No performance difference between 1.5, 1.75 & 2GHz

Moderators: mdrjr, odroid

No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Tue Aug 09, 2016 10:15 pm

Hello Odroid lovers and makers!

While digging around for some information, I came across this link. It claimed the C2 would only be running at 1.536 GHz, since at the two faster modes benchmark results (they are compiling code) would be the same and no further increasing.
I did some tests with 7z and sysbench (primes up to 10000) myself and am unfortunately experiencing the same issue.
I confirmed this on two systems, updated to the latest official Ubuntu levels. Frequency was set with cpufreq-set -u $FREQ and queried with cpufreq-info.

Please find my results below:

Code: Select all
7z b, total MIPS
kHZ   MIPS   
100000   221   
250000   605   
500000   1252   
1000000   2488   
1296000   3182   
1536000   3705   
1752000   3691   
2016000   3682   
   
avg. of 3 x sysbench --num-threads=4 --test=cpu run
kHZ   seconds @4threads   seconds @1threads
100000   40   151
250000   15.3   61
500000   7.6   30
1000000   3.76   14.9
1296000   2.9   11.5
1536000   2.46   9.64
1752000   2.43   9.7
2016000   2.43   9.64


While the results scale almost linearly between 0.25, 0.5, 1 and 1.5GHz, they hit a wall there.
Regardless if run on 4 or just 1 threads.

Quickly scanning the forums here I couldn't find anything on the topic yet, just Exploring CPU clock speeds on the C2, trying to overclock to 2.2GHz with no visible effect, now making sense in that light...

Any idea whats going one here?
Are the devices really running at 2GHz? If so, what is that barrier at 1.5GHz, why is there no performance increase at the last two clock settings?
Running just one thread didn't benefit, I'd say the prime computation isn't bound by memory bandwidth, but what else?

I own 8 C2 myself, in our BOINC team we have dozens of them computing for science. Of course we are trying to maximize their performance. Usually these are actively cooled and could certainly run at a bit more voltage (if needed) and full throttle. If they could deliver some 20-30% more performance, by benefiting from the last two clock steps, that would be awesome...
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby majorowe » Wed Aug 10, 2016 12:06 am

Awaiting response with bated breath :o
majorowe
 
Posts: 33
Joined: Fri Jul 29, 2016 12:51 am
languages_spoken: english français deutsch espanol
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby crashoverride » Wed Aug 10, 2016 1:38 am

koschi wrote:Quickly scanning the forums here I couldn't find anything on the topic yet

This topic sounds related:
http://forum.odroid.com/viewtopic.php?f=138&t=22983

I think this issue warrants further investigation by HardKernel. It would be helpful if you could provide a script that reliably reproduces the issue.
crashoverride
 
Posts: 2974
Joined: Tue Dec 30, 2014 8:42 pm
languages_spoken: english
ODROIDs: C1

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Wed Aug 10, 2016 6:35 am

Thanks for the link crashoveride, I didn't check "Other OS" to be honest.
The problem seems related...

7z benchmark (package: p7zip)
Code: Select all
#!/bin/bash
for FREQ in 2016000 1752000 1536000 1296000 1000000 500000 250000 100000; do
   echo "Setting CPU frequency: cpufreq-set -u $FREQ, then sleeping 5s"
   cpufreq-set -u $FREQ
   sleep 5
   printf "CPU frequency (cpufreq-info): "; cpufreq-info | awk '/current CPU frequency is/ { print $5 }' | xargs
   7z b | egrep "Rating|MIPS|Tot:"
   7z b | egrep "Tot:"
   7z b | egrep "Tot:"
   echo
done


Sysbench (package: sysbench)
Code: Select all
#!/bin/bash
for FREQ in 2016000 1752000 1536000 1296000 1000000 500000 250000 100000; do
   echo "Setting CPU frequency: cpufreq-set -u $FREQ, then sleeping 5s"
   cpufreq-set -u $FREQ
   sleep 5
   printf "CPU frequency (cpufreq-info): "; cpufreq-info | awk '/current CPU frequency is/ { print $5 }' | xargs
   echo '3 x sysbench --num-threads=4 --test=cpu run | grep "total time:"'
   sysbench --num-threads=4 --test=cpu run | grep "total time:"
   sysbench --num-threads=4 --test=cpu run | grep "total time:"
   sysbench --num-threads=4 --test=cpu run | grep "total time:"
   echo
done
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Wed Aug 10, 2016 7:23 am

Could this be related to thermal throttling?
What happens if the benchmarks are short, so any thermal throttling doesn't start?
endecotp
 
Posts: 227
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Wed Aug 10, 2016 7:44 am

No this isn't related to throttling, the devices are actively cooled through a 12cm Noctua fan at 5v (usb), temperatures <70°C running several BOINC projects at consistent work unit durations.

Have a look at the sys bench runs, at high frequencies they take only few seconds, I had nothing else running in that moment, so the device should be cooled down... Running high frequencies first, with added throttling after some runs the performance should degrade, but it doesn't at 2, 1.75 and 1.5 GHz, just once we are below 1.5 GHz...
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby odroid » Wed Aug 10, 2016 12:28 pm

We will investigate this issue seriously.
User avatar
odroid
Site Admin
 
Posts: 23714
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Wed Aug 10, 2016 4:57 pm

Thank you!

