A few days ago I received my first Pi3 and got it online. The Pi3 now joins two Pi2s in their search for the "elusive". All Piz are operating in the same room with an ambient room temperature of 79F. Yes summer has arrived early this year.
What follows are some numbers for each of the Piz. All Piz are running Ubuntu Mate as released, no overclocking mods have been made.
[pre]
pi type ambient room temp Pi temp Pi enclosure Run Time
pi2 26C ~59C open ~55k
pi2 26C ~39C pi tunnel ~71k
pi3 26C ~57C new pi tunnel ~37k
[/pre]
The Pi3 has only validated 4 jobs as of this time. All Piz are running 3 concurrent jobs utilizing the same E@H "profile" - school.
The Pi3 has had /root moved from the microSD card to a USB-to-mSATA drive. This increased the overall length of the PI since the mSATA plugs into one of the USB ports.
Here is a look at the "new" tunnel with the USB-to-mSATA plugged in:
Why just 3? Four jobs for the Pi3 should result in slower runtimes near 42k sec per task, but still more productive overall.
Quote:
pi tunnel ~71k
Hmmm... why should the Pi2 in the tunnel , running cooler, take longer for the tasks?? Are you sure about this value?
Today I had to cool one of my new PI3s that is running 4 BRP4 jobs in parallel, I even had a spare fan from an old notebook-cooling stand, but no "tunnel".... so inspired by your design I used a Pringles (TM) can :-). Cut off one end, fit the fan to it, insulate the inside (aluminum foil!!!) with a sheet of paper, put the Pi inside..done. It does work and had exactly the right diameter for the fan I happend to have! Too bad it has this aluminum foil inside (at least here in Germany), you have to insulate it to prevent bad things from happening when you put your Pi inside. Still interferes with WiFi I guess. For the moment it's ok. Maybe I can find a brand of potato chips that comes in un-metalized, plain cardboard cans :-).
Why just 3? Four jobs for the Pi3 should result in slower runtimes near 42k sec per task, but still more productive overall.
wanted to get a feel for temps so thought I would start with 3
Quote:
pi tunnel ~71k
Hmmm... why should the Pi2 in the tunnel , running cooler, take longer for the tasks?? Are you sure about this value?
I too noticed this when I posted. This Pi2 was built for robotics so it has a lot of bloat. This could be effecting it.The others were built for crunching so they don't have as much "stuff" loaded. I will have to look into it. My mistake. The "open" Pi was built for robotics. The Pi in question was built like the Pi3 so now I am really at a loss for this large disparity. I will look at its /boot/config.txt file and compare it to the other Piz.
[EDIT] I compared its setting in Boinc Manager to the other Pi's settings and they are the same. All 3 Piz reference the same "profile". So for the moment I am at a loss for the answer.
Quote:
Today I had to cool one of my new PI3s that is running 4 BRP4 jobs in parallel, I even had a spare fan from an old notebook-cooling stand, but no "tunnel".... so inspired by your design I used a Pringles (TM) can :-). Cut off one end, fit the fan to it, insulate the inside (aluminum foil!!!) with a sheet of paper, put the Pi inside..done. It does work and had exactly the right diameter for the fan I happend to have! Too bad it has this aluminum foil inside (at least here in Germany), you have to insulate it to prevent bad things from happening when you put your Pi inside. Still interferes with WiFi I guess. For the moment it's ok. Maybe I can find a brand of potato chips that comes in un-metalized, plain cardboard cans :-).
Quote:
Hmm. Crunching with a bouquet of chips. There truly is no end to the possibilities. :>)
The first thing to check is the CPU frequency while crunching (sudo vcgencmd measure_clock arm ) . It is possible to configure the CPU frequency scaling mechanism to throttle down the frequency when the CPU is not under load (as you would do for stuff running on batteries), and it is possible to configure it in a way so that jobs running with low prio (see Linux's "nice", this applies also to the BOINC apps!) do not count as processor load. You would then see the PI throttling down even if under full load from E@H jobs!
All 3 Piz reference the same "profile". So for the moment I am at a loss for the answer.
