Thanks! Also much thanks to Robl - if it weren't for the thread they mentioned, I wouldn't have found out about the '--with-boinc-platform' and '--with-boinc-alt-platform' arguments for configuration. I completely missed those when going through BOINC's documentation for compiling the client yourself.
It looks like the scheduler is looking at the CPU features for “neon” and if it finds it gives out the 1.47 app. As far as I know the Pi2 can do neon.
I will try and fire up a Raspberry Pi2 using Raspbian to see what it does. Most of my fleet is Pi3’s and a couple of Pi4’s that are doing Rosetta work at the moment. I don’t tend to use the Pi2 due to their speed. The Pi3’s take around 11:30 to do a BRP4 work unit. I have no idea how long a Pi2 will take.
Standard install of Raspbian buster-lite, although after doing an apt upgrade it tried to install Pi-Bluetooth. That failed because the Pi2 doesn’t have Bluetooth. Attaching to Einstein worked as normal and it downloaded the 1.47 app and its crunching away.
Definitely seeing that they're taking their sweet time to do work, myself. Might have some bad CFLAGS - the ones I'm using now are "-O2 -pipe -march=armv7-a -mfpu=neon-vfpv4 -mfloat-abi=hard", but I should probably change from -O2 to -O3 for BOINC and doing
suggests my flags should be "-mfloat-abi=hard -mfpu=vfpv3-d16 -mtls-dialect=gnu -mtune=cortex-a7 -marm -march=armv7ve+simd" excluding -O3 and -pipe. Maybe I should try recompiling after the tasks they're working on now are done?
I've seen mention in a number of places of some "wisdom" thing and how it's improved performance for people. What's this about? I'm getting a lot of results when searching for that term on the site, but all I've found so far is on what it is.
First batch of 4 work units done using the project supplied app. Approx run time was 16 hours. I've got another 4 tasks going, which I will let it complete. After these 4 I will mothball it. Its no surprise they are somewhat slower that the Pi3B.
Average runtime so far has been 16.25 hours per task here, with a minimum of about 15.4 hours and a maximum of 17 hours. I'm still finding mentions of custom clients and using FFTW here and there, but I'm either missing something staring me right in the face or I can't find any guides for making custom clients using FFTW and the RPi GPU (if it's even worth it).
Thanks! Also much thanks to
)
Thanks! Also much thanks to Robl - if it weren't for the thread they mentioned, I wouldn't have found out about the '--with-boinc-platform' and '--with-boinc-alt-platform' arguments for configuration. I completely missed those when going through BOINC's documentation for compiling the client yourself.
It looks like the scheduler
)
It looks like the scheduler is looking at the CPU features for “neon” and if it finds it gives out the 1.47 app. As far as I know the Pi2 can do neon.
I will try and fire up a Raspberry Pi2 using Raspbian to see what it does. Most of my fleet is Pi3’s and a couple of Pi4’s that are doing Rosetta work at the moment. I don’t tend to use the Pi2 due to their speed. The Pi3’s take around 11:30 to do a BRP4 work unit. I have no idea how long a Pi2 will take.
MarksRpiCluster
Okay fired up one of my old
)
Okay fired up one of my old Pi2B’s. It’s this one https://einsteinathome.org/host/12824814
Standard install of Raspbian buster-lite, although after doing an apt upgrade it tried to install Pi-Bluetooth. That failed because the Pi2 doesn’t have Bluetooth. Attaching to Einstein worked as normal and it downloaded the 1.47 app and its crunching away.
MarksRpiCluster
Definitely seeing that
)
Definitely seeing that they're taking their sweet time to do work, myself. Might have some bad CFLAGS - the ones I'm using now are "-O2 -pipe -march=armv7-a -mfpu=neon-vfpv4 -mfloat-abi=hard", but I should probably change from -O2 to -O3 for BOINC and doing
suggests my flags should be "-mfloat-abi=hard -mfpu=vfpv3-d16 -mtls-dialect=gnu -mtune=cortex-a7 -marm -march=armv7ve+simd" excluding -O3 and -pipe. Maybe I should try recompiling after the tasks they're working on now are done?
I've seen mention in a number
)
I've seen mention in a number of places of some "wisdom" thing and how it's improved performance for people. What's this about? I'm getting a lot of results when searching for that term on the site, but all I've found so far is on what it is.
First batch of 4 work units
)
First batch of 4 work units done using the project supplied app. Approx run time was 16 hours. I've got another 4 tasks going, which I will let it complete. After these 4 I will mothball it. Its no surprise they are somewhat slower that the Pi3B.
MarksRpiCluster
Average runtime so far has
)
Average runtime so far has been 16.25 hours per task here, with a minimum of about 15.4 hours and a maximum of 17 hours. I'm still finding mentions of custom clients and using FFTW here and there, but I'm either missing something staring me right in the face or I can't find any guides for making custom clients using FFTW and the RPi GPU (if it's even worth it).
Hi folks! Well, there is
)
Hi folks!
Well, there is more simpler solution for wrong platform without recompiling.
Just create/edit cc_config.xml and add alt platform with right name:
and force client to read it:
boinccmd --read_cc_config
P.S.
Looks like multi-line code block not working -_-