I just noticed that (while running 400% Einstein@home CPU load) the CPU temperature between 2, 1.75 and 1.5GHz doesn't change at all (solid 65°C), while running at 1.3 (61°C), 1GHz (52°C) or 0.5GHz (41°C) it significantly decreases.
Freq. change to temperature measurement interval is 10 minutes, again the device is slightly active cooled.
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Thu Aug 11, 2016 12:15 am

On aarch64 (Arch ARM), I set the frequency with the systemd service in the cpupower-4.6-4 package and timed a call to `xz --threads 0 test.txt` which is a 36 meg file I generated by taking 250,000 lines out of a httpd access log file. I did this on a tmpfs (RAM disk) mounted partition so there is no micro SD card I/O to consider.

I varied the CPU frequency and ran the compression task a total of 3 times. On my system, the time to complete the compression task is absolutely correlated to the CPU frequency, albeit in a non-linear fashion:
Image

Data you can play with yourself:
Code: Select all
clock rate (MHz),time to compress (sec)
1000,62.72
1000,62.69
1000,62.8
2016,43.45
2016,43.55
2016,45.23
500,127.84
500,127.77
500,118.53
1536,45.22
1536,43.55
1536,45.18
250,229.47
250,234.33
250,230.62


Example:
Code: Select all
xz --threads 0 test  74.65s user 0.54s system 149% cpu 50.335 total
Last edited by graysky on Mon Aug 22, 2016 2:40 pm, edited 1 time in total.
User avatar
graysky
 
Posts: 111
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Thu Aug 11, 2016 12:31 am

So you are running Arch Linux?
Can you please also provide either sysbench or 7z results, to compare against the HK Ubuntu?
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Thu Aug 11, 2016 12:36 am

koschi wrote:So you are running Arch Linux?
Can you please also provide either sysbench or 7z results, to compare against the HK Ubuntu?


Yes, Arch. Sorry, I don't have the time to do that... you or someone else can easily repeat my xz experiment using a similarly sized text file. I do know that the most recent version of cpupower (4.7-1) does not change the CPU frequency of not only ARM processors but x86 processors as well. See: https://bugs.archlinux.org/task/50202

What version are the affected users running?

EDIT: I see you're using cpufreq-set not cpupower. I don't find an Arch package that provides cpufreq-set. Do you have cpupower on your distro? Does it work and what version?
User avatar
graysky
 
Posts: 111
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Thu Aug 11, 2016 4:42 am

The seems to be no cpupower in the ARM nor x86-64 Ubuntu.

edit:

now that I'm online with my home computer I can see your full data set (the mobile browser showed it only partly).

freq seconds
2016 43.45
2016 43.55
2016 45.23

looks about the same as
freq seconds
1536 45.22
1536 43.55
1536 45.18

You are saying "On my system, the time to complete the compression task is absolutely correlated to the CPU frequency, albeit in a non-linear fashion".
I see virtually no difference between your results taken at a 500MHz frequency difference.
*crystal ball on* From that I would conclude that Arch is equally affected and that its most likely not the CPU frequency management tool-set that is at fault, but the Kernel... *crystal ball off*
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Thu Aug 11, 2016 5:03 am

koschi wrote:The seems to be no cpupower in the ARM nor x86-64 Ubuntu.


I haven't used Debian/Ubuntu in a long time but perhaps linux-utils: http://packages.ubuntu.com/search?suite ... n=contents

koschi wrote:You are saying "On my system, the time to complete the compression task is absolutely correlated to the CPU frequency, albeit in a non-linear fashion".
I see virtually no difference between your results taken at a 500MHz frequency difference.
*crystal ball on* From that I would conclude that Arch is equally affected and that its most likely not the CPU frequency management tool-set that is at fault, but the Kernel... *crystal ball off*


Again, I can't explain why it's not linear, but it is highly correlated to clock rate. You can model my data with an R^2 of 0.9996 using a symmetrical sigmoidal model.
Code: Select all
1000,62.74
2016,44.08
500,124.71
1536,44.65
250,231.47


Fit parameters:
Code: Select all
y = 36.83126 + (329.5832 - 36.83126)/(1 + (x/341.0039)^2.206052)


EDIT: Ah! I took the title at face value and didn't inspect the OP's data. His too can be nearly perfectly modeled with the symmetrical sigmoidal fit. So I agree with your conclusion.

EDIT2: The next thing to do is to repeat one of these experiments on an x86 machine to see if it's the fault of the C2 or of the software driving it.

As an aside, https://mycurvefit.com/ is a great resource for simple plots and fits.
User avatar
graysky
 
Posts: 111
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Thu Aug 11, 2016 11:42 pm

Looking at drivers/amlogic/cpufreq/voltage.c, there is a table that relates frequency and supply voltage; its fastest entry is for 1.5 GHz. BUT this file doesn't actually seems to be used; the driver that is compiled is in cpufreq-scpi.c, which doesn't use voltage.h. I don't see anything related to voltage scaling in the code that's actually used.

Looking at the datasheet, e.g. figure III.18.1 on page 46, apparently any voltage scaling is the responsibility of an external power management IC. That figure shows the core voltage as being 0.81 ~ 1.2 V. I don't think the C2 has anything like that; looking at the schematic, it seems to use a fixed regulator (U14, on page 9), and I think the chosen resistors give an output of 1.05 V.

The datasheet also uses the term "overdrive" in places to refer to the 2 GHz clock speed and 1.2 V supply voltage.

HK, where did your power supply design come from? Did you get a reference design from Amlogic, or is this your own design?
endecotp
 
Posts: 227
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Fri Aug 12, 2016 4:30 am

From my post on: "Exploring CPU clock speeds on the C2" way back upon release I discovered the freq table is controlled via the ARM Trustzone controller
There is no hackable way of changing the voltage / freq table without modifying the binary blobs and signing them somehow. We "assume" that Amlogic
set them up properly and they are working.

