hatahata Posted May 3, 2016 Posted May 3, 2016 hi all i try unixbench (Version 5.1.3) compared with orange pi PC's System Benchmarks Index Score 202.0 Benchmark Run: ç« 5月 03 2016 23:39:28 - 00:07:310 CPUs in system; running 1 parallel copy of testsDhrystone 2 using register variables 7827597.5 lps (10.0 s, 7 samples)Double-Precision Whetstone 915.0 MWIPS (9.8 s, 7 samples)Execl Throughput 459.3 lps (29.9 s, 2 samples)File Copy 1024 bufsize 2000 maxblocks 176763.8 KBps (30.0 s, 2 samples)File Copy 256 bufsize 500 maxblocks 53800.0 KBps (30.0 s, 2 samples)File Copy 4096 bufsize 8000 maxblocks 431407.5 KBps (30.0 s, 2 samples)Pipe Throughput 430477.6 lps (10.0 s, 7 samples)Pipe-based Context Switching 73425.8 lps (10.0 s, 7 samples)Process Creation 1360.4 lps (30.0 s, 2 samples)Shell Scripts (1 concurrent) 1188.6 lpm (60.0 s, 2 samples)Shell Scripts (8 concurrent) 783.3 lpm (60.1 s, 2 samples)System Call Overhead 969171.8 lps (10.0 s, 7 samples)System Benchmarks Index Values BASELINE RESULT INDEXDhrystone 2 using register variables 16700.0 7827597.5 670.7Double-Precision Whetstone 55.0 915.0 166.4Execl Throughput 43.0 459.3 106.8File Copy 1024 bufsize 2000 maxblocks 3960.0 176763.8 446.4File Copy 256 bufsize 500 maxblocks 1655.0 53800.0 325.1File Copy 4096 bufsize 8000 maxblocks 5800.0 431407.5 743.8Pipe Throughput 12440.0 430477.6 346.0Pipe-based Context Switching 4000.0 73425.8 183.6Process Creation 126.0 1360.4 108.0Shell Scripts (1 concurrent) 42.4 1188.6 280.3Shell Scripts (8 concurrent) 6.0 783.3 1305.6System Call Overhead 15000.0 969171.8 646.1 ========System Benchmarks Index Score 336.2surely odroid-c2's score is higer than orane pi PC ( 336.2 X 202.0) . and old intel PC , smpboot: CPU0: Intel® Celeron® CPU 540 @ 1.86GHz (fam: 06, model: 16, stepping: 01) System Benchmarks Index Score 616.3 only result , i cannot analyze . --- regards
tkaiser Posted May 3, 2016 Posted May 3, 2016 i try unixbench https://twitter.com/brendangregg/status/618573280375484417 https://github.com/GoogleCloudPlatform/PerfKitBenchmarker/issues/125 Please people, stop fooling yourself using such crap and comparing numbers without meaning! It's absolutely useless to run this stuff from last century.
hatahata Posted May 3, 2016 Author Posted May 3, 2016 i see unixbench was outdated . i want to mesure cpu speed not hard disk speed . i do not know the tool . Once upon a time there was super pi and i used it . at present whic one is easy to do to measure cpu speed ? orange pi PC with armbian is very tuned up and is fast and i think it may be faster than raspberry pi 3 with rasbian . i think odroid-c2 must be fast considering its hardware .
