic 152.34 550.92 Mflop/sec
AltiVec Basic 154.00 4.47 Gflop/sec
vecLib FFT 153.75 2.39 Gflop/sec
Floating Point Library 152.34 6.10 Mops/sec
Thread Test 110.61
Computation 80.35 1.08 Mops/sec, 4 threads
Lock Contention 177.40 2.23 Mlocks/sec, 4 threads
Memory Test 131.56
System 145.77
Allocate 585.07 381.64 Kalloc/sec
Fill 159.18 1267.09 MB/sec
Copy 79.44 397.19 MB/sec
Stream 119.88
Copy 124.49 910.02 MB/sec [altivec]
Scale 126.01 929.93 MB/sec [altivec]
Add 118.55 758.74 MB/sec [altivec]
Triad 111.57 681.70 MB/sec [altivec]
Quartz Graphics Test 169.78
Line 129.41 3.29 Klines/sec [50% alpha]
Rectangle 156.87 11.04 Krects/sec [50% alpha]
Circle 159.57 3.68 Kcircles/sec [50% alpha]
Bezier 150.49 1.64 Kbeziers/sec [50% alpha]
Text 410.44 6.69 Kchars/sec
OpenGL Graphics Test 100.24
Spinning Squares 100.24 70.15 frames/sec
User Interface Test 229.27
Elements 229.27 73.75 refresh/sec
Disk Test 48.81
Sequential 48.36
Uncached Write 44.06 18.37 MB/sec [4K blocks]
Uncached Write 38.06 15.59 MB/sec [256K blocks]
Uncached Read 105.39 16.68 MB/sec [4K blocks]
Uncached Read 41.23 16.66 MB/sec [256K blocks]
Random 49.26
Uncached Write 56.43 0.85 MB/sec [4K blocks]
Uncached Write 41.32 9.32 MB/sec [256K blocks]
Uncached Read 53.66 0.35 MB/sec [4K blocks]
Uncached Read 48.44 9.97 MB/sec [256K blocks]





Results 135.71
System Info
Xbench Version 1.1.3
System Version 10.3.4 (7H63)
Physical RAM 768 MB
Model PowerBook5,2
Processor PowerPC G4 @ 1.25 GHz
L1 Cache 32K (instruction), 32K (data)
L2 Cache 512K @ 1.25 GHz
Bus Frequency 167 MHz
Video Card ATY,RV350M10
Drive Type FUJITSU MHT2080AT
CPU Test 151.86
GCD Loop 146.59 5.72 Mops/sec
Floating Point Basic 153.17 553.91 Mflop/sec
AltiVec Basic 153.48 4.46 Gflop/sec
vecLib FFT 157.97 2.45 Gflop/sec
Floating Point Library 148.60 5.95 Mops/sec
Thread Test 109.95
Computation 79.99 1.08 Mops/sec, 4 threads
Lock Contention 175.78 2.21 Mlocks/sec, 4 threads
Memory Test 140.12
System 155.63
Allocate 786.50 513.04 Kalloc/sec
Fill 163.43 1300.91 MB/sec
Copy 84.13 420.67 MB/sec
Stream 127.41
Copy 128.81 941.59 MB/sec [altivec]
Scale 130.21 960.96 MB/sec [altivec]
Add 128.28 820.96 MB/sec [altivec]
Triad 122.63 749.27 MB/sec [altivec]
Quartz Graphics Test 154.54
Line 136.05 3.46 Klines/sec [50% alpha]
Rectangle 128.48 9.04 Krects/sec [50% alpha]
Circle 156.77 3.61 Kcircles/sec [50% alpha]
Bezier 152.73 1.66 Kbeziers/sec [50% alpha]
Text 232.83 3.80 Kchars/sec
OpenGL Graphics Test 99.02
Spinning Squares 99.02 69.29 frames/sec
User Interface Test 217.21
Elements 217.21 69.87 refresh/sec
Disk Test 61.72
Sequential 65.00
Uncached Write 60.71 25.31 MB/sec [4K blocks]
Uncached Write 54.43 22.29 MB/sec [256K blocks]
Uncached Read 126.76 20.07 MB/sec [4K blocks]
Uncached Read 53.17 21.48 MB/sec [256K blocks]
Random 58.76
Uncached Write 67.88 1.02 MB/sec [4K blocks]
Uncached Write 52.18 11.77 MB/sec [256K blocks]
Uncached Read 57.41 0.38 MB/sec [4K blocks]
Uncached Read 59.68 12.28 MB/sec [256K blocks]The fact that this is a developers' release means nothing to you?

[stoid]

From what I hear, the early releases of Panther were pretty unoptimized and slow. I would expect that this code sees a major optimization overhaul before the Gold Master release.

[danman]

that's precisely your problem - you're "judging by the Developer Preview"

folks, this release is AT LEAST 7 months away. keep that in perspective.

Neither of you can read can you? What is the very first thing I said in my post?

[blue&whiteman]

that's precisely your problem - you're "judging by the Developer Preview"

folks, this release is AT LEAST 7 months away. keep that in perspective.

exactly. when panther was at the dp stage it got faster with each build and even a little faster with each update since the final. the difference with panther though is that it was showing big gains in performance over jaguar in the very first dp build. tiger will get faster for sure as builds are released but the question is how much faster.

[bousozoku]

From what I hear, the early releases of Panther were pretty unoptimized and slow. I would expect that this code sees a major optimization overhaul before the Gold Master release.

Early releases of each Mac OS X version have been slow until about 60 percent through the development cycle. It's always been a developing adage to get it right first, then make it fast.

danman:

While it's possible that Apple might use IBM's compilers to help Tiger code, there is no good reason for this to happen. In fact, it might lead to negative results among those trying to re-compile the Darwin portion with gcc 3.x. Also, the IBM compilers explicitly support G5 and G4 processors, but not G3s. Do you really want to alienate that many users?

[blue&whiteman]

Also, the IBM compilers explicitly support G5 and G4 processors, but not G3s. Do you really want to alienate that many users?

ibm is now the sole manufacturer of G3 cpu's so why would they not support their own product?

[JFreak]

beta software always carries debug code and that is jus unnecessary slowdown compared to final release. even if nothing else changes but the debug code is removed, there will be a significant performance boost.

just focus on features, not performance. the function of a developer preview is to give developers new frameworks and other hooks that they can use in programming their own code - those frameworks are not necessarily even finished yet, and you guys whine about developer release not being optimized ;) come on... give apple some time. if you NEED to use the developer preview, you will value the features and couldn't frankly care less about the performance. if you're a developer, you know how the process goes.

as someone said, apple first does it right and then makes it fast. it takes time. microsoft however is very fast in doing it wrong ;)

[bousozoku]

ibm is now the sole manufacturer of G3 cpu's so why would they not support their own product?

They may support the G3 but they don't say explicitly that there is support in the Mac OS X version of the XLC compiler for it. This may only mean that the code exists but they have not tested it or it may mean that it doesn't exist at all. Obviously, since the AIX version has more legacy, it may have to support the G3 and perhaps, even the 604/604e but there is always gcc.