| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: Android Support Forum
Thread: Performance difference in big.LITTLE cores |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 5
|
|
| Author |
|
|
fuzzydice555
Advanced Cruncher Joined: Mar 25, 2015 Post Count: 89 Status: Offline Project Badges: 
|
Hello All!
----------------------------------------I recently got my hands on a few used Odroid XU4 and Rock Pi 4 machines. Both of those are big.LITTLE designs, with some cores built for efficiency, some for performance. However, when I check the results, there is no major difference between runtimes between tasks. All machines are running on 100% CPU, in BoincTasks I see there are indeed the max 6 / 8 tasks running concurrently. What could be the reason that I see no short runtime tasks (big core) vs long runtime (little cores) tasks? I have two guesses: - Maybe the scheduler is actually rotating tasks between cores, so a single task has some runtime on big and some runtime on a little core? - Or the difference between big & little cores is not significant enough to show up in statistics?  |
||
|
|
PMH_UK
Veteran Cruncher UK Joined: Apr 26, 2007 Post Count: 786 Status: Offline Project Badges: 
|
Windows and Linux appear to rotate between cores so likely Android does too.
----------------------------------------I found some tasks sometimes slowed until I locked some/all tasks to a core/thread. Paul.
Paul.
|
||
|
|
Acibant
Advanced Cruncher USA Joined: Apr 15, 2020 Post Count: 126 Status: Offline Project Badges:
|
The only performance difference I see on my Odroid N2+ units is between different MCM tasks that use one of two algorithms. So I would agree that tasks are getting rotated between cores.
---------------------------------------- |
||
|
|
JLDun
Cruncher Joined: Aug 17, 2025 Post Count: 2 Status: Offline |
Due to the way Android operates, the high performance cores are primarily reserved for "important" stuff, which is usually OS functions as needed, everything else is on the low performance cores; most devices are split 50/50 on having each type of core. With BOINC, setting "cores in use" to 50% or less keeps everything in the lower cores, while setting "cores in use" to higher than 50% actually results in tasks being swapped in and out of the lower cores, avoiding the higher cores altogether. The net result is running up the "wall clock" time reported for each task while waiting for an available core, for the same amount of actual processing done as if you kept it set to 50% or lower.
|
||
|
|
TonyEllis
Senior Cruncher Australia Joined: Jul 9, 2008 Post Count: 286 Status: Offline Project Badges: 
|
fuzzydice555 scribed :-
----------------------------------------Maybe the scheduler is actually rotating tasks between cores Running Einstein WUs on a Radxa ROCK 4C+. OS is Debian GNU/Linux 11 (bullseye) aarch64 distribution. CPUs are dual-core Arm Cortex A72 and Arm Cortex quad-core A53 Using "top" with the "1" option shows that the WUS are being continuously moved from one CPU to another. This system has two performance CPUs and 4 lesser ones. There is a difference in frequency between the 2 CPU types. Have not thoroughly investigated the difference between the CPUs, but some bechmarks seem to indicate an A72 CPU is some 1.3x higher than an A53 at the same clock speed. I maintain a constant moderate CPU temperature, hence the faster 2 CPUs are running slower, as shown below, than the maximum possible (1512000). root@rocky:~# cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq
Run Time Stats https://grassmere-productions.no-ip.biz/
----------------------------------------[Edit 1 times, last edit by TonyEllis at Aug 30, 2025 4:02:37 AM] |
||
|
|
|