Overclocking of OP+, my experience

Vlad · Published in 2015-11-7 04:16:06

Edited by Vlad at 2015-11-7 06:30

According to temprerature monitor and frequency control i would have max freq as:
* 1008000 for a full load of all the cores by ffmpeg conversion (t is 65C). This case is ok, i can it without problem.
* 2000000, may be more, for single core compilation of a sources (t is 55C on 1536000 freq). This limit isn't available

But!
on the new kernel the cpu freq can't be assigned above 1536000

How to assign some above 1536000 to /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq ???

P.S. Heatsinks are attached to cpu and mem chips on my OP+

hojnikb · Published in 2015-11-9 18:55:38

Looks like its hardcoded into the kernel.

Although i doubt CortexA7 can handle any more than ~1.6Ghz.... 1.53G is already pretty overclocked for such design... Normally, cortex a7 designs runs at ~1000-1200Mhz tops.

Vlad · Published in 2015-11-15 23:38:33

hojnikb replied at 2015-11-9 18:55
Looks like its hardcoded into the kernel.

Although i doubt CortexA7 can handle any more than ~1.6Gh ...

I have to point on the insteresting lines in the official ubuntu vivid mini distribution (/etc/rc.local):

# ** Overclock to 1.728 GHz 1728000
# echo 1728000 > /sys/devices/system/cpu/cpu0/cpufreq/scaling_max_freq

So. I am pretty shure kernel can overclock CPU to 1728000hz without rebuild. Can't?

What about 2,000,000 hz?

bronco · Published in 2015-11-18 03:28:55

Edited by bronco at 2015-11-18 03:29

Vlad replied at 2015-11-15 23:38
I am pretty shure kernel can overclock CPU to 1728000hz without rebuild.

That's bullsh*t: Please have a look at http://linux-sunxi.org/Orange_Pi_PC#CPU_clock_speed_limit

The "up to 1.6 GHz" is marketing stuff.

tokka · Published in 2015-11-18 17:27:13

Usually these cpu can run at 2ghz, but....this freq is shared with gpu, so you have to take off 300mhz for it.
If you want to do some test, you have to change the values in script.bin too, setting up voltages related to freq..
It's not wasy to find the correct balance with cpu, gpu and voltages.
Good luck

bronco · Published in 2015-11-18 18:19:35

Edited by bronco at 2015-11-18 18:24

tokka replied at 2015-11-18 17:27
Usually these cpu can run at 2ghz, but....this freq is shared with gpu, so you have to take off 300m ...

Memory is shared between CPU and GPU. How could 'frequencies' be shared between different IC components?

There's a 'thermal budget' associated with running both CPU and GPU intensive tasks and that's what thermal throttling (and the H3's "Thermal Sensor Controller (TSC) providing over-temperature protection interrupt and over-temperature alarm interrupt") is for. And there are voltage settings and a maximum core voltage of 1.4V defined in the H3's datasheet. Again: http://linux-sunxi.org/Orange_Pi_PC#CPU_clock_speed_limit

; extremity_freq(Hz): cpu extremity frequency when run benckmark or demo apk
; 1536MHz@1500mV with radiator, 1296MHz@1340mV without radiator
; max_freq: cpu maximum frequency, based on Hz, can not be more than 1200MHz
; min_freq: cpu minimum frequency, based on Hz, can not be less than 60MHz

Even the 1.5 GHz @ 1.5V are insane and for benchmarking only. These so called 'extremity frequencies' are for tablet and OTT manufacturers and should be used while thermal throttling is disabled to fool their customers since this way high Antutu scores can be shown when they produce product videos demonstrating the performance of their device. These speeds are not meant for reliable use.

It seems none of the 'this chip is able to run at 1.5 GHz or above' apologists do care about stability and data integrity? I've found not a single reference to cpufreq-ljt-stress-test in this forum or anywhere else related to H3. It's all about overheating and instability issues.

bronco · Published in 2015-11-18 19:56:11

Is any of you overclocking fans able to provide the results (whole output) of these three short benchmarks:

sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=4 && echo -e "\n\n" && 7za b && echo -e "\n\n" && mbw -t0 256 ; mbw -t1 256 ; mbw -t2 256

Copy code

(you might have to do an "sudo apt-get install sysbench p7zip-full mbw" before and it would be interesting whether CPU cores were shut down and which cpufreq settings you used)

I found only references on the net where measurements were wrong or couldn't be used due to thermal throttling and unknown cpu-max-prime value:

http://tech.scargill.net/orange-pi-pc-battle-of-the-pis/
http://www.orangepi.org/orangepi ... 9&fromuid=29411

loboris · Published in 2015-11-18 23:02:55

Edited by loboris at 2015-11-18 16:10

bronco replied at 2015-11-18 12:56
Is any of you overclocking fans able to provide the results (whole output) of these three short benc ...