The only issue I have with all these tests is you can see from my post on thermal tests for the C2 that when it thermally throttles the CPU cools off after the frequency
is dropped, then warms up again when it increases the frequency. Exactly what you would expect to happen.
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Fri Aug 12, 2016 5:47 am

endecotp wrote: I don't think the C2 has anything like that; looking at the schematic, it seems to use a fixed regulator (U14, on page 9), and I think the chosen resistors give an output of 1.05 V.


Ah, sorry, I see that I've overlooked something important: the regulator IS adjustable, using a PWM output from the processor. This is more complicated than I thought....
endecotp
 
Posts: 227
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Fri Aug 12, 2016 5:57 am

mlinuxguy wrote: We "assume" that Amlogic
set them up properly and they are working.


After seeing these statistics it seems perhaps not, of course it could be thermal because 2ghz is essentially overclocked like the XU3/4.

No amount of heatsink/fan will keep the XU4 from throttling in my experience, even on short maximum loads.
User avatar
rooted
 
Posts: 3652
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1
C1+
C2
XU3 Lite
XU4
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby meveric » Fri Aug 12, 2016 6:33 am

rooted wrote:No amount of heatsink/fan will keep the XU4 from throttling in my experience, even on short maximum loads.

There are some tests that indicate with the right cooling it does withstand bursts of up to 20 seconds before throttling https://www.youtube.com/watch?v=q7_ZCHpAngo
Donate to support my work on the ODROID GameStation Turbo Image for U2/U3 XU3/XU4 X2 X C1 as well as many other releases.
Check out the Games and Emulators section to find some of my work or check the files in my repository to find the software i build for ODROIDs.
If you want to add my repository to your image read my HOWTO integrate my repo into your image.
User avatar
meveric
 
Posts: 7416
Joined: Mon Feb 25, 2013 2:41 pm
languages_spoken: german, english
ODROIDs: ODROID-X2,ODROID-U2,ODROID-U3,ODROID-XU-Lite, ODROID-XU3, ODROID-XU3-Lite, ODROID-C1, ODROID-XU4, ODROID-C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Fri Aug 12, 2016 11:28 am

I only ran a few of the tests on a C2 with the stock cooler
Based on what I'm seeing from the temperatures I don't think its actually changing the CPU clock frequency. You would think the temps
would drop even under load by the time you drop to 1.5ghz.
I have another C2 with my custom giant heatsink I will re-run the tests on
Code: Select all
root@odroid64-2:~# ./7bench.sh
Setting CPU frequency: cpufreq-set -u 2016000, then sleeping 5s
CPU frequency (cpufreq-info): 2.02 2.02 2.02 2.02
      Speed Usage    R/U Rating  |    Speed Usage    R/U Rating
       KB/s     %   MIPS   MIPS  |     KB/s     %   MIPS   MIPS
Tot:          354   1019   3691
Tot:          356   1019   3732
Tot:          356   1019   3725

Setting CPU frequency: cpufreq-set -u 1752000, then sleeping 5s
CPU frequency (cpufreq-info): 1.75 1.75 1.75 1.75
      Speed Usage    R/U Rating  |    Speed Usage    R/U Rating
       KB/s     %   MIPS   MIPS  |     KB/s     %   MIPS   MIPS
Tot:          357   1020   3737
Tot:          358   1019   3742
Tot:          358   1019   3742

Setting CPU frequency: cpufreq-set -u 1536000, then sleeping 5s
CPU frequency (cpufreq-info): 1.54 1.54 1.54 1.54
      Speed Usage    R/U Rating  |    Speed Usage    R/U Rating
       KB/s     %   MIPS   MIPS  |     KB/s     %   MIPS   MIPS
Tot:          356   1019   3721


Temps during test
Code: Select all
root@odroid64-2:~# ./watchtemp.sh
time,temp,freq
0,68,2016000
5,66,2016000
10,67,2016000
15,67,2016000
20,69,2016000
25,68,2016000
30,67,2016000
35,67,2016000
40,68,2016000
45,69,2016000
50,69,2016000
55,70,2016000
60,69,2016000
65,63,1752000
70,63,1752000
75,69,1752000
80,67,1752000
85,69,1752000
90,70,1752000
95,67,1752000
100,69,1752000
105,70,1752000
110,71,1752000
115,67,1752000
120,69,1752000
125,69,1752000
130,70,1752000
135,71,1752000
140,71,1752000
145,71,1752000
150,72,1752000
155,68,1752000
160,69,1752000
165,70,1752000
170,72,1752000
175,69,1752000
180,70,1752000
185,72,1752000
190,72,1752000
195,67,1752000
200,70,1752000
205,70,1752000
210,71,1752000
215,72,1752000
220,72,1752000
225,71,1752000
230,73,1752000
235,72,1752000
240,73,1752000
245,71,1752000
250,72,1752000
255,73,1752000
260,72,1752000
265,72,1752000
270,73,1752000
275,74,1752000
280,71,1752000
285,72,1752000
290,72,1752000
295,73,1752000
300,74,1752000
305,74,1752000
310,75,1752000
315,75,1752000
320,63,1536000
325,72,1536000
330,72,1536000
335,73,1536000
340,75,1536000
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Fri Aug 12, 2016 11:54 am

Testing the big (slightly cut-down by 1 row) Northbridge heatsink
There is no throttling going on as this C2 is far below the thermal limits. The temps hang around a max of 55C even after I re-ran the
tests with the CPU still hot from previous. However the MIPS values do not change as the frequency does....

