ARM Linux
1 Dec 2013 22:59:36 UTC
Topic 197292
(moderation:
Have installed a Ubuntu version on my odroid. BM is running, first Einstein tasks downloaded.
http://einsteinathome.org/host/9699248/tasks
The wu's have an estimated runtime of 7 hrs 10 min.
Have also tried the Albert wu's; they have an estimated runtime of > 14 hrs. Both the 1.06 versions. Have cancelled the Alberts.
It looks like computation is faster under Linux (more than one hr / wu). Would like to run the 1.07 version with neon support. Is there anything I have to do to get these wu's? The log file says no usable gpu found; but is 'NEON' a gpu in BOINC terms?
Cheers
Alexander
ARM Linux
)
Hi!
Thanks for the feedback. Android is not ideal for floating point heavy applications because it uses (at least for versions so far) a mechanism to pass around floating point numbers that is backwards-compatible, but doesn't make full use of floating point hardware. Most Linux distributions for ARM now are assuming hardware that is capable to handle floating point data in a more efficient way and are optimized accordingly.
NEON is a CPU feature, not a GPU (or co-processor in BOINC-speak), so it's more like SSE for x86 CPUs.
The NEON enabled ARM Linux app isn't yet installed on Einstein, just on Albert. I guess I can 'risk' it now to install it on Einstein@Home as well.
Cheers
HB
I looked around in the
)
I looked around in the hardkernel forum and found a kernel update for Mali (the hardware for NEON).
Downloading this was not a problem, but installation was (and is).
It's now time to ask for help.
I double-clicked the tar file in the download area and found some files that i copied into a temp directory.
The readme gives instructions how to install. The result can be seen here:
https://dl.dropboxusercontent.com/u/50246791/odroid/install%20error%201.PNG
OK, it did not work.
But double clicking the .deb files opened the Ubuntu software center and started some installation.
The result was that my odroid was running on 3 cores only, also after rebooting it.
Fortunately I had a backup. But 2 wu's were lost.
I'm out of ideas of how to install it.
Help apreciated!
Isn't Mali the GPU? MrS
)
Isn't Mali the GPU?
MrS
Scanning for our furry friends since Jan 2002
RE: Isn't Mali the
)
Ahmm .. yes, i mixed some things in my brain and need to run a defrag there.
I have read so much about kernel updates (i hope this will bring my 'neon' to live) which are named 'Mali kernel update', found a new version of image file and the magic comment 'This image is compatible with the kernel update script :) '.
So hope is back again, but I need to pick up another SD-card because I don't like to overwrite the current, working sd-card.
In addition I wanted to make something like an 'autostart entry' for the X11 VNC server; it's very different to windows and the usual wiki's do not help here, this is a smaller version and not all features are implemented.
Anyway, a toy that's good to learn a lot.
Alexander
Hi I finally installed the
)
Hi
I finally installed the new app version with NEON support for Linux on ARM at Einstein@Home.
For this to work, your BOINC client has to be able to detect the presence of NEON support in the CPU, you can check by looking at the startup messages in the Event Log (in BOINC Manager) where the detected processor features should be listed.
Cheers
HB
RE: RE: Isn't Mali the
)
here is what I did on a raspberry pi to get tightvnc server to autostart.
cd /home/pi
cd .config
mkdir autostart
cd autostart
nano (vi) tightvnc.desktop
enter the following
[Desktop Entry]
Type=Application
Name=TightVNC
Exec=vncserver :1
StartupNotify=false
save the file
DONE
RE: here is what I did on
)
Great, THX Robl.
I'm currently writing the new image to the SD-card, takes >2 hrs under windows.
I'll try that on sunday.
Cheers,
Alexander
Sat 07 Dec 2013 07:21:13 AM
)
Sat 07 Dec 2013 07:21:13 AM UTC | | Processor features: swp half thumb fastmult vfp edsp neon vfpv3 tls vfpv4 idiva idivt
;))))
https://dl.dropboxusercontent.com/u/50246791/odroid/first%20neons.PNG
edit:
http://einsteinathome.org/host/9744481/tasks
and this is the same device running Android
http://einsteinathome.org/host/9525126/tasks
Excellent, thanks for
)
Excellent, thanks for sharing, looking forward to seeing the results.
It would be interesting to see whether the FFT phase can be optimized by what is called 'wisdom' in the terminology of the FFT library used, FFTW: With a little tool you let the machine try out different possible ways to compute the FFT and by benchmarking, the tool finds a near optimal way for the given hardware. This can take a very long time tho (many hours to a day or more) so we cannot do this inside the app.
Since I still don't have a NEON capable ARM board under Linux (thanks to the delays of Parallella, see another thread), I would need someone to volunteer for that.
Cheers
HBE
RE: Since I still don't
)
Pls check your PM
Alexander