Pine 64 benchmarks

[longsleep]

Oh cool what a change in temperature. Looks good for a first improvement and I think it is pretty good with a larger/better heat sink. Though the end of the monitor graph looks strange, high temp no CPU cores .. What happened there?

[Andrew2]

Here we go: http://openbenchmarking.org/result/16030...603051GA21

Everything necessary to interpret the results available here: http://kaiser-edv.de/tmp/mc6CyL

[Andrew2]

And another one: http://openbenchmarking.org/result/16030...603075GA69

[janjwerner]

you've been busy while I was sleeping
Those are all with:
export CXXFLAGS="-march=armv8-a -mtune=cortex-a53"
export CFLAGS="-march=armv8-a -mtune=cortex-a53"
and aggressive throttling & heatsink + fan?

[Andrew2]

Less aggressive throttling And the most important part was adjusting the cooler table since you can influence throttling behavior this way. Please don't think about it without going through the whole process outlined here: https://github.com/longsleep/build-pine64-image/pull/3

My A64 is stable with these settings but it's a huge amount of work to verify this.

The submitted results were meant for the Phoronix founder so he might get the idea that the most important thing when testing modern SBC is to look at thermal and especially throttling behavior (something he still ignores more or less).

[janjwerner]

ok, optimized version, no aggressive throttling just better compiler options:
export CXXFLAGS="-march=armv8-a -mtune=cortex-a53"
export CFLAGS="-march=armv8-a -mtune=cortex-a53"

results (pine64-optimized): http://openbenchmarking.org/result/16030...603051GA05
cpufreq: http://pastebin.com/YXWRnpxA
rpimonitor: http://imgur.com/x2T6QcN

for some reason john benchmark did not run, all the other results look promising
RPI monitor for some reasons show cpu count during C-Ray test. this is clearly incorrect:
there is no CPU killed in dmesg
and this is htop output: http://imgur.com/igH14q3

[Andrew2]

Promising? The results aren't better than before, quite the opposite. Unless you improve heat dissipation the whole thing is useless since compiler optimisation might generate more heat, throttling happens earlier and the result looks slower.

Disable longsleep's core-keeper and try cpuburn-a53 to get the idea ;-)

BTW: without 'force-install' tests might not be rebuilt.

[janjwerner]

promising in terms of in line of resulsts with heatsink / fan with less agressive throttling.
I burnt my fingers trying cpuburn-a53 checking the heatsink

[Andrew2]

Funny: John the ripper ignores CFLAGS by default. After patching the Makefile the score went from 763 up to 932 and would've exceeded 1000 if heavy throttling wouldn't happened.

The only thing we could learn from the Phoronix Test Suite is that compiler settings and thermal settings matter (the stuff that will be ignored normally )

---

http://openbenchmarking.org/result/16030...603082GA36

-O3 not active with every test. I stop here since passive benchmarking this way is just a waste of time :-)

[longsleep]

Thanks Andrew2 and janwerner. Those results are very helpful for those willing to understand them.

For now i do not see a reason to change the cooler table in the default device tree. It does not loose cores with a8 load on all 4 cores and it is essentially impossible/pointless to benchmark a53 optimized code on all cores without a good heat dissipation solution. At least the cores come back now and i avoid reports why the Pine64 is faster when rebooted