GPU GROMACS

Message boards : Number crunching : GPU GROMACS

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Profile Jack Shultz
Avatar

Send message
Joined: 10 Apr 09
Posts: 503
Credit: 120,150
RAC: 0
Message 631 - Posted: 17 Jun 2009, 22:59:15 UTC

Ok so if anyone has the time to work on the GPU code for gromacs I'll tell you where I'm stuck.

In cygwin I did the the following to build the openMM API
cmake .
make
make install
These scripts seemed to execute without any errors

Then I took the source code from the linux distribution and pasted the
source, makefile and autoconf files to the corresponding gromacs
directories.

kernel -> Makefile.am md.c md_openmm.cpp md_openmm.h
gromacs-4.0.3 -> configure.ac

I performed the following steps

autoconf
./configure
make

Now I'm doing this on two systems, one of them does not have a gpu but
has the cuda developer toolkit and sdk
Without the gpu I'm stuck here

md.o:md.c:(.text+0x1d4e): undefined reference to `_openmm_init'
md.o:md.c:(.text+0x2364): undefined reference to `_openmm_take_one_step'
md.o:md.c:(.text+0x2499): undefined reference to `_openmm_cleanup'
md.o:md.c:(.text+0x2738): undefined reference to `_openmm_copy_state'
collect2: ld returned 1 exit status
make[3]: *** [mdrun.exe] Error 1
make[3]: Leaving directory `/home/jshultz/gromacs-4.0.3/src/kernel'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/home/jshultz/gromacs-4.0.3/src'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/home/jshultz/gromacs-4.0.3/src'
make: *** [all-recursive] Error 1

With the gpu machine


make: *** [all-recursive] Error 1
$ make
Making all in include
make[1]: Entering directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/include'
Making all in .
make[2]: Entering directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/include'
make[2]: Nothing to be done for `all-am'.
make[2]: Leaving directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/include'
Making all in types
make[2]: Entering directory
`/cygdrive/c/Users/jshultz/gromacs-4.0.5/include/types'
make[2]: Nothing to be done for `all'.
make[2]: Leaving directory
`/cygdrive/c/Users/jshultz/gromacs-4.0.5/include/types'
make[1]: Leaving directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/include'
Making all in src
make[1]: Entering directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/src'
make all-am
make[2]: Entering directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/src'
make[2]: *** No rule to make target `add_par.lo', needed by
`libgmxpreprocess.la'. Stop.
make[2]: Leaving directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/src'
make[1]: *** [all] Error 2
make[1]: Leaving directory `/cygdrive/c/Users/jshultz/gromacs-4.0.5/src'
make: *** [all-recursive] Error 1
$

So either way I hit wall. I hope they provide the binaries for windows
and possibly some compile instructions. We want to integrate the BOINC
api.


ID: 631 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Tim Turner
Avatar

Send message
Joined: 1 May 09
Posts: 570
Credit: 184,322
RAC: 0
Message 635 - Posted: 19 Jun 2009, 4:23:19 UTC - in response to Message 631.  
Last modified: 19 Jun 2009, 4:37:38 UTC

well, it seems that you released some wu for gpu, well damn these output files....

06/19/09 00:21:25 DrugDiscovery Starting task run_1245375086989221465_4 using gromacs_gpu version 0
06/19/09 00:21:35 DrugDiscovery [error] Can't rename output file run_1245375086989221465_4_0 to projects/boinc.drugdiscoveryathome.com/run_1245375086989221465_4_0: Error 2
06/19/09 00:21:35 DrugDiscovery Computation for task run_1245375086989221465_4 finished


server (att) had 12 and gave me 12 gpu app wu all of which failed.. but i told i would be your tester for gpu windows app.

wu

<core_client_version>6.6.31</core_client_version>
<![CDATA[
<message>
- exit code 195 (0xc3)
</message>
<stderr_txt>
wrapper: starting
wrapper: running ../../projects/boinc.drugdiscoveryathome.com/unzip_0.000_windows_intelx86__cuda.exe (-qq -n bin.zip &nbsp;--device 0)
caution: filename not matched: &nbsp;--device
caution: filename not matched: 0
app exit status: 0xb
called boinc_finish

