Parallella, Raspberry Pi, FPGA & All That Stuff

robl
robl
Joined: 2 Jan 13
Posts: 1709
Credit: 1454481596
RAC: 8753

bowguy wrote:N30dG-ARM

bowguy wrote:
N30dG-ARM wrote:
Still using the wisdom that I've created on my RK3399 as the wisdom generated on the PI4 is way slower. Not shure why.

Is your version available for download?  Does it work on pi3B+?

great question!!  and the answer is ....

N30dG-ARM
N30dG-ARM
Joined: 20 Oct 17
Posts: 23
Credit: 22094059
RAC: 0

robl wrote:bowguy

robl wrote:
bowguy wrote:
N30dG-ARM wrote:
Still using the wisdom that I've created on my RK3399 as the wisdom generated on the PI4 is way slower. Not shure why.

Is your version available for download?  Does it work on pi3B+?

great question!!  and the answer is ....

In theory this will run on a PI3B/B+ but due to lack of RAM you could only run one Job at a Time. To run 4 task you need at least 4GB of Memory. This only makes sense on the 4GB - PI4 or other Boards with that amount of RAM. And it requires a 64bit OS to run.

 

As for a download. Still looking for a place to upload.

Meanwhile I've send a PM, with download instructions to both of you.

 

bowguy
bowguy
Joined: 2 Oct 17
Posts: 37
Credit: 10710134
RAC: 3830

Got it !  Thanks - it may

Got it !  Thanks - it may take a while to get to this.

robl
robl
Joined: 2 Jan 13
Posts: 1709
Credit: 1454481596
RAC: 8753

got it the PM but will not

got the PM but will not get to it until Tuesday at the earliest.

poppageek
poppageek
Joined: 13 Aug 10
Posts: 259
Credit: 2473733122
RAC: 0

https://www.raspberrypi.org/b

https://www.raspberrypi.org/blog/new-price-raspberry-pi-4-2gb

"A birthday gift: 2GB Raspberry Pi 4 now only $35"

"TL;DR: it’s our eighth birthday, and falling RAM prices have allowed us to cut the price of the 2GB Raspberry Pi 4 to $35."

PorkyPies
PorkyPies
Joined: 27 Apr 16
Posts: 197
Credit: 28020942
RAC: 8709

Rosetta@home released an app

Rosetta@home released an app "Rosetta for Portable Devices" that has both aarch64 and android versions. I did a blog post on how to get Raspbian and BOINC to use aarch64. See http://marksrpicluster.blogspot.com/2020/04/do-something-useful-with-your-pi4.html for details.

The drawback if you do this is you'll have to use N30DG-ARM's aarch64 app as Einstein doesn't have aarch64 apps.

bowguy
bowguy
Joined: 2 Oct 17
Posts: 37
Credit: 10710134
RAC: 3830

Thanks! I have 5 pi4's now

Thanks!

I have 5 pi4's now running Rosetta.  Stay safe

PorkyPies
PorkyPies
Joined: 27 Apr 16
Posts: 197
Credit: 28020942
RAC: 8709

PorkyPies wrote:The drawback

PorkyPies wrote:
The drawback if you do this is you'll have to use N30DG-ARM's aarch64 app as Einstein doesn't have aarch64 apps.

Unfortunately it doesn't work due to the 32 bit userland that Raspbian has. N30DG-ARM and I had a bit of discussion about this. I ended up grabbing the 1.47 armhf app that the project supplies from another Pi and setting it up to work as anonymous platform.

robl
robl
Joined: 2 Jan 13
Posts: 1709
Credit: 1454481596
RAC: 8753

There seems to be a lot of

There seems to be a lot of discussion regarding Piz.  Some in this thread and some in the "Getting Started" thread.  I decided to post here.  I am not claiming anyone's coding effort for my own.  I give credit where it is due.  The purpose of this post/thread is to provide info on how to crunch FGRP WUs on a Pi 4.  Two downloadable tar files are required.  One from one location and the second from one of my servers.  Both are required.  In the "Getting Started" thread there were some questions about "wisdom" files.  The second tar file contains a couple as well as how to make your own.   That I will leave to the "Einsteins" of this project.  

Here we go:

This procedure is based on the following:

  1.  I am using ubuntu 19.  It should work on other distros however.
  2.  the login/user I am using is ubuntu so you may have to change the “chown” commands   below to accommodate your login/user.
  3.  my $HOME will be different then yours.  just adjust.
  4.  While I have tried for accuracy there could be errors.  I will help when I can but I do not   have a lot of free time. Not trying to be a jerk but ….
  5.  There are two downloads – one from one location and another from my server.