I didn't notice if you commented on this, but 71K is close to the speed of a Pi 2 running the non beta work units w/o a wisdom file in /etc/fftw/. If you are running the beta work units, it was commented above that the wisdom is included. Either add /etc/fftw/wisdomf following instructions from this thread: http://einsteinathome.org/node/198011 or try enabling beta work units in your profile.
Out of curiosity, why are your running Mate over Raspbian?
This returned 900000 on the behaving Pi2 and 600000 on the Pi2 with the 70k response times. I am assuming that this difference would cause the longer times. On the Pi3 it returned 1200000.
I found a suggestion on a website and after trying it lost the desktop. I tried to recover but could not. I was also concerned with what else might have been lost so I rebuild this node. The above "cat" command is now showing 900000 for the rebuild. I read " KF7IJZ" response and there was no wisdom file on any of my Ubuntu nodes and I did have "beta work" checked. I note HB's comment above about the internal wisdom, but not understanding the significance. Are there different type of jobs or are these with internal wisdom required for non Rasparian distros? I am not sure how the scaling_cur_freq was different on the problematic node. I suppose in a few hours I will know if this Pi2 is performing like the other one and that things make sense.
@KF7IJZ you asked why Ubuntu and not Raspian. I have several other Linux nodes all running Ubuntu. It easier to manage when all nodes are running the same distro. Also, and this is just my interpretation, I like the look and feel of Ubuntu Mate over Raspian. Again just my likes and dislikes. Raspian does seem more like an easier fit at times.
HB liked the "Chip Cooler". Whoda thunk. Achieves the same goal but at a much better price point.
Yes, that must be the problem: the CPU was downclocking for some reason.
Quote:
I note HB's comment above about the internal wisdom, but not understanding the significance.
Just means that the tasks in question used an app that has a "wisdom file" already built in, so there is no need (or benefit) to put a wisdom file in /etc/fftw/wisdomf .
RE: Just got my first
)
Seems about right. Mine has been averaging 39k seconds running the 1.47 BRP4 beta app.
BOINC blog
A few days ago I received my
)
A few days ago I received my first Pi3 and got it online. The Pi3 now joins two Pi2s in their search for the "elusive". All Piz are operating in the same room with an ambient room temperature of 79F. Yes summer has arrived early this year.
What follows are some numbers for each of the Piz. All Piz are running Ubuntu Mate as released, no overclocking mods have been made.
[pre]
pi type ambient room temp Pi temp Pi enclosure Run Time
pi2 26C ~59C open ~55k
pi2 26C ~39C pi tunnel ~71k
pi3 26C ~57C new pi tunnel ~37k
[/pre]
The Pi3 has only validated 4 jobs as of this time. All Piz are running 3 concurrent jobs utilizing the same E@H "profile" - school.
The Pi3 has had /root moved from the microSD card to a USB-to-mSATA drive. This increased the overall length of the PI since the mSATA plugs into one of the USB ports.
Here is a look at the "new" tunnel with the USB-to-mSATA plugged in:
More to see here
If interested I have made the ".stl" files and python code available for download at the above link, bottom of page.
Nice! RE: All Piz
)
Nice!
Why just 3? Four jobs for the Pi3 should result in slower runtimes near 42k sec per task, but still more productive overall.
Hmmm... why should the Pi2 in the tunnel , running cooler, take longer for the tasks?? Are you sure about this value?
Today I had to cool one of my new PI3s that is running 4 BRP4 jobs in parallel, I even had a spare fan from an old notebook-cooling stand, but no "tunnel".... so inspired by your design I used a Pringles (TM) can :-). Cut off one end, fit the fan to it, insulate the inside (aluminum foil!!!) with a sheet of paper, put the Pi inside..done. It does work and had exactly the right diameter for the fan I happend to have! Too bad it has this aluminum foil inside (at least here in Germany), you have to insulate it to prevent bad things from happening when you put your Pi inside. Still interferes with WiFi I guess. For the moment it's ok. Maybe I can find a brand of potato chips that comes in un-metalized, plain cardboard cans :-).
RE: Nice! RE: All Piz
)
wanted to get a feel for temps so thought I would start with 3
I too noticed this when I posted.
