Boinc - to help science
#11
(04-20-2016, 04:52 PM)Keex Wrote:
(04-20-2016, 03:52 PM)kermitas Wrote: Can somebody pleeeeease post Pine's CPU benchmark made by BOINC app? (boincmgr -> advanced -> Run CPU benchmarks and then advanced -> event log)

Can't wait to see how many whetstones and dhrystones there are Smile.
Thanks!

Did it:
#1st run:
Benchmark results:
Number of CPUs: 4
1013 floating point MIPS (Whetstone) per CPU
3583 integer MIPS (Dhrystone) per CPU

#2nd run:
Benchmark results:
Number of CPUs: 4
1015 floating point MIPS (Whetstone) per CPU
3593 integer MIPS (Dhrystone) per CPU

Ran it on Ubuntu using the 2GB Pine64+. Most surprisingly I got to use a program in Linux that just installed and had a user-friendly GUI that just worked without messing with config files. I'm shocked.

Cheers.

Thanks Keex Smile. Now if you find app easy and useful you can easily donate your Pine's CPU power Smile).
Check my blog post where I mentioned you.

Interesting finding: if I would have 4 Pines then I would have as much dhrystones as ... Pentium Core 2 Q9450 (see attached chart).


Attached Files Thumbnail(s)
   
  Reply
#12
(04-21-2016, 02:44 AM)kermitas Wrote: Thanks Keex Smile. Now if you find app easy and useful you can easily donate your Pine's CPU power Smile).
Check my blog post where I mentioned you.

Interesting finding: if I would have 4 Pines then I would have as much dhrystones as ... Pentium Core 2 Q9450 (attachment attached).

Not the worst idea Wink I will check it out
  Reply
#13
Well I tried, but the hackless-experience of course didn't last long and I got something about an unknown platform which led me to vague forums talking about recompiling stuff and so on. Can't have an application on Linux without it, it seems.

If you get around to it and post a clear guide I will happily follow along ;-)
  Reply
#14
(04-05-2016, 07:59 AM)Andrew2 Wrote:
(04-05-2016, 07:42 AM)pine.tree Wrote: I'm guessing Boinc is very resource-intensive?

Sure, on Pine64 and on any other ARMv8 board it's important compile the programs since some algorithms are 20 times slower when not being compiled for ARMv8:

Code:
   export CXXFLAGS="-O3 -march=armv8-a -mtune=cortex-a53"
   export CFLAGS="-O3 -march=armv8-a -mtune=cortex-a53"

Then the board gets as hot as with optimised settings but is still slow as hell. Have a look at the commonly used sysbench: http://forum.odroid.com/viewtopic.php?f=136&t=19158

Hi Andrew,
are you talking about compilation of BOINC app?

I am asking because I wonder if this app is doing any calculations by itself?

I have a feeling that it downloads ready apps and spawns separate processes for them.

Here are some of "ps aux" from my PC machines:
Code:
boinc      982  0.0  0.0 118988  2792 ?        SNs  Apr20   0:00 /bin/bash /usr/bin/boinc --allow_remote_gui_rpc --dir /var/lib/boinc
boinc      987  0.0  0.6 502312 27384 ?        SNl  Apr20   0:54 /usr/bin/boinc_client --allow_multiple_clients --allow_remote_gui_rpc --dir /var/lib/boinc
boinc     2795 98.2  1.8 198376 75592 ?        RNl  12:23 288:21 ../../projects/www.worldcommunitygrid.org/wcgrid_fahb_bedam_7.15_x86_64-pc-linux-gnu -seed 304808031 -trickle 1 -upload 1
boinc     2867 99.8  0.9  77492 37516 ?        RNl  13:23 232:45 ../../projects/www.worldcommunitygrid.org/wcgrid_mcm1_7.36_x86_64-pc-linux-gnu -SettingsFile MCM1_0122506_1960.txt -DatabaseFile dataset-curatedOvarian_EarlyLate_v1.0.txt
boinc     3012  0.0  0.1  19152  5336 ?        SNl  15:33   0:02 ../../projects/wuprop.boinc-af.org/data_collect_v4_419_x86_64-pc-linux-gnu__nci -n 360 -c 60 -g 14
boinc     3031 99.7  0.9  77224 37440 ?        RNl  15:43  92:25 ../../projects/www.worldcommunitygrid.org/wcgrid_mcm1_7.36_x86_64-pc-linux-gnu -SettingsFile MCM1_0122506_2721.txt -DatabaseFile dataset-curatedOvarian_EarlyLate_v1.0.txt
boinc     3197 99.5  0.9  76960 37252 ?        RNl  16:55  20:35 ../../projects/www.worldcommunitygrid.org/wcgrid_mcm1_7.36_x86_64-pc-linux-gnu -SettingsFile MCM1_0122507_4519.txt -DatabaseFile dataset-curatedOvarian_EarlyLate_v1.0.txt
  