i'm not an overclocking fan, but here are the tests (1.53 GHz constant during the test, no thermal throttling during the test; OPI-2; heatsink&fan):

Options -md23 in 7za and 112 in mbw used not to get Out of memory error.

root@OrangePI:~# uname -r
3.4.39-01-lobo

root@OrangePI:~# sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=4 && echo -e "\n\n" && 7za b -md23 && echo -e "\n\n" && mbw -t0 112
; mbw -t1 112 ; mbw -t2 112
sysbench 0.4.12:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 4

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000


Test execution summary:
total time:                         123.3567s
total number of events:             10000
total time taken by event execution: 493.3652
per-request statistics:
      min:                               48.98ms
      avg:                               49.34ms
      max:                               108.22ms
      approx.  95 percentile:             49.50ms

Threads fairness:
events (avg/stddev):          2500.0000/20.14
execution time (avg/stddev): 123.3413/0.01





7-Zip (A) 9.20  Copyright (c) 1999-2010 Igor Pavlov  2010-11-18
p7zip Version 9.20 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,4 CPUs)

RAM size: 1002 MB,  # CPU hardware threads: 4
RAM usage: 226 MB,  # Benchmark threads:    4

Dict       Compressing       |       Decompressing
   Speed Usage R/U Rating  | Speed Usage R/U Rating
   KB/s    % MIPS MIPS  |    KB/s    % MIPS MIPS

22: 1640 307 520 1595  | 42528 382 1005 3837
23: 1656 315 536 1687  | 43460 393 1012 3977
----------------------------------------------------------------
Avr:       311 528 1641             387 1008 3907
Tot:       349 768 2774



Long uses 4 bytes. Allocating 2*29360128 elements = 234881024 bytes of memory.
Getting down to business... Doing 10 runs per test.
0    Method: MEMCPY  Elapsed: 0.28276       MiB: 112.00000  Copy: 396.101 MiB/s
1    Method: MEMCPY  Elapsed: 0.28270       MiB: 112.00000  Copy: 396.177 MiB/s
2    Method: MEMCPY  Elapsed: 0.28286       MiB: 112.00000  Copy: 395.958 MiB/s
3    Method: MEMCPY  Elapsed: 0.28269       MiB: 112.00000  Copy: 396.198 MiB/s
4    Method: MEMCPY  Elapsed: 0.28273       MiB: 112.00000  Copy: 396.143 MiB/s
5    Method: MEMCPY  Elapsed: 0.28274       MiB: 112.00000  Copy: 396.129 MiB/s
6    Method: MEMCPY  Elapsed: 0.28276       MiB: 112.00000  Copy: 396.093 MiB/s
7    Method: MEMCPY  Elapsed: 0.28283       MiB: 112.00000  Copy: 395.995 MiB/s
8    Method: MEMCPY  Elapsed: 0.28268       MiB: 112.00000  Copy: 396.205 MiB/s
9    Method: MEMCPY  Elapsed: 0.28301       MiB: 112.00000  Copy: 395.753 MiB/s
AVG    Method: MEMCPY  Elapsed: 0.28277       MiB: 112.00000  Copy: 396.075 MiB/s
Long uses 4 bytes. Allocating 2*29360128 elements = 234881024 bytes of memory.
Getting down to business... Doing 10 runs per test.
0    Method: DUMB Elapsed: 0.15066       MiB: 112.00000  Copy: 743.376 MiB/s
1    Method: DUMB Elapsed: 0.14534       MiB: 112.00000  Copy: 770.586 MiB/s
2    Method: DUMB Elapsed: 0.15054       MiB: 112.00000  Copy: 744.013 MiB/s
3    Method: DUMB Elapsed: 0.14568       MiB: 112.00000  Copy: 768.798 MiB/s
4    Method: DUMB Elapsed: 0.15076       MiB: 112.00000  Copy: 742.922 MiB/s
5    Method: DUMB Elapsed: 0.14549       MiB: 112.00000  Copy: 769.828 MiB/s
6    Method: DUMB Elapsed: 0.15060       MiB: 112.00000  Copy: 743.682 MiB/s
7    Method: DUMB Elapsed: 0.14586       MiB: 112.00000  Copy: 767.833 MiB/s
8    Method: DUMB Elapsed: 0.15029       MiB: 112.00000  Copy: 745.211 MiB/s
9    Method: DUMB Elapsed: 0.14574       MiB: 112.00000  Copy: 768.502 MiB/s
AVG    Method: DUMB Elapsed: 0.14810       MiB: 112.00000  Copy: 756.263 MiB/s
Long uses 4 bytes. Allocating 2*29360128 elements = 234881024 bytes of memory.
Using 262144 bytes as blocks for memcpy block copy test.
Getting down to business... Doing 10 runs per test.
0    Method: MCBLOCK Elapsed: 0.15554       MiB: 112.00000  Copy: 720.095 MiB/s
1    Method: MCBLOCK Elapsed: 0.15603       MiB: 112.00000  Copy: 717.806 MiB/s
2    Method: MCBLOCK Elapsed: 0.15597       MiB: 112.00000  Copy: 718.073 MiB/s
3    Method: MCBLOCK Elapsed: 0.15585       MiB: 112.00000  Copy: 718.658 MiB/s
4    Method: MCBLOCK Elapsed: 0.15582       MiB: 112.00000  Copy: 718.792 MiB/s
5    Method: MCBLOCK Elapsed: 0.15573       MiB: 112.00000  Copy: 719.198 MiB/s
6    Method: MCBLOCK Elapsed: 0.15546       MiB: 112.00000  Copy: 720.429 MiB/s
7    Method: MCBLOCK Elapsed: 0.15553       MiB: 112.00000  Copy: 720.100 MiB/s
8    Method: MCBLOCK Elapsed: 0.15586       MiB: 112.00000  Copy: 718.603 MiB/s
9    Method: MCBLOCK Elapsed: 0.15605       MiB: 112.00000  Copy: 717.709 MiB/s
AVG    Method: MCBLOCK Elapsed: 0.15578       MiB: 112.00000  Copy: 718.945 MiB/s
root@OrangePI:~#