Last I knew the Amlogic kernel L2 cache code was not implemented. So the tests could be severely limited by the size of the L1 cache
Assuming the Freq/Volt tables are correct, the test code could be so large it doesn't fit in the L1 cache and since L2 isn't implemented we
are limited to pulling from ram.
Cache info for S905
Code: Select all
L1 cache  8KB-64KB I$ + 8KB-64KB D$
L2 cache  128KB - 2MB (Optional)

From the S905 PDF:
The main system CPU is a quad-core ARM Cortex-A53 CPU with 32KB L1 instruction
 and 32KB data cache for each core and a large 512KB L2 unified cache to improve system performance

It would be interesting to run some benchmark that fits in the 32kb L1 cache and see if its perf varies by clock speed

Code: Select all
root@odroid64-1:~# ./7bench.sh
Setting CPU frequency: cpufreq-set -u 2016000, then sleeping 5s
CPU frequency (cpufreq-info): 2.02 2.02 2.02 2.02
      Speed Usage    R/U Rating  |    Speed Usage    R/U Rating
       KB/s     %   MIPS   MIPS  |     KB/s     %   MIPS   MIPS
Tot:          351   1003   3619
Tot:          353   1000   3626
Tot:          351   1000   3612

Setting CPU frequency: cpufreq-set -u 1752000, then sleeping 5s
CPU frequency (cpufreq-info): 1.75 1.75 1.75 1.75
      Speed Usage    R/U Rating  |    Speed Usage    R/U Rating
       KB/s     %   MIPS   MIPS  |     KB/s     %   MIPS   MIPS
Tot:          351   1000   3608
Tot:          351   1001   3609
Tot:          353   1002   3623

Setting CPU frequency: cpufreq-set -u 1536000, then sleeping 5s
CPU frequency (cpufreq-info): 1.54 1.54 1.54 1.54
      Speed Usage    R/U Rating  |    Speed Usage    R/U Rating
       KB/s     %   MIPS   MIPS  |     KB/s     %   MIPS   MIPS
Tot:          354   1001   3631
Tot:          352   1001   3619
Tot:          353   1001   3623


Temperatures
Code: Select all
root@odroid64-1:~# ./watchtemp.sh
time,temp,freq
0,45,2016000
5,45,2016000
10,44,2016000
15,44,2016000
20,42,2016000
25,42,2016000
30,47,2016000
35,50,2016000
40,49,2016000
45,47,2016000
50,50,2016000
55,48,2016000
60,49,2016000
65,50,2016000
70,51,2016000
75,49,2016000
80,49,2016000
85,50,2016000
90,50,2016000
95,50,2016000
100,50,2016000
105,47,2016000
110,51,2016000
115,50,2016000
120,51,2016000
125,50,2016000
130,49,2016000
135,52,2016000
140,50,2016000
145,50,2016000
150,51,2016000
155,52,2016000
160,49,2016000
165,50,2016000
170,51,2016000
175,51,2016000
180,52,2016000
185,52,2016000
190,49,2016000
195,53,2016000
200,51,2016000
205,53,2016000
210,51,2016000
215,51,2016000
220,53,2016000
225,52,2016000
230,52,2016000
235,52,2016000
240,53,2016000
245,48,2016000
250,52,2016000
255,52,2016000
260,52,2016000
265,52,2016000
270,52,2016000
275,52,2016000
280,54,2016000
285,48,1752000
290,51,1752000
295,54,1752000
300,52,1752000
305,53,1752000
310,54,1752000
315,52,1752000
320,52,1752000
325,53,1752000
330,54,1752000
335,49,1752000
340,52,1752000
345,52,1752000
350,53,1752000
355,53,1752000
360,52,1752000
365,53,1752000
370,54,1752000
375,51,1752000
380,54,1752000
385,53,1752000
390,53,1752000
395,54,1752000
400,52,1752000
405,52,1752000
410,53,1752000
415,54,1752000
420,50,1752000
425,53,1752000
430,53,1752000
435,53,1752000
440,53,1752000
445,54,1752000
450,54,1752000
455,54,1752000
460,51,1752000
465,55,1752000
470,52,1752000
475,53,1752000
480,54,1752000
485,52,1752000
490,53,1752000
495,53,1752000
500,54,1752000
505,50,1752000
510,52,1752000
515,52,1752000
520,54,1752000
525,53,1752000
530,53,1752000
535,54,1752000
540,54,1752000
545,51,1752000
550,47,1536000
555,54,1536000
560,53,1536000
565,53,1536000
570,54,1536000
575,53,1536000
580,53,1536000
585,54,1536000
590,54,1536000
595,51,1536000
600,53,1536000
605,53,1536000
610,54,1536000
615,53,1536000
620,54,1536000
625,54,1536000
630,55,1536000
635,52,1536000
640,55,1536000
645,54,1536000
650,54,1536000
655,55,1536000
660,53,1536000
665,53,1536000
670,55,1536000
675,55,1536000
680,56,1536000
685,54,1536000
690,54,1536000
695,54,1536000
700,54,1536000
705,54,1536000
710,55,1536000
715,56,1536000
720,56,1536000
725,53,1536000
730,56,1536000
735,54,1536000
740,57,1536000
745,54,1536000
750,54,1536000
755,55,1536000
760,52,1536000
765,57,1536000
770,54,1536000
775,54,1536000
780,56,1536000
785,56,1536000
790,56,1536000
795,56,1536000
800,57,1536000
805,57,1536000
810,49,1296000
815,54,1296000
820,55,1296000
825,54,1296000
830,54,1296000
835,56,1296000
840,54,1296000
845,54,1296000
850,54,1296000
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Fri Aug 12, 2016 1:19 pm