This Pi2 was built for robotics so it has a lot of bloat. This could be effecting it.The others were built for crunching so they don't have as much "stuff" loaded. I will have to look into it.My mistake. The "open" Pi was built for robotics. The Pi in question was built like the Pi3 so now I am really at a loss for this large disparity. I will look at its /boot/config.txt file and compare it to the other Piz.[EDIT] I compared its setting in Boinc Manager to the other Pi's settings and they are the same. All 3 Piz reference the same "profile". So for the moment I am at a loss for the answer.
Hmm. Crunching with a bouquet of chips. There truly is no end to the possibilities. :>)
you might want to look at this page: It has adjustments for running a Pi in a hot environment
RE: So for the moment I
)
The first thing to check is the CPU frequency while crunching (
sudo vcgencmd measure_clock arm) . It is possible to configure the CPU frequency scaling mechanism to throttle down the frequency when the CPU is not under load (as you would do for stuff running on batteries), and it is possible to configure it in a way so that jobs running with low prio (see Linux's "nice", this applies also to the BOINC apps!) do not count as processor load. You would then see the PI throttling down even if under full load from E@H jobs!E.g. see http://kubuntutunings.blogspot.de/2013/04/better-performance-with-ondemand-cpu.html , especially the "ignore_nice_load" part.
EDIT:
some pictures of the potato chip cooling PI3 cruncher:
RE: All 3 Piz reference the
)
I didn't notice if you commented on this, but 71K is close to the speed of a Pi 2 running the non beta work units w/o a wisdom file in /etc/fftw/. If you are running the beta work units, it was commented above that the wisdom is included. Either add /etc/fftw/wisdomf following instructions from this thread: http://einsteinathome.org/node/198011 or try enabling beta work units in your profile.
Out of curiosity, why are your running Mate over Raspbian?
My YouTube Channel: https://www.youtube.com/user/KF7IJZ
Follow me on Twitter: https://twitter.com/KF7IJZ
RE: I didn't notice if you
)
The tasks we are so puzzled about must be these here: https://einsteinathome.org/host/12210278/tasks&offset=0&show_names=0&state=4&appid= and as you can see, they were done by beta apps, so w/ internal wisdom.
RE: RE: I didn't notice
)
regarding the Pi2 with 70K times:
I started looking at HB's earlier suggestion and did the following command on all 3 PI nodes:
sudo cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
This returned 900000 on the behaving Pi2 and 600000 on the Pi2 with the 70k response times. I am assuming that this difference would cause the longer times. On the Pi3 it returned 1200000.
I found a suggestion on a website and after trying it lost the desktop. I tried to recover but could not. I was also concerned with what else might have been lost so I rebuild this node. The above "cat" command is now showing 900000 for the rebuild. I read " KF7IJZ" response and there was no wisdom file on any of my Ubuntu nodes and I did have "beta work" checked. I note HB's comment above about the internal wisdom, but not understanding the significance. Are there different type of jobs or are these with internal wisdom required for non Rasparian distros? I am not sure how the scaling_cur_freq was different on the problematic node. I suppose in a few hours I will know if this Pi2 is performing like the other one and that things make sense.
@KF7IJZ you asked why Ubuntu and not Raspian. I have several other Linux nodes all running Ubuntu. It easier to manage when all nodes are running the same distro. Also, and this is just my interpretation, I like the look and feel of Ubuntu Mate over Raspian. Again just my likes and dislikes. Raspian does seem more like an easier fit at times.
HB liked the "Chip Cooler". Whoda thunk. Achieves the same goal but at a much better price point.
Hi! Yes, that must be the
)
Hi!
Yes, that must be the problem: the CPU was downclocking for some reason.
Just means that the tasks in question used an app that has a "wisdom file" already built in, so there is no need (or benefit) to put a wisdom file in /etc/fftw/wisdomf .
Cheers
HB
Raspberry Pi microSD card
)
Raspberry Pi microSD card performance comparison - 2015
http://www.jeffgeerling.com/blogs/jeff-geerling/raspberry-pi-microsd-card
Sorry if this was posted somewhere else.