bronco · Published in 2015-11-18 23:55:04

loboris replied at 2015-11-18 23:02
i'm not an overclocking fan, but here are the tests (1.53 GHz constant during the test, no thermal ...

Great! Thx a lot. I guess, DRAM frequency was set to 672 MHz?

Do you have an idea why the 'out of memory errors' occur? I used the very same setup on a couple of SBC with default settings before without problems. If anyone's interested to compare with these SBC, the results can be found somewhere inside this review of another board:

Raspberry Pi 2, BroadCom BCM2836 quad core SoC, 1.0 GHz
Hardkernel ODROID-C1+, Amlogic S805 quad core SoC, 1.7 GHz
LeMaker Banana Pi, Allwinner A20 dual core SoC, 0.96 GHz
LeMaker Guitar, Actions Semi S500/ATM7059 quad core SoC, 1.1 GHz
Wandboard Quad, Freescale i.MX6 quad core SoC, 1.0 GHz

Thank you again, that gives a good impression what to expect when trying to run the H3 without a fan and conservative settings. Will be a bit faster compared to RPi 2 when it's about integer performance, a lot faster regarding I/O since the 'single USB 2.0 connection' bottleneck all RPi suffer from doesn't exist with the H3 (and will loose when it's about HW accelerated video encoding/decoding since I doubt we will ever see working CedarX stuff ion Linux). Perfect, I just ordered 1 Orange Pi PC to play with :-)

And thanks a lot for all your work to provide useable H3 OS images.

loboris · Published in 2015-11-19 01:21:05

bronco replied at 2015-11-18 16:55
Great! Thx a lot. I guess, DRAM frequency was set to 672 MHz?

Do you have an idea why the 'out of ...

DRAM frequency is 672. "Out of memory" is because oom killer kills one of the threads. It occurs also on i386 machines when memory is low.  I dont think the results are influenced by lower settings.

Here are the results on OPI-PLUS without the heatsink (frequency trottles to 1.2/1.008 GHz). I had to lower the memory settings because it was running with desktop active.

root@OrangePI:~# sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=4 && echo -e "\n\n" && 7za b -md22 && echo -e "\n\n" && mbw -t0 64;
mbw -t1 64 ; mbw -t2 64
sysbench 0.4.12:  multi-threaded system evaluation benchmark

Running the test with following options:
Number of threads: 4

Doing CPU performance benchmark

Threads started!
Done.

Maximum prime number checked in CPU test: 20000

Test execution summary:
total time:                         183.7352s
total number of events:             10000
total time taken by event execution: 734.7864
per-request statistics:
      min:                               49.00ms
      avg:                               73.48ms
      max:                               446.61ms
      approx.  95 percentile:          132.30ms

Threads fairness:
events (avg/stddev):          2500.0000/27.88
execution time (avg/stddev): 183.6966/0.02

7-Zip (A) 9.20  Copyright (c) 1999-2010 Igor Pavlov  2010-11-18
p7zip Version 9.20 (locale=hr_HR.UTF-8,Utf16=on,HugeFiles=on,4 CPUs)

RAM size: 1000 MB,  # CPU hardware threads: 4
RAM usage: 122 MB,  # Benchmark threads:    4

Dict       Compressing       |       Decompressing
   Speed Usage R/U Rating  | Speed Usage R/U Rating
   KB/s    % MIPS MIPS  |    KB/s    % MIPS MIPS