PART 1:

  1. install boinc
  2. sudo apt-get install boinc-client boinc-manager
  3. get n30dg’s modified code
  4. cd /var/lib/boinc-client
  5. sudo wget http://kerbodyne.com/einstein/einstein_64bit_0.13.tar
  6. sudo chown boinc:boinc einstein_64bit_0.13.tar
  7. sudo tar -xvf einstein_64bit_0.13.tar
  8. restart boinc-client: systemctl restart boinc-client
  9. be sure that directories have proper ownership, i.e.,einstein.phys.uwm.edu should be owned by boinc:boinc and not root:root

PART 2:

Here I want to explain the files in the second tar file as well as providing some additional information.  But first the bulk of this effort was done by:  N30DG-Arm (christian beer – who did the bulk of the code modification),  PORKYPIES,  koschi, bikeman.

—————————–cut here ——————————————

Some Notes on the Files in this second tar file (below):

HSgammapulsar: Single-threaded Binary.

HSgammapulsar_n2: Multithreaded Binary (Highly experimental). Uses 2cores per Task, but only the FFTW part is multithreaded so performance is poor. Avg. CPU-usage is around 1.25. But you can give it a try on devices with low Memory. Boinc didn’t notice that it uses 2Cores per Task, so you have to manually limit Boinc to use only half the cores of your system.

FGRPB1wisdom.dat.S905.n2: Wisdom for the Multithreaded App, generated on a Amlogic S905 (Odroid C2)

FGRPB1wisdom.dat.rk3399.LowMem: Wisdom for the single threaded app, generated on a Rockchip RK3399, uses around 600MB per Task. Please use max 5 concurrent tasks on a system with 4GB.

FGRPB1wisdom.dat.rk3399.HighMem: Same as above but uses around 800MB per Task. Run’s much faster that the lowMem-wisdom. Perfect for a quad-core-system with 4GB of Memory. That’s what I’m using on my PI4’s.

fftwf-wisdom: App for generating your own wisdom.

 

As a starting point you can use one of my wisdoms. Please copy the desired wisdom-file to /etc/fftw/FGRP1Bwisdom.dat and make shure boinc can access it.

For ideal performance, you will have to create your own wisdom using the fftwf-wisdom tool. But notice that this will take at least several Day’s up to several Week’s to finish.

Use

./fftwf-wisdom -o wisdom.dat rib67108864to generate a  wisdom file with low memory profile.

or

./fftwf-wisdom -x -o wisdom.dat rib67108864to generate a  wisdom file with high memory profile. Takes way longer than above, around a month on a A53.

—————————–cut here ——————————————

2. download the tar file from http://usefulramblings.org/?page_id=12404 <—there will be a large blue button to the right titled:  “Pi 4 arm 64 code”
3. place this file in a directory called $HOME/eh
4. to view the contents of the tar file: tar -tvf eh.tar.gz (8 files)
5. goal: to move these files from the $HOME/eh directory to where they belong
in the boinc directory structure
6. cd to the directory where the the tar file is: $HOME/eh
7. extract the files: tar -xf eh.tar.gz
8 files are now in $HOME/eh
8. we now need to move them into the appropriate directories for boinc.
a. sudo mkdir /etc/fftw
b. sudo chown boinc:boinc /etc/fftw
c. cd $HOME/eh
d. sudo cp FGRPB1wisdom.dat.rk3399.LowMem /etc/fftw
e. sudo mv /etc/fftw/FGRPB1wisdom.dat.rk3399.LowMem FGRPB1wisdom.dat
f. sudo chown boinc:boinc FGRPB1wisdom.dat
g. sudo cp $HOME/eh/app_info.xml /var/lib/boinc-client/projects/einstein.phys.uwm.edu
h. sudo chown boinc:boinc /var/lib/boinc-client/projects/einstein.phys.uwm.edu/app_info.xml
i. sudo cp $HOME/eh/HSgammapulsar /var/lib/boinc-client/projects/einstein.phys.uwm.edu
j. sudo chown boinc:boinc /var/lib/boinc-client/projects/einstein.phys.uwm.edu/HSgammapulsar
k. sudo chmod +x /var/lib/boinc-client/projects/einstein.phys.uwm.edu/HSgammapulsar

Part 3:

9. on the Einstein@Home website for your location: “Use CPU” in project setting set to “yes
10. on the Einstein@Home website for your location: “Run CPU versions of applications for which GPU versions are
available set to “Yes

10.  on the Einstein@Home website under account–>preferences–project–> preference-set check “gamma ray pulsar search #5”

11. systemctl restart boinc-client <– restart boinc-client (all currently running jobs will be lost.

12.  join the einstein@home project.

I have made every effort for accuracy but you know how that goes.   

Sid
Sid
Joined: 17 Oct 10
Posts: 160
Credit: 920880079
RAC: 280183

robl

robl wrote:

HSgammapulsar_n2: Multithreaded Binary (Highly experimental). Uses 2cores per Task, but only the FFTW part is multithreaded so performance is poor. Avg. CPU-usage is around 1.25. But you can give it a try on devices with low Memory. Boinc didn’t notice that it uses 2Cores per Task, so you have to manually limit Boinc to use only half the cores of your system.

I'm using this application for a couple of months and I like it. No any issue.

Thank you again Robl.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.