</stderr_txt>
]]>

http://boinc.drugdiscoveryathome.com/result.php?resultid=653525
Tim Turner
Public Relations Admin
Secunia PSI: http://secunia.com/vulnerability_scanning/personal/
If you need help via voice or Convo; PM me and i will give you details on where i will be; Teamspeak, Yahoo Messenger, or Skype.
ID: 635 · Rating: 0 · rate: Rate + / Rate - Report as offensive
zombie67 [MM]
Volunteer tester
Avatar

Send message
Joined: 25 Apr 09
Posts: 58
Credit: 1,785,257
RAC: 135
Message 637 - Posted: 19 Jun 2009, 5:30:31 UTC

If you are going to release GPU tasks, a few suggestions:

- The prefs page needs to have more than "allow GPU tasks Y/N". We need to select only GPU, only CPU, or both. And then another box for accept work from the unselected if work for the selected is not available.
- The apps page needs to show GPU apps too. I'm not seeing anything but CPU apps.
- The front-front page has no link to the apps page. You have to know to go to the BOINC page to then go to the apps page. (An aside: I really don't like the non-BOINC front end to the site. It makes many of the "list" pages (stats, tasks, etc) way too narrow. I prefer just the normal BOINC format.
@zombie_67

ID: 637 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 638 - Posted: 19 Jun 2009, 6:08:40 UTC - in response to Message 637.  
Last modified: 19 Jun 2009, 6:11:28 UTC

- The prefs page needs to have more than "allow GPU tasks Y/N". We need to select only GPU, only CPU, or both. And then another box for accept work from the unselected if work for the selected is not available.

That's a backend software update including David knows what bugs it extra gives. But can be done.

- The apps page needs to show GPU apps too. I'm not seeing anything but CPU apps.

Not sure which page you're looking at, but DrugDiscovery Apps shows "GROMACS with Nvidia GPU" for me.

- The front-front page has no link to the apps page. You have to know to go to the BOINC page to then go to the apps page. (An aside: I really don't like the non-BOINC front end to the site. It makes many of the "list" pages (stats, tasks, etc) way too narrow. I prefer just the normal BOINC format.

You have two buttons at the bottom here in the BOINC side. One going Home (to Drupal), one to Project (BOINC). Click project and the applications link is easily available.

Edit: Completely aside anything, but DD is testing Drupal. BOINC eventually wants to do away with the old style forums and such and go totally Drupal. Best be ready. ;-)
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 638 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Jack Shultz
Avatar

Send message
Joined: 10 Apr 09
Posts: 503
Credit: 120,150
RAC: 0
Message 642 - Posted: 19 Jun 2009, 22:26:37 UTC - in response to Message 638.  

That last gpu release was an accident. Sorry...GPU needs more work.
ID: 642 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Gerry Rough
Avatar

Send message
Joined: 1 May 09
Posts: 19
Credit: 91,491
RAC: 0
Message 645 - Posted: 19 Jun 2009, 23:43:33 UTC - in response to Message 642.  

That last gpu release was an accident. Sorry...GPU needs more work.


It sounds to me like the programmer needs less work!! :-)

(Click for detailed stats)
ID: 645 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile nenym

Send message
Joined: 23 Apr 09
Posts: 99
Credit: 530,784
RAC: 534
Message 651 - Posted: 22 Jun 2009, 17:09:33 UTC

Messages from server:
22/06/2009 19:06:10 DrugDiscovery Scheduler request completed: got 0 new tasks
22/06/2009 19:06:10 DrugDiscovery Message from server: No work sent
22/06/2009 19:06:10 DrugDiscovery Message from server: GROMACS with Nvidia GPU is not available for your type of computer.