22: 1262 280 438 1228  | 29045 322 812 2620
----------------------------------------------------------------
Avr:       280 438 1228             322 812 2620
Tot:       301 625 1924

Long uses 4 bytes. Allocating 2*16777216 elements = 134217728 bytes of memory.
Getting down to business... Doing 10 runs per test.
0    Method: MEMCPY  Elapsed: 0.14754       MiB: 64.00000 Copy: 433.775 MiB/s
1    Method: MEMCPY  Elapsed: 0.15243       MiB: 64.00000 Copy: 419.857 MiB/s
2    Method: MEMCPY  Elapsed: 0.15263       MiB: 64.00000 Copy: 419.301 MiB/s
3    Method: MEMCPY  Elapsed: 0.15258       MiB: 64.00000 Copy: 419.438 MiB/s
4    Method: MEMCPY  Elapsed: 0.16614       MiB: 64.00000 Copy: 385.222 MiB/s
5    Method: MEMCPY  Elapsed: 0.17646       MiB: 64.00000 Copy: 362.684 MiB/s
6    Method: MEMCPY  Elapsed: 0.16827       MiB: 64.00000 Copy: 380.348 MiB/s
7    Method: MEMCPY  Elapsed: 0.15096       MiB: 64.00000 Copy: 423.959 MiB/s
8    Method: MEMCPY  Elapsed: 0.15009       MiB: 64.00000 Copy: 426.405 MiB/s
9    Method: MEMCPY  Elapsed: 0.15331       MiB: 64.00000 Copy: 417.466 MiB/s
AVG    Method: MEMCPY  Elapsed: 0.15704       MiB: 64.00000 Copy: 407.535 MiB/s
Long uses 4 bytes. Allocating 2*16777216 elements = 134217728 bytes of memory.
Getting down to business... Doing 10 runs per test.
0    Method: DUMB Elapsed: 0.08016       MiB: 64.00000 Copy: 798.453 MiB/s
1    Method: DUMB Elapsed: 0.07934       MiB: 64.00000 Copy: 806.604 MiB/s
2    Method: DUMB Elapsed: 0.08479       MiB: 64.00000 Copy: 754.842 MiB/s
3    Method: DUMB Elapsed: 0.08501       MiB: 64.00000 Copy: 752.835 MiB/s
4    Method: DUMB Elapsed: 0.08185       MiB: 64.00000 Copy: 781.937 MiB/s
5    Method: DUMB Elapsed: 0.08109       MiB: 64.00000 Copy: 789.276 MiB/s
6    Method: DUMB Elapsed: 0.09141       MiB: 64.00000 Copy: 700.112 MiB/s
7    Method: DUMB Elapsed: 0.08055       MiB: 64.00000 Copy: 794.567 MiB/s
8    Method: DUMB Elapsed: 0.08131       MiB: 64.00000 Copy: 787.140 MiB/s
9    Method: DUMB Elapsed: 0.08635       MiB: 64.00000 Copy: 741.152 MiB/s
AVG    Method: DUMB Elapsed: 0.08319       MiB: 64.00000 Copy: 769.367 MiB/s
Long uses 4 bytes. Allocating 2*16777216 elements = 134217728 bytes of memory.
Using 262144 bytes as blocks for memcpy block copy test.
Getting down to business... Doing 10 runs per test.
0    Method: MCBLOCK Elapsed: 0.09939       MiB: 64.00000 Copy: 643.928 MiB/s
1    Method: MCBLOCK Elapsed: 0.08989       MiB: 64.00000 Copy: 711.997 MiB/s
2    Method: MCBLOCK Elapsed: 0.09726       MiB: 64.00000 Copy: 658.003 MiB/s
3    Method: MCBLOCK Elapsed: 0.10718       MiB: 64.00000 Copy: 597.104 MiB/s
4    Method: MCBLOCK Elapsed: 0.09798       MiB: 64.00000 Copy: 653.201 MiB/s
5    Method: MCBLOCK Elapsed: 0.09871       MiB: 64.00000 Copy: 648.331 MiB/s
6    Method: MCBLOCK Elapsed: 0.10879       MiB: 64.00000 Copy: 588.311 MiB/s
7    Method: MCBLOCK Elapsed: 0.10049       MiB: 64.00000 Copy: 636.905 MiB/s
8    Method: MCBLOCK Elapsed: 0.10045       MiB: 64.00000 Copy: 637.108 MiB/s
9    Method: MCBLOCK Elapsed: 0.11052       MiB: 64.00000 Copy: 579.075 MiB/s
AVG    Method: MCBLOCK Elapsed: 0.10107       MiB: 64.00000 Copy: 633.245 MiB/s
root@OrangePI:~#