ARM CPU Stress Test

[jzhang18]

I saw this thread:
https://core-electronics.com.au/tutorial...ry-pi.html

sudo apt-get install stress
wget https://raw.githubusercontent.com/ssvb/c...burn-a53.S
gcc -o cpuburn-a53 cpuburn-a53.S

For Raspberry PI 3,  there are 4 A53 cores only.  for RockPro64.   How to test the other A72 cores? 

Can I use cpuburn-a53 to test A72 cores?

[Bullet64]

Will this work???

This however, will only stress a single core, but to to put the whole CPU (all 4 cores) under full stress (oh come on dudes, have mercy! ) I’ll have to open 4 terminals/tabs and enter “burnP6” under each Terminal window. So altogether, 4 “burnP6” programs will be running simultaneously. That’s it!.

Source: https://www.hecticgeek.com/2012/03/cpubu...ntu-linux/

[jzhang18]

Will this work???

This however, will only stress a single core, but to to put the whole CPU (all 4 cores) under full stress (oh come on dudes, have mercy! ) I’ll have to open 4 terminals/tabs and enter “burnP6” under each Terminal window. So altogether, 4 “burnP6” programs will be running simultaneously. That’s it!.

Source: https://www.hecticgeek.com/2012/03/cpubu...ntu-linux/

That is only for x86 CPU, not for ARM CPU,  am I correct?

[dukla2000]

Can I use cpuburn-a53 to test A72 cores?

Try it - you will see it does. But the basic cpuburn-a53 code is flawed as it assumes 2, 4, 8 processors and so only runs 4 threads on the RK3399 hexa core. Linux big.LITTLE schedulling is smart enough to ensure 2 go on the A72s, but even so only heats my system to 65 degrees after 10 minutes.

Better is arguably cpuminer benchmarks which use neon instructions. Seem to remember a thread somewhere of folks measuring actual power usage to substantiate the "better" part of this. But is e.g. part of sbc-bench which also has some neat background monitoring (if you are not already running a monitor). And it is easy to modify the script to avoid the other benchmark bits and run cpuminer for 10 minutes if you wish (which gets my setup to 72 degrees).

Code:

...
        CheckRelease
#        CheckLoad
        BasicSetup >/dev/null 2>&1
        InstallPrerequisits
        InitialMonitoring
        CheckClockspeeds
        CheckTimeInState before
#        RunTinyMemBench
#        RunOpenSSLBenchmark
#        Run7ZipBenchmark
        if [ "${TestNEON}" = "yes" -a -x "${InstallLocation}"/cpuminer-multi/cpuminer ]; then
            RunCpuminerBenchmark
        fi
        CheckTimeInState after
        CheckClockspeeds # again after heating the SoC to the max
#        DisplayResults
    fi
} # Main
...
    "${InstallLocation}"/cpuminer-multi/cpuminer --benchmark --cpu-priority=2 >${TempLog} &
    MinerPID=$!
    sleep 600
...

[jzhang18]

It seems both cpuminer and cpuburn, need to be compiled from source code.

Let me install gcc-6, and switch from gcc-7 to gcc-6, then compile those software.

thanks for the suggestions.

[dukla2000]

If you download and run sbc-bench as per the github instructions it downloads and installs a version of cpuminer for you.

[jzhang18]

Yes, started to run sbc-bench.

It seems it will take quite some time to finish. I just leave it running there.

---

This is my result:
http://ix.io/1lBI

This is the sample result for RockPro64 from sbc-bench README.
http://ix.io/1iEb

---

But where is cpuminer downloaded and installed?

---

It seems my board is slow with the 4GB RAM. not sure why.

[dukla2000]

---

But where is cpuminer downloaded and installed?

---

try

Code:

sudo /bin/bash ./sbc-bench.sh neon

He has just updated the tool (and removed neon example - not sure if it still works like that but would expect so!)

His results are on 4.18 kernel which is generally faster.

PPS - see the tool revision went to 0.6 then 0.6.1 today, main change is the parameters! neon retained for backwards compatibility
My results today.
Not sure if it is significant but seems my 2G memory is marginally faster (and lower latency) than your 4G!
Interesting the new -T option, with
sudo /bin/bash ./sbc-bench.sh -T 70
my system heats from 70.6 at start of 5 minutes to 75.6 at end so figure my cooling OK (Pine64 30mm tall heatsink, with arctic alumina thermal paste instead of thermal pad and vertically mounted. Ambient 20.)

[jzhang18]

sudo ./sbc-bench.sh -c
result - http://ix.io/1lC4
sudo ./sbc-bench.sh NEON
result - http://ix.io/1lD5
Not sure there is much difference or not.

[Luke]

Take a look here: https://forum.armbian.com/topic/7819-sbc...ment=60880