tkaiser Posted May 3, 2016 Posted May 3, 2016 Why do you care about measuring cpu speed if you already noticed what's important? It always depends 100% on the use case. If you want to do number crunching... then CPU performance is important if you can't use GPGPU. Trying number crunching on slow ARM SoCs is moronic. So why not just dropping the whole idea to measure CPU performance at all? When you use a desktop Linux then get fast storage with superiour random I/O (modern eMMC) and have a look whether you get 2D acceleration (not CPU related) and 3D acceleration (not CPU related) and video acceleration (not CPU related). And then stuff like IRQ balancing, good dvfs operating point tuning to get the best thermal behaviour and so on is more important (that's what Armbian does: focusing on the relevant tuning parameters!) Simply stop stupid CPU performance benchmarking. Most if not all of the times this is only producing worthless numbers (that look pretty nice when graphs were made of so that really dumb people can draw conclusions from the length of graphs instead of switching on their brains to understand what's really important to get good 'real world' performance). These useless kitchen-sink micro-benchmarks are for morons and/or for phoronix.com. Stuff like "quad core is better than dual core IF the workload makes use of parallelism" are so easy to understand that no one needs benchmarks to 'verify' this.
hatahata Posted June 6, 2016 Author Posted June 6, 2016 i tried sysbench . sysbench --test=cpu run the result is dmesg | grep CPU0 CPU0: Intel® Celeron® CPU 540 @ 1.86GHz (fam: 06, model: 16, stepping: 01)Threads started!Done.Maximum prime number checked in CPU test: 10000Test execution summary: total time: 23.2413s total number of events: 10000 total time taken by event execution: 23.2258 per-request statistics: min: 2.20ms avg: 2.32ms max: 10.99ms approx. 95 percentile: 2.39msThreads fairness: events (avg/stddev): 10000.0000/0.00 execution time (avg/stddev): 23.2258/0.00 odoroid-c2 over internet Threads started!Done.Maximum prime number checked in CPU test: 10000Test execution summary: total time: 9.6456s total number of events: 10000 total time taken by event execution: 9.6433 per-request statistics: min: 0.96ms avg: 0.96ms max: 1.42ms approx. 95 percentile: 0.97msThreads fairness: events (avg/stddev): 10000.0000/0.00 execution time (avg/stddev): 9.6433/0.00 odroid-c2 is much faster than Celeron® 1.86GHz
tkaiser Posted June 6, 2016 Posted June 6, 2016 odroid-c2 is much faster than Celeron® 1.86GHz Plain wrong, sysbench ist just one of the many broken 'benchmarks' that can not be used to compare between different platforms. Unless your use case is calculating prime numbers sysbench results are 100 percent worthless. If you let it run on a more recent ultra slow Intel Atom it will outperform S905. And that S905 looks that good with this one specific sysbench test is due to using ARMv8 instruction set. Some details: http://forum.odroid.com/viewtopic.php?f=136&t=19158 And the basic problem regarding mornonic kitchen-sink benchmarks: http://www.brendangregg.com/blog/2014-05-02/compilers-love-messing-with-benchmarks.html (by choosing wrong compiler settings your ODROID-C2 might be 100 times slower. And that's why this sort of 'fire and forget' benchmarking is always crap)
hatahata Posted June 6, 2016 Author Posted June 6, 2016 i give up benchmark . at odroid-c2(armbian and over internet ) date ; ./pi.bat ; date2016å¹´ 6月 7æ—¥ ç«æ›œæ—¥ 05:15:29 JST3.141592653592016å¹´ 6月 7æ—¥ ç«æ›œæ—¥ 05:16:06 JST37 sec at Intel® Celeron® CPU 540 @ 1.86GHz (kali linux) $ date ; ./pi.bat ; date2016å¹´ 6月 7æ—¥ ç«æ›œæ—¥ 05:16:17 JST3.141592653592016å¹´ 6月 7æ—¥ ç«æ›œæ—¥ 05:16:28 JST11 sec here cat pi.bat #!/usr/bin/env pythondef f(x): return 4 / ( 1.0 + x**2 )n = 10000000sum = 0step = 1.0 / nfor i in range(0, n): x = ( i + 0.5 ) * step sum += f(x) pi = sum * stepprint pi i get pi information at http://netbuffalo.doorblog.jp/archives/4783572.html i tried arch linux on odroid-c2 , but almost all aur fails . and in openbsd , source(ports) is common to all architecture . .this is suprising .
Recommended Posts