I decided to run a test where I can control the size of the test program
ref: http://www.brendangregg.com/blog/2014-0 ... hmark.html
Test script
Code: Select all
root@odroid64-1:~# cat nopbench.sh
#!/bin/bash
for FREQ in 2016000 1752000 1536000 1296000 1000000 500000 250000 100000; do
   echo "Setting CPU frequency: cpufreq-set -u $FREQ, then sleeping 5s"
   cpufreq-set -u $FREQ
   sleep 5
   printf "CPU frequency (cpufreq-info): "; cpufreq-info | awk '/current CPU frequency is/ { print $5 }' | xargs
   time /root/noploop
   echo
done


For the noploop test program I followed his example and did 2000 nop instructions above this:
noploop.s
Code: Select all
    nop
    nop
    ble .L3    <---- 2,000 nop's added above this
    mov w0, 0
    add sp, sp, 16
    ret
    .size   main, .-main


The results show there is NO significant difference in execution speed for the top 3 frequencies
Code: Select all
root@odroid64-1:~# ./nopbench.sh
Setting CPU frequency: cpufreq-set -u 2016000, then sleeping 5s
CPU frequency (cpufreq-info): 2.02 2.02 2.02 2.02

real    0m6.568s
user    0m6.560s
sys     0m0.000s

Setting CPU frequency: cpufreq-set -u 1752000, then sleeping 5s
CPU frequency (cpufreq-info): 1.75 1.75 1.75 1.75

real    0m6.802s
user    0m6.770s
sys     0m0.000s

Setting CPU frequency: cpufreq-set -u 1536000, then sleeping 5s
CPU frequency (cpufreq-info): 1.54 1.54 1.54 1.54

real    0m6.569s
user    0m6.550s
sys     0m0.010s

Setting CPU frequency: cpufreq-set -u 1296000, then sleeping 5s
CPU frequency (cpufreq-info): 1.30 1.30 1.30 1.30

real    0m7.785s
user    0m7.780s
sys     0m0.000s

Setting CPU frequency: cpufreq-set -u 1000000, then sleeping 5s
CPU frequency (cpufreq-info): 1000 1000 1000 1000

real    0m10.092s
user    0m10.080s
sys     0m0.000s

Setting CPU frequency: cpufreq-set -u 500000, then sleeping 5s
CPU frequency (cpufreq-info): 500 500 500 500

real    0m20.222s
user    0m20.200s
sys     0m0.000s

Setting CPU frequency: cpufreq-set -u 250000, then sleeping 5s
CPU frequency (cpufreq-info): 250 250 250 250

real    0m41.099s
user    0m41.010s
sys     0m0.010s

Setting CPU frequency: cpufreq-set -u 100000, then sleeping 5s
CPU frequency (cpufreq-info): 100.0 100.0 100.0 100.0

real    1m42.968s
user    1m42.610s
sys     0m0.030s

I don't think the frequencies are really varying for the top 3 settings .....
Notes: all my tests were run on a HK distributed kernel for Ubuntu
root@odroid64-1:~# uname -r
3.14.65-73
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby odroid » Fri Aug 12, 2016 3:47 pm

We could reproduce the issue and reported what we observed to Amlogic.
I guess it can be a silicon design issue in worst case. :(
User avatar
odroid
Site Admin
 
Posts: 23714
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Fri Aug 12, 2016 4:01 pm

odroid wrote:We could reproduce the issue and reported what we observed to Amlogic.
I guess it can be a silicon design issue in worst case. :(

Hope not, you have reason to think it may be hardware?
User avatar
rooted
 
Posts: 3652
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1
C1+
C2
XU3 Lite
XU4
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Fri Aug 12, 2016 4:18 pm

I found an interesting result from my above nop test
If you follow the reference link he talks about computing Ghz based off of instruction count / time

Calculations
I got: user 0m6.560s
20 billion NOPs (10 million x 2000)

20 / 6.56 = 3.05 GHz
A53 is dual-issue so:
3.05 / 2 = 1.5244 GHz

Conclusion:
We are stuck at 1.5ghz, higher clock speeds are merely "Computer generated engine noises piped in to make you think your car is fast"
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Fri Aug 12, 2016 5:48 pm

Still the CPU performs well to be stuck at 1.5ghz, downside...how hot this would get at 2ghz. It must be intentionally locked at a maximum of 1.5ghz due to overheating.
User avatar
rooted
 
Posts: 3652
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1
C1+
C2
XU3 Lite
XU4
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby koschi » Fri Aug 12, 2016 7:46 pm

In that case it should be marketed at 1.5GHz, not 2GHz...
Most SBC don't even come with a heat sink and throttle right away. The C2 has a heat sink, runs perfectly fine under full load at old 2GHz / new 1.5GHz. Now if it would offer a turbo mode to 2GHz that would kick in at load and being under 70"C, it would be up to the customer to decide for a fan, accept that extra bit of noise and enjoy the extra power offered by 2GHz.
Of course the performance of the C2 is good, but just imagine where it would be at the marketed 2GHz?
With only few direct comparisons between the various SBC on the market available, one (at least those primarily interested in processing power) decides looking at the CPU architecture and its clock rate.
Then 2GHz A53 looks much stronger than 1.2GHz A53. At 1.5GHz it's not such a big difference, especially against a Pi3 that can be overclocked...

As Willy Tarreau points out in his article, the C2 was the same speed as a Samsung S5P6818 (8x1.4GHz A53) based competitor at 4 threads, now available for 35$.
If the C2 can't deliver the advertised GHz and hence higher performance, it simply looses some of its attractiveness, which would be a pity...
koschi
 
Posts: 15
Joined: Tue Mar 22, 2016 4:32 am
languages_spoken: english
ODROIDs: 8 x C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Pepes » Fri Aug 12, 2016 11:57 pm

Samsung S5P6818 won't have mainline support and its almost w/o support - check this link
Pepes
 
Posts: 97
Joined: Mon Feb 15, 2016 5:28 am
Location: Praha | Czech republic
languages_spoken: Czech | English
ODROIDs: Odroid C2 (2 dvb-c tuner & tvheadend)
Odroid U2
-->both without eMMC (overpriced)<--

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Raybuntu » Sat Aug 13, 2016 3:09 am

I wonder what happens if we try a different kernel. There are some users on this forum that use KH's 4.7 kernel with serial console. Maybe it's a kernel issue after all.
BTC: 13M7sUBpS9uVpo4mzukV9CH8HGg7vDUC7Y
Raybuntu
 
Posts: 1100
Joined: Mon Nov 30, 2015 4:23 pm
languages_spoken: english, german
ODROIDs: C1+, C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby brad » Sun Aug 14, 2016 3:55 pm

I have been attempting to get the userspace driver working on 4.7 kernel but with no luck as yet. Anyway here is the nopbench test (with 2000 nop's)...

Code: Select all
root@odroid64:~# uname -a
Linux odroid64 4.7.0-next-20160805 #5 SMP PREEMPT Sun Aug 14 02:32:18 EDT 2016 aarch64 aarch64 aarch64 GNU/Linux
root@odroid64:~# ./nopbench.sh
Setting CPU frequency: cpufreq-set -u 2016000, then sleeping 5s
Error setting new values. Common errors:
- Do you have proper administration rights? (super-user?)
- Is the governor you requested available and modprobed?
- Trying to set an invalid policy?
- Trying to set a specific frequency, but userspace governor is not available,
   for example because of hardware which cannot be set to a specific frequency
   or because the userspace governor isn't loaded?
CPU frequency (cpufreq-info):

real   0m6.569s
user   0m6.564s
sys   0m0.000s


Interestingly lmbench detects the frequency as 1535 MHz, 0.6515 nanosec clock but not sure it is acurate.....

Calculating mhz, please wait for a moment...
I think your CPU mhz is

1535 MHz, 0.6515 nanosec clock

but I am frequently wrong. If that is the wrong Mhz, type in your
best guess as to your processor speed. It doesn't have to be exact,
but if you know it is around 800, say 800.

Please note that some processors, such as the P4, have a core which
is double-clocked, so on those processors the reported clock speed
will be roughly double the advertised clock rate. For example, a
1.8GHz P4 may be reported as a 3592MHz processor.

Processor mhz [default 1535 MHz, 0.6515 nanosec clock]:


Edit: look as though the user-space governor is not yet working in 4.7, but patches are on the way starting with the SCPI proptcol drivers. Sorry Edited 2nd time: Sorry wrong link here is the correct one https://patchwork.kernel.org/patch/9270825/
Last edited by brad on Mon Aug 15, 2016 8:09 pm, edited 1 time in total.
brad
 
Posts: 526
Joined: Tue Mar 29, 2016 1:22 pm
Location: Australia
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Mon Aug 15, 2016 11:33 am

I'm fairly confident the problem is not in the kernel
Based on how ARM trustzone has its own volt/freq table, and how Amlogic had to "fix" it back when we had the thermal issues you would think
that they have recently double-checked the tables.
So that implies that what they thought was getting set for the frequency (those greater than 1.5ghz):
[1] is being ignored by the hardware
[2] Is hitting a coded freq limit of 1.5ghz that someone left in the trustzone secure code (if we're lucky)
[3] they have buggy PWM code that is not setting up the clock properly on the higher frequencies
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby graysky » Tue Aug 16, 2016 3:30 am

graysky wrote:The next thing to do is to repeat one of these experiments on an x86 machine to see if it's the fault of the C2 or of the software driving it.


As the positive control, I repeated on an x86_64 machine and as expected, the performance scales in a linear fashion with the frequency. Data below. Good work by folks here to formulate an underlying hypothesis as to why the C2 seems choked around 1.5 GHz. Hopefully, it's a software fix.

Image

Code: Select all
freq(Hz),time(sec)
2395000,24.025
2395000,24.123
2395000,23.948
2261000,27.355
2261000,27.528
2261000,27.341
1995000,30.547
1995000,30.205
1995000,30.526
1729000,34.650
1729000,34.104
1729000,33.947
1596000,36.364
1596000,36.449
1596000,37.303
2394000,26.114
2394000,26.266
2394000,25.913
2128000,28.672
2128000,28.968
2128000,29.332
1995000,30.588
1995000,30.125
1995000,30.293
1862000,31.894
1862000,31.956
1862000,32.248
User avatar
graysky
 
Posts: 111
Joined: Wed Jul 13, 2016 9:11 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby brad » Tue Aug 16, 2016 10:05 pm

mlinuxguy wrote:I'm fairly confident the problem is not in the kernel


I was able to compile and perform a Dhrystone test in u-boot with similar results, Im hoping the issue can be patched in the trusted firmware or the frequency can be adjusted at boot :?:

Code: Select all
=> help dhry
dhry - [iterations] - run dhrystone benchmark

Usage:
dhry
    - run the Dhrystone 2.1 benchmark, a rough measure of CPU speed

=> dhry 100000000
100000000 iterations in 36257 ms: 2758088/s, 1569 DMIPS
brad
 
Posts: 526
Joined: Tue Mar 29, 2016 1:22 pm
Location: Australia
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Wed Aug 17, 2016 6:36 am

mlinuxguy wrote:ARM trustzone has its own volt/freq table


Can anyone point to any documentation anywhere about that?
endecotp
 
Posts: 227
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby seandepagnier » Fri Aug 19, 2016 10:20 pm

any update to this? I am seeing no improvements from 1.5 to 2ghz, and lmbench detects 1.5.
seandepagnier
 
Posts: 16
Joined: Thu Aug 18, 2016 5:11 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby luxmile » Mon Aug 22, 2016 1:07 pm

Looking forward to this issue before making any decision... :)
luxmile
 
Posts: 33
Joined: Wed Feb 25, 2015 10:52 am
languages_spoken: english

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby odroid » Mon Aug 22, 2016 2:10 pm

We are still having some discussion with Amlogic.
There are a few things we can try. But the possibility and feasibility are not clear yet.
Give us one more week and we will let you know what we are trying now.
User avatar
odroid
Site Admin
 
Posts: 23714
Joined: Fri Feb 22, 2013 11:14 pm
languages_spoken: English
ODROIDs: ODROID

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Snk » Tue Aug 23, 2016 7:51 am

odroid wrote:We are still having some discussion with Amlogic.
There are a few things we can try. But the possibility and feasibility are not clear yet.
Give us one more week and we will let you know what we are trying now.


It is known whether the chip is "false" or is it something related to the device software?
Well this complicated story, you buy a product with 2Ghz and 1.5Ghz actually be.
I would even sue the Amlogic for false advertising.
User avatar
Snk
 
Posts: 195
Joined: Sun Jul 31, 2016 6:43 am
languages_spoken: Portuguese
ODROIDs: XU4 + eMMC 32GB + UART

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Tue Aug 23, 2016 8:27 am

The performance of the C2 is impressive, especially running at 1.5ghz.

Let's not get all worked up, even if the device is stuck at 1.5ghz. (I believe it is, couldn't pass thermal test)

After all look at the Pi 3, it's claimed to be 1.2ghz but in real use that's false because of overheating. The amlogic SoC being 2.0ghz isn't a lie, I believe it just isn't usable without a fan.
User avatar
rooted
 
Posts: 3652
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1
C1+
C2
XU3 Lite
XU4
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby TanteJu » Thu Aug 25, 2016 4:21 am

rooted wrote:Let's not get all worked up, even if the device is stuck at 1.5ghz.

You are not a manager at VW, are you?

Image
User avatar
TanteJu
 
Posts: 18
Joined: Fri Jul 29, 2016 5:32 pm
languages_spoken: german english french russian
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Thu Aug 25, 2016 5:11 am

TanteJu wrote:
rooted wrote:Let's not get all worked up, even if the device is stuck at 1.5ghz.
You are not a manager at VW, are you?

Lol, no but I wish I were.

It's not HK that wouldn't be telling the facts, they believed the SoC had 2ghz of usable processing power. This is amlogic, perhaps they made a mistake in configuration which will be corrected in software.

It should be up to the end user if they want to use a fan if it's held back intentionally due to overheating.
User avatar
rooted
 
Posts: 3652
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1
C1+
C2
XU3 Lite
XU4
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby TanteJu » Thu Aug 25, 2016 4:29 pm

rooted wrote:It's not HK that wouldn't be telling the facts, they believed the SoC had 2ghz of usable processing power. This is amlogic, perhaps they made a mistake ...


Again sound like VW to me ("It's not our fault, it's Bosch [subcontractor] fault").

rooted wrote:It should be up to the end user if they want to use a fan if it's held back intentionally due to overheating.


If I buy a 400HP Ferrari and then get told "It actually has only 200HP, but if you want 400, just install a Nitrous Oxide injection system" I'd be pretty pissed off.
User avatar
TanteJu
 
Posts: 18
Joined: Fri Jul 29, 2016 5:32 pm
languages_spoken: german english french russian
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby rooted » Thu Aug 25, 2016 7:15 pm

I'm not going to debate with you, everyone is entitled to an opinion.

Exactly what do you expect HK should do about it? You want your money back?

Amlogic sells the S905 based SoC as 2ghz, this is in no way the fault of HK .
User avatar
rooted
 
Posts: 3652
Joined: Fri Dec 19, 2014 9:12 am
Location: Gulf of Mexico, US
languages_spoken: english
ODROIDs: C1
C1+
C2
XU3 Lite
XU4
VU7+
HiFi Shield 2
Smart Power (original)

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Bl4ckD0g » Thu Aug 25, 2016 8:45 pm

I don't know about you all, but I'm eager to get more (much more) details about it.
For sure it doesn't smell good. :roll:
Bl4ckD0g
 
Posts: 48
Joined: Sat Apr 09, 2016 3:18 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby andy5macht » Thu Aug 25, 2016 10:17 pm

you just have to log UART while bootup to see its actually CPU-Freq.:
Code: Select all
BL2 Built : 11:44:26, Nov 25 2015.
gxb gfb13a3b-c2 - jcao@wonton

Board ID = 8
set vcck to 1100 mv
set vddee to 1050 mv
CPU clk: 1536MHz
DDR channel setting: DDR0 Rank0+1 same
DDR0: 2048MB(auto) @ 912MHz(2T)-13
DataBus test pass!
AddrBus test pass!
Load fip header from SD, src: 0x0000c200, des: 0x01400000, size: 0x000000b0
Load bl30 from SD, src: 0x00010200, des: 0x01000000, size: 0x00009ef0
Sending bl30........................................OK.
Run bl30...
Load bl301 from SD, src: 0x0001c200, des: 0x01000000, size: 0x00001ae0
Wait bl30...Done
Sending bl301.......OK.
Run bl301...
c: 0x00020200, des: 0x10100000, size: 0x00011130

i would really like to knwo why this isnt running at 2Ghz...
andy5macht
 
Posts: 168
Joined: Mon May 16, 2016 5:25 am
languages_spoken: english, german
ODROIDs: U3, C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Bl4ckD0g » Thu Aug 25, 2016 10:41 pm

And why the OS/kernel reports 2016MHz...
Bl4ckD0g
 
Posts: 48
Joined: Sat Apr 09, 2016 3:18 am
languages_spoken: english
ODROIDs: C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby mlinuxguy » Fri Aug 26, 2016 1:30 am

Bl4ckD0g wrote:And why the OS/kernel reports 2016MHz...

I can tell you why.... if you refer to my attempts to overclock the C2 I found out that no matter what I set the clock frequencies to on the linux side
(i.e. 2.3ghz) linux would report it was running at that frequency.... The full thread has the details: viewtopic.php?f=139&t=18738#p125407

At the time I wrote that I didn't know the volt/freq table was controlled by the M3 embedded controller and set by the manufacturer. Given it took me a while to figure out I was really not achieving those clock frequencies that I thought I had set.... I suspect Amlogic just assumed all was working properly at the higher clocks than 1.5ghz.
mlinuxguy
 
Posts: 639
Joined: Thu Feb 28, 2013 10:28 am
languages_spoken: english
ODROIDs: X, X2, XU, XU3, XU4, C1, C1+, C2, USB-IO

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Fri Aug 26, 2016 1:37 am

rooted wrote:Exactly what do you expect HK should do about it?


At this stage, I'm surprised that their web page still prominently says "ARM 64bit 2Ghz quad core single board computer!".
endecotp
 
Posts: 227
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby meveric » Fri Aug 26, 2016 1:43 am

Cause that is what AmLogic sold their SoCs as and as long as AmLogic doesn't say otherwise, it's considered still to be a 2GHz just, with an issue.

If due to a firmware update a iPhone 6 Plus suddenly does only show 512MB RAM instead of 1GB, do you think they would change their product page and say it's only a 512 MB phone?
Donate to support my work on the ODROID GameStation Turbo Image for U2/U3 XU3/XU4 X2 X C1 as well as many other releases.
Check out the Games and Emulators section to find some of my work or check the files in my repository to find the software i build for ODROIDs.
If you want to add my repository to your image read my HOWTO integrate my repo into your image.
User avatar
meveric
 
Posts: 7416
Joined: Mon Feb 25, 2013 2:41 pm
languages_spoken: german, english
ODROIDs: ODROID-X2,ODROID-U2,ODROID-U3,ODROID-XU-Lite, ODROID-XU3, ODROID-XU3-Lite, ODROID-C1, ODROID-XU4, ODROID-C2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby endecotp » Fri Aug 26, 2016 2:22 am

meveric wrote:If due to a firmware update a iPhone 6 Plus suddenly does only show 512MB RAM instead of 1GB, do you think they would change their product page and say it's only a 512 MB phone?


I don't think Apple have ever said publicly how much RAM an iPhone has (see e.g. http://www.apple.com/uk/iphone-6/specs/ ). But ignoring that detail, I think that if Apple did something like this then they would quickly find themselves subject to a class-action lawsuit. This is exactly what happened when the iOS 7 to iOS 8 upgrade reduced the amount of Flash storage available by about 25%: https://www.theguardian.com/technology/ ... d-capacity .

Passing the blame to a supplier isn't really acceptable, legally.
endecotp
 
Posts: 227
Joined: Sun May 08, 2016 12:30 am
languages_spoken: english
ODROIDs: c2

Re: No performance difference between 1.5, 1.75 & 2GHz

Unread postby Bl4ckD0g » Fri Aug 26, 2016 2:23 am

mlinuxguy wrote:
Bl4ckD0g wrote:And why the OS/kernel reports 2016MHz...

I can tell you why.... if you refer to my attempts to overclock the C2 I found out that no matter what I set the clock frequencies to on the linux side
(i.e. 2.3ghz) linux would report it was running at that frequency.... The full thread has the details: viewtopic.php?f=139&t=18738#p125407

At the time I wrote that I didn't know the volt/freq table was controlled by the M3 embedded controller and set by the manufacturer. Given it took me a while to figure out I was really not achieving those clock frequencies that I thought I had set.... I suspect Amlogic just assumed all was working properly at the higher clocks than 1.5ghz.


Thanks for the inputs!
So you already had the "issue" at that time.
Hopefully there'll be a software solution for this...
Bl4ckD0g
 
Posts: 48
Joined: Sat Apr 09, 2016 3:18 am
languages_spoken: english
ODROIDs: C2

Next

Return to Issues

Who is online

Users browsing this forum: No registered users and 1 guest