Both GTX260 (host ID: 169) and 9600GT (host ID: 364).
ID: 651 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Gerry Rough
Avatar

Send message
Joined: 1 May 09
Posts: 19
Credit: 91,491
RAC: 0
Message 652 - Posted: 22 Jun 2009, 17:39:59 UTC

Actually, the prior posts may be incorrect in the assumption that BOINC is asking for GPU work. I think this may not be the case. Here is what I am getting in my messages tab. Note the first lines are for CPU work fetch:


6/22/2009 12:27:00 PM DrugDiscovery Sending scheduler request: To fetch work.
6/22/2009 12:27:00 PM DrugDiscovery Requesting new tasks for CPU
6/22/2009 12:27:05 PM DrugDiscovery Scheduler request completed: got 0 new tasks
6/22/2009 12:27:05 PM DrugDiscovery Message from server: No work sent
6/22/2009 12:27:05 PM DrugDiscovery Message from server: GROMACS with Nvidia GPU is not available for your type of computer.


This is what the prior posts were addressing. The messages were for CPU work, not GPU work. I am still not sure why this message is sent since I also have a 9800 GT GPU. So we still have a problem with GPU work not sent to GPU enabled hosts either way.


There is also another another related issue that is popping up. Not sure what this means:


6/22/2009 10:00:47 AM DrugDiscovery Sending scheduler request: To fetch work.
6/22/2009 10:00:47 AM DrugDiscovery Requesting new tasks for CPU
6/22/2009 10:00:52 AM DrugDiscovery Scheduler request completed: got 0 new tasks
6/22/2009 10:00:52 AM DrugDiscovery Message from server: No work sent
6/22/2009 10:00:52 AM DrugDiscovery Message from server: GROMACS with Nvidia GPU is not available for your type of computer.
6/22/2009 10:00:52 AM DrugDiscovery Message from server: (won't finish in time) BOINC runs 99.9% of time, computation enabled 100.0% of that

(Click for detailed stats)
ID: 652 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile nenym

Send message
Joined: 23 Apr 09
Posts: 99
Credit: 530,784
RAC: 534
Message 654 - Posted: 22 Jun 2009, 19:11:10 UTC - in response to Message 653.  
Last modified: 22 Jun 2009, 19:15:23 UTC

Full message (6.6.36 client):
22.6.2009 21:02:39 DrugDiscovery Sending scheduler request: To fetch work.
22.6.2009 21:02:39 DrugDiscovery Requesting new tasks for GPU
22.6.2009 21:02:44 DrugDiscovery Scheduler request completed: got 0 new tasks
22.6.2009 21:02:44 DrugDiscovery Message from server: No work sent
22.6.2009 21:02:44 DrugDiscovery Message from server: GROMACS with Nvidia GPU is not available for your type of computer.

if set <wfd> flag
22.6.2009 21:14:20 DrugDiscovery [wfd] request: CPU (0.00 sec, 0) CUDA (86400.86 sec, 1)
22.6.2009 21:14:20 DrugDiscovery Sending scheduler request: To fetch work.
22.6.2009 21:14:20 DrugDiscovery Requesting new tasks for GPU
22.6.2009 21:14:26 DrugDiscovery Scheduler request completed: got 0 new tasks
22.6.2009 21:14:26 DrugDiscovery Message from server: No work sent
22.6.2009 21:14:26 DrugDiscovery Message from server: GROMACS with Nvidia GPU is not available for your type of computer.
22.6.2009 21:14:26 DrugDiscovery [wfd] backing off CUDA 718 sec
22.6.2009 21:14:26 [work_fetch_debug] Request work fetch: RPC complete
ID: 654 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Gerry Rough
Avatar

Send message
Joined: 1 May 09
Posts: 19
Credit: 91,491
RAC: 0
Message 655 - Posted: 22 Jun 2009, 19:48:40 UTC - in response to Message 653.  

The second message typically indicates that one of the time metrics may have gotten whacked. In this case, I'd say take a look at your indicated CPU efficiency since the two shown in the snippet seem to be OK.

Alinator



Not sure which lines to look at for that host here at DD@h. Here are the three that might fit your query:


% of time BOINC client is running 99.9138 %
While BOINC running, % of time host has an Internet connection 98.9259 %
While BOINC running, % of time work is allowed 99.9897 %


I have all of my hosts set at the following on my BAM! account:

On a multiprocessor system, use at most 4 processors
On a multiprocessor system, use at most 100% of the processors
Use at most 100% of CPU time

(Click for detailed stats)
ID: 655 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 657 - Posted: 22 Jun 2009, 23:23:18 UTC - in response to Message 655.  