Code:
boinc      902  0.0  0.2   5588  2848 ?        SNs  11:24   0:00 /bin/bash /usr/bin/boinc --allow_remote_gui_rpc --dir /var/lib/boinc
boinc      910  0.2  1.6  66356 20916 ?        SNl  11:24   0:51 /usr/bin/boinc_client --allow_multiple_clients --allow_remote_gui_rpc --dir /var/lib/boinc
boinc     1176 94.1  5.8 133204 74296 ?        RNl  11:25 331:50 ../../projects/www.worldcommunitygrid.org/wcgrid_fahb_bedam_7.15_i686-pc-linux-gnu -seed 1905709651 -trickle 1 -upload 1
boinc     1711 93.7  1.4  39436 18452 ?        RNl  12:59 242:41 ../../projects/www.worldcommunitygrid.org/wcgrid_ugm1_7.28_i686-pc-linux-gnu -E 0.000000001 -Z 10000 -S -Q -m 9c ugm1_ugm1_24999_a_0018.txt ugm1_ugm1_24999_b_0027.txt
boinc     1912  0.1  0.3   5576  4328 ?        SNl  15:41   0:07 ../../projects/wuprop.boinc-af.org/data_collect_v4_419_i686-pc-linux-gnu__nci -n 360 -c 60 -g 14

And here from MK809III (ARMv7-A):
Code:
boinc      637  0.3  0.3  14568  7084 ?        SN   Apr20   4:12 /usr/bin/boinc --check_all_logins --redirectio --dir /var/lib/boinc-client --allow_remote_gui_rpc
boinc     2953 98.8  4.3  95216 91824 ?        RNl  16:59  22:19 ../../projects/einstein.phys.uwm.edu/einsteinbinary_BRP4_1.06_arm-unknown-linux-gnueabihf -i p2030.20151006.G46.15+01.55.N.b2s0g0.00000_1254.bin4 -t stochastic_full.bank -l p2030.20151006.G46.15+01.55.N.b2s0g0.00000.zap -o results.cand0 -c p2030.20151006.G46.15+01.55.N.b2s0g0.00000_1254.cpt -A 0.08 -P 3.0 -f 400.0 -W -z
boinc     2956 98.4  4.3  95216 91824 ?        RNl  16:59  22:10 ../../projects/einstein.phys.uwm.edu/einsteinbinary_BRP4_1.06_arm-unknown-linux-gnueabihf -i p2030.20151006.G46.15+01.55.N.b2s0g0.00000_124.bin4 -t stochastic_full.bank -l p2030.20151006.G46.15+01.55.N.b2s0g0.00000.zap -o results.cand0 -c p2030.20151006.G46.15+01.55.N.b2s0g0.00000_124.cpt -A 0.08 -P 3.0 -f 400.0 -W -z
boinc     2960 98.4  4.3  95216 91824 ?        RNl  16:59  22:04 ../../projects/einstein.phys.uwm.edu/einsteinbinary_BRP4_1.06_arm-unknown-linux-gnueabihf -i p2030.20151006.G46.15+01.55.N.b2s0g0.00000_1252.bin4 -t stochastic_full.bank -l p2030.20151006.G46.15+01.55.N.b2s0g0.00000.zap -o results.cand0 -c p2030.20151006.G46.15+01.55.N.b2s0g0.00000_1252.cpt -A 0.08 -P 3.0 -f 400.0 -W -z
boinc     2962 98.9  0.0   3088  1348 ?        RNl  16:59  22:08 ../../projects/universeathome.pl_universe/universe-BHspin_10_arm-unknown-linux-gnueabihf
boinc     2969  0.2  0.1   4180  2576 ?        SNl  17:08   0:02 ../../projects/wuprop.boinc-af.org/data_collect_v4_4.19_arm-unknown-linux-gnueabihf__nci -n 360 -c 60 -g 14

By looking at them - do you still think that dedication compilation will give any boost?
  Reply
#15
I'll give this a try - I've been running BOINC for the POEM project on my main PC.

Note that you can limit percentage of CPU to use.
  Reply
#16
Sat 23 Apr 2016 10:20:20 PM UTC | | Benchmark results:
Sat 23 Apr 2016 10:20:20 PM UTC | | Number of CPUs: 4
Sat 23 Apr 2016 10:20:20 PM UTC | | 1005 floating point MIPS (Whetstone) per CPU
Sat 23 Apr 2016 10:20:20 PM UTC | | 3506 integer MIPS (Dhrystone) per CPU


With cpus running at 1.15GHz, with some temporary drops to 1.10GHz.

