Page 1 of 1
GPU computation
Posted: Thu Nov 11, 2010 4:57 pm
by M Cluster
Hello Cummity,
It would be very nice if the Eon2 - client 1.0 can also be used by GPU.
Is it being considered for the Eon Project is an Cuda support to programming, if so, by when this would be done?
I could add some TFlops because my GPU cluster is a cluster rather.
If no Cuda support is planned? Could the source be disclosed to me? Respectively. the community to incorporate this support?
Thank you
Regards
Re: GPU computation
Posted: Fri Nov 12, 2010 6:27 pm
by skipdashu
Or... would this application lend it itself better to the multi-thread capability of the ATI cards.
Skip,
Hutto, TX
PS: Personally only interested in Linux apps.
Re: GPU computation
Posted: Tue Nov 16, 2010 12:00 am
by M Cluster
Yes Ati would support inside. Only problem is you'd need the source code to programmers for anything.
Eon are the source to write freely about a GPU support?
Regards
Re: GPU computation
Posted: Tue Nov 16, 2010 3:43 am
by graeme
These are good questions. We'll work on a multithreaded client first (perhaps using mpi). We have not found an efficient way to use GPUs yet, but there is some possibility with one of our potentials. Anyway, GPUs will not be supported any time soon.
Re: GPU computation
Posted: Tue Nov 16, 2010 10:14 am
by M Cluster
Ok, there is a possibility to write their own clients for Eon? I may find a way for multithreading or porting to GPU computation.
What are the chemical equations calculated their files?
Is it allowed an own client to write and test this?
Regards
Re: GPU computation
Posted: Tue Nov 16, 2010 4:39 pm
by graeme
Sure, by all means. While there is quite a bit of code, the computational work is localized to a few function calls that evaluate the force and energy between the atoms. These are all in the potentials subdirectory of the client code.
Re: GPU computation
Posted: Tue Nov 16, 2010 7:52 pm
by M Cluster
Well, I look at the files in the tar at times. uncompiled version of Linux could be published clientes?
I would then begin.
Regards
Re: GPU computation
Posted: Tue Nov 16, 2010 7:58 pm
by chill
We now have anonymous read-only access to our subversion repository for eon. The url is
http://theory.cm.utexas.edu/svn/eon. It can be checked out like so:
Code: Select all
svn co http://theory.cm.utexas.edu/svn/eon
Re: GPU computation
Posted: Thu Nov 18, 2010 11:54 pm
by M Cluster
Hi,
I have now started a cuda class to write.
now I have tried to compile your revision 640 without my improvement.
have generated the client file and renamed the copied project folder.
You experience the following error.
looking a what have I done wrong? standing on the hose: P
19-Nov-2010 00:36:33 [eon2] Message from server: This computer has reached a limit on tasks in progress
19-Nov-2010 00:36:57 [eon2] Task 1709711176_48_66543_0 exited with zero status but no 'finished' file
19-Nov-2010 00:36:57 [eon2] If this happens repeatedly you may need to reset the project.
19-Nov-2010 00:36:57 [eon2] Restarting task 1709711176_48_66543_0 using eonclient version 110
19-Nov-2010 00:37:07 [eon2] Task 1709711176_48_66541_0 exited with zero status but no 'finished' file
19-Nov-2010 00:37:07 [eon2] If this happens repeatedly you may need to reset the project.
19-Nov-2010 00:37:07 [eon2] Restarting task 1709711176_48_66541_0 using eonclient version 110
Re: GPU computation
Posted: Fri Nov 19, 2010 1:06 pm
by M Cluster
The new errors.
What is it wrong SVN Rev 640th
without my extensions.
Code: Select all
19-Nov-2010 14:03:56 [eon2] Sending scheduler request: To fetch work.
19-Nov-2010 14:03:56 [eon2] Requesting new tasks
19-Nov-2010 14:03:57 [eon2] Started download of 1709711176_85_367436.tgz
19-Nov-2010 14:03:58 [eon2] Finished download of 1709711176_85_367436.tgz
19-Nov-2010 14:03:59 [eon2] Scheduler request completed: got 7 new tasks
19-Nov-2010 14:04:01 [eon2] Started download of 1709711176_85_367845.tgz
19-Nov-2010 14:04:01 [eon2] Started download of 1709711176_85_367846.tgz
19-Nov-2010 14:04:01 [eon2] [error] File eonclient_1.10_i686-pc-linux-gnu has wrong size: expected 2931279, got 8476514
19-Nov-2010 14:04:04 [eon2] Finished download of 1709711176_85_367845.tgz
19-Nov-2010 14:04:04 [eon2] Finished download of 1709711176_85_367846.tgz
19-Nov-2010 14:04:04 [eon2] Started download of 1807230871_105_1133250.tgz
19-Nov-2010 14:04:04 [eon2] Started download of 1807230871_105_1133275.tgz
19-Nov-2010 14:04:07 [eon2] Finished download of 1807230871_105_1133250.tgz
19-Nov-2010 14:04:46 [eon2] Task 1709711176_85_367436_0 exited with zero status but no 'finished' file
19-Nov-2010 14:04:46 [eon2] If this happens repeatedly you may need to reset the project.
