Research SBC consumption/performance comparisons

[hojnikb]

7 hours ago, superjamie said:

 

The configuration tool allowed users to try for 1GHz but it definitely wasn't achievable on all boards. I had a first-batch 256MiB RAM Pi 1 (which I bought the week they were released in 2012) and a later 512 MiB Pi 1, both of which could not reliably go past 900MHz. I've spoken to other Pi 1 owners who could achieve 950MHz or 1000MHz, and one owner whose board couldn't even get past 850MHz reliably.

 

If you had a Pi 1 reaching 1150MHz, you were very lucky and your experience was definitely not typical of most users.

You needed to bump the voltage to ~1.4V to hit those speeds. On stock voltage, most boards went to ~850-950Mhz (mine topped out at ~930mhz). If you cranked the voltage to 1.4V and didn't hit 1.0.Ghz reliably, then you definitely had the biggest turd of an example. 

[tkaiser]

On 4.8.2016 at 8:38 PM, tkaiser said:

Performance: sysbench takes 120 seconds (constantly at 1200 MHz, 80°C)

 

Little update on the most useless benchmark around. Since I glued together an Armbian armhf Debian userland with RPi bootloader/kernel I was able to test this on RPi 3 again: With normal Debian Jessie armhf sysbench binary execution is 134 seconds (12% slower). Now I upgraded to Debian stretch and then execution time is just 105 seconds (15% faster). At least Raspbian folks use better compiler switches when baking their packages compared to Debian armhf.

 

 

root@raspberrypi:~# sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=$(grep -c '^processor' /proc/cpuinfo)
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:                          134.4076s
    total number of events:              10000
    total time taken by event execution: 537.5008
    per-request statistics:
         min:                                 52.98ms
         avg:                                 53.75ms
         max:                                182.82ms
         approx.  95 percentile:              54.94ms

Threads fairness:
    events (avg/stddev):           2500.0000/7.25
    execution time (avg/stddev):   134.3752/0.02

root@raspberrypi:~# lsb_release -c
Codename:	jessie



root@raspberrypi:~# sysbench --test=cpu --cpu-max-prime=20000 run --num-threads=$(grep -c '^processor' /proc/cpuinfo)
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:                          105.6523s
    total number of events:              10000
    total time taken by event execution: 422.4823
    per-request statistics:
         min:                                 41.47ms
         avg:                                 42.25ms
         max:                                120.22ms
         approx.  95 percentile:              42.98ms

Threads fairness:
    events (avg/stddev):           2500.0000/3.08
    execution time (avg/stddev):   105.6206/0.01

root@raspberrypi:~# lsb_release -c
Codename:	stretch

 

And if we allow RPi 3 to run with ARMv8 code execution time will be below 8 seconds (if you don't run in under-voltage situations otherwise RPi firmware will silently play frequency capping and then your benchmark will run twice as long or half as fast). 'sysbench --test=cpu' is still a great compiler optimization benchmark and load/heat generator but you can't test different CPUs with it

 

Edit: Tests repeated with RPi 2 (BCM2836) -- no throttling occured, I used the default 900 MHz clockspeed:

 

• Raspbian/ARMv6: 192 sec
• Debian Jessie/ARMv7: 208 sec
• Debian Stretch/ARMv7: 153 sec

 

Edit 2: And just recently other SBC distros started to use sysbench to 'measure CPU performance' -- funny idea