Not sure which lines to look at for that host here at DD@h.

The duration correction factor.
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 657 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Gerry Rough
Avatar

Send message
Joined: 1 May 09
Posts: 19
Credit: 91,491
RAC: 0
Message 658 - Posted: 22 Jun 2009, 23:28:15 UTC - in response to Message 657.  

Not sure which lines to look at for that host here at DD@h.

The duration correction factor.


I don't think that is what he meant. Nonetheless, here is the requested DCF for that host:

Task duration correction factor 1.202254

(Click for detailed stats)
ID: 658 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile nenym

Send message
Joined: 23 Apr 09
Posts: 99
Credit: 530,784
RAC: 534
Message 660 - Posted: 23 Jun 2009, 9:34:06 UTC

I can't download any work for CPU and also for GPU. I have no LTD/STD/DCF problem, no high priority run. Clients at host without CUDA ask for CPU work, clients at hosts with CUDA ask for both CPU/GPU work. Still the same msg from server "GROMACS with Nvidia GPU is not available for your type of computer". With CUDA I run successfully GPUGRID, AQUA, Seti@beta "VLAR" and did run Lattice GPU tasks...
BTW the server batch is CPU or GPU work?
Is the problem in my settings or computers? What can I try more to get work? Any idea what to do? I feel as newbie :-((.
ID: 660 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 661 - Posted: 23 Jun 2009, 10:10:17 UTC - in response to Message 660.  

BTW the server batch is CPU or GPU work?

Checking the server... it seems it's neither. ;-)
None of the work available seems to have gone out yet, while there's nothing I can do about that.
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 661 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 662 - Posted: 23 Jun 2009, 10:12:21 UTC - in response to Message 659.  

At least that was how it used to work until they started screwing it all up with 6.4.x generation backends. ;-)

Yeah me too, until I got hit by similar messages on Primegrid, DD and H2, where my TDCF was above 5. I had to lower it to below 1 (0.92 to be exact) to get work.
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 662 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 671 - Posted: 23 Jun 2009, 19:38:52 UTC - in response to Message 662.  

Well, I was now hit with that one as well, on one of my computers:

23-Jun-09 21:36:02 DrugDiscovery Message from server: GROMACS with Nvidia GPU is not available for your type of computer.
23-Jun-09 21:36:02 DrugDiscovery Message from server: (won't finish in time) BOINC runs 98.3% of time, computation enabled 100.0% of that

I reset the project and that 'fixed' it. It's downloading work again.
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 671 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 672 - Posted: 23 Jun 2009, 19:42:58 UTC - in response to Message 671.  

I think it has to do with the 100% ... it's an impossibility that Boinc runs 100% of the time. It is normally always under this, if only by 0.1% ... I'll check in the code what goes on when the second number is 100%.
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 672 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Gerry Rough
Avatar

Send message
Joined: 1 May 09
Posts: 19
Credit: 91,491
RAC: 0
Message 673 - Posted: 23 Jun 2009, 21:39:58 UTC - in response to Message 672.  

I think it has to do with the 100% ... it's an impossibility that Boinc runs 100% of the time. It is normally always under this, if only by 0.1% ... I'll check in the code what goes on when the second number is 100%.


I just tried resetting the project. I am now getting work: I received one WU from the server.

(Click for detailed stats)
ID: 673 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Ageless
Avatar

Send message
Joined: 11 Apr 09
Posts: 172
Credit: 7,631
RAC: 0
Message 674 - Posted: 24 Jun 2009, 0:18:31 UTC - in response to Message 673.  

I sent it off to the Boinc_Dev email list. More and more people are now reporting the same issue on other projects. And all report the 100% computation time. Gotta be a bug, either in the client or the scheduler. I'll keep you updated.
Jord

'Cause you seem like an orchard of mines, Just take one step at