Re: GPU computation
Posted: Fri Nov 19, 2010 4:30 pm
by chill
M Cluster wrote:The new errors.
What is it wrong SVN Rev 640th
without my extensions.
I wouldn't try to test your client the way you are doing it. It probably doesn't work because has not been built with the boinc library (by running "BOINC=1 make" in the client directory). The client also supports being compiled in a standalone mode, which is most likely what you have built. I would instead run one of the samples like in "client/samples/al". Just run the client you built in that directory after running the init script in that directory to setup the files.
Re: GPU computation
Posted: Sun Nov 21, 2010 5:02 pm
by M Cluster
Hello So I try to explain my whole steps.
1. Creating Directories
Code: Select all
cd / home
mkdir boincadm
cd boincadm
Code: Select all
svn co http://boinc.berkeley.edu/svn/trunk/boinc
so that the BOINC source downloaded / revision currently 22 729
Code: Select all
svn co http://theory.cm.utexas.edu/svn/eon
Eon Source - Revision 645
3. present term folder structure
Dir: /home/boincadm
We now have 2 new subfolders.
1. / Home / boincadm / boinc
2. / Home / boincadm / eon
see in append
boinctree.txt
eontree.txt
4. Compiling BOINC
The following commands must be executed
We are in the folder
"/ Home / boincadm"
Code: Select all
cd boinc
. / _autosetup
. / Configure - disable-server
make
We must now copy the eon Boinc Libraries
boinc_api.h
diagnostics.h
filesys.h
boinc_win.h
win_util.h
Code: Select all
cp / home / boincadm / boinc / api / boinc_api.h / home / boincadm / eon / boinc / boinc_api.h
cp / home / boincadm / boinc / lib / diagnostics.h / home / boincadm / eon / boinc / diagnostics.h
cp / home / boincadm / boinc / lib / filesys.h / home / boincadm / eon / boinc / filesys.h
cp / home / boincadm / boinc / lib / boinc_win.h / home / boincadm / eon / boinc / boinc_win.h
cp / home / boincadm / boinc / lib / win_util.h / home / boincadm / eon / boinc / win_util.h
*******
we certainly need to configure the client eon
boinc config - "/ home / boincadm / eon / boinc / config.ini"
- Here I have no plan now which come in this folder
I tried several variants.
Now we compile the client Eon
6. test of client Eon
Copy the client in the test folder
Code: Select all
cp / home / boincadm / eon / client / client / home / boincadm / eon / client / samples / al / client
init files for creating the chem
Code: Select all
. Cd / home / boincadm / eon / client / samples / al
. / Init
. / Client
. / Clean
if no errors occur here is how the client
Question:
So my client is great in about 8 mb
now how do I put the configs so that the Boinc lib is used?
Code: Select all
EON client Compilation Options
BOINC: No
DEBUG: Yes
POTENTIAL: EMT + + + MORSE LJ-EAM + QSC + Pt + H2O-VASP + + + Aluminum bopfox Lenosky + SW + + Tersoff EDIP
because I get the Eon client only compiled in debug mode
He works because he expects the samples correctly.
[/color]
Re: GPU computation
Posted: Mon Nov 22, 2010 12:16 am
by chill
M Cluster wrote:
Question:
So my client is great in about 8 mb
now how do I put the configs so that the Boinc lib is used?
Code: Select all
EON client Compilation Options
BOINC: No
DEBUG: Yes
POTENTIAL: EMT + + + MORSE LJ-EAM + QSC + Pt + H2O-VASP + + + Aluminum bopfox Lenosky + SW + + Tersoff EDIP
because I get the Eon client only compiled in debug mode
He works because he expects the samples correctly.
[/color]
You can compile the client with the boinc libraries by setting the environment variable BOINC to any value at all. If you use the bash shell you can build it like so:
I don't understand why you want to go to all the effort of building a boinc client and running a boinc server if your goal is to add GPU support to the client. That part has nothing to do with boinc specific apis. You should just build the client without boinc support in order to do your tests. By the way, 99% of the cpu time is spent calculating the force, which is done by the potential. Thus it probably only makes sense to move these parts to the GPU. The potential that is used is set in the parameters_passed.dat file and the code for the different potentials is in client/potentials/. The most simple potential is probably the morse or lennard jones potential.