The question is which projects provide arm64 binaries for they work.

If you are using Debian or Ubuntu based distro, and want to join SETI@Home, just install boinc-app-seti package. It will make boinc-client use it automatically

pine64user@debianpine64:/var/lib/boinc-client$ ps aux | grep seti
boinc 23364 97.8 3.5 89668 72972 ? RNl 22:29 2:59 ../../projects/setiathome.berkeley.edu/setiathome_v8

pine64user@debianpine64:/var/lib/boinc-client$ file projects/setiathome.berkeley.edu/setiathome_v8
projects/setiathome.berkeley.edu/setiathome_v8: symbolic link to /usr/lib/boinc-app-seti/setiathome_v8

$ file -L /var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_v8
/var/lib/boinc-client/projects/setiathome.berkeley.edu/setiathome_v8: ELF 64-bit LSB shared object, ARM aarch64, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux-aarch64.so.1, for GNU/Linux 3.7.0, BuildID[sha1]=7ff21f252684c89976528b28922eca4f79b208e2, stripped

It looks it will finish my work unit in about 36 hours. I know RPI2 is doing about 77h per WU.

Still this is only at one core, with 1.15GHz. Will see how things heat up with 4 WUs in progress and 4 cores busy. I do have small heatsink, but it might become a bit warm.

It looks every instance of setiathome_v8 uses about 80MB of RES memory. The application itself is small (3MB shared memory).
  Reply
#17
Thanks baryluk.

I am attaching      power efficiency charts of my devices. Here is a link to the spreadsheet if anybody is interested in.

Notes:
  • my gauge (measurer) works from 5W and for many hours it shows 0Wh for MK809III so I assumed that MK809III is consuming 4W
  • I don't have Pine64 yet so I assumed that it needs 5W to work
  • Android app (on my Samsung Galaxy S4) does not have an option to do CPU benchmark (or at least I didn't find it) so I put zeroes
Last chart shows power efficiency (how much millions of whetstones|dhrystones is delivered in one kW). As you can see, for now Pine64 is winning.

Interesting finding: monthly cost of power for old Pentium4 and Pentium Core-2 machines is equal to cost of one, brand new Pine64+ 2GB.
  Reply
#18
(04-24-2016, 01:21 AM)kermitas Wrote: Thanks baryluk.

I am attaching  power efficiency charts of my devices. Here is a link to the spreadsheet if anybody is interested in.

Notes:
  • my gauge (measurer) works from 5W and for many hours it shows 0Wh for MK809III so I assumed that MK809III is consuming 4W
  • I don't have Pine64 yet so I assumed that it needs 5W to work
  • Android app (on my Samsung Galaxy S4) does not have an option to do CPU benchmark (or at least I didn't find it) so I put zeroes
Last chart shows power efficiency (how much millions of whetstones|dhrystones is delivered in one kW). As you can see, for now Pine64 is winning.

Interesting finding: monthly cost of power for old Pentium4 and Pentium Core-2 machines is equal to cost of one, brand new Pine64+ 2GB.

5W is too high I recon. I never really measured more than 3W consumption. I'll hook it up with a stress test and try again and post the results.

----------------
Edit:
Ok sorry, I was thoroughly mistaken. Using StabilityTest 2.7 I ran the CPU stress test and it was pulling consistently even 5.8W, so your number seems good.
  Reply
#19
I am measuring around 2.6W when idle and with screen turned off. 4.3W under 50% load (2 heavily loaded cores). It goes to about 5W when loaded 100% on all cores, communicating over Ethernet heavily also adds a bit of power.


From the charts it looks that Pine64 is pretty impressive in terms of efficiency.
  Reply
#20
(04-23-2016, 04:21 PM)baryluk Wrote: Hey thanks!  The key package was boinc-app-seti which shows up in the package catalogue using synaptics but not visible using dpkg -l *UNTIL* you install it! Now, if we can only get the other boinc apps (E.g. pogs) to do the same.

By the way, boinc-manager (GUI) works just fine as it is the same version as boinc-client i.e. 7.6.31.

I do not yet have a heat sink.
Not over-clocked.

With 4 busy cores, I observed a steady temperature of 90C (ouch!).
I backed off to 2 busy cores and observe the temperature steady in the 60-70C range.

A heat sink is a good idea!
  Reply


Forum Jump:


Users browsing this thread: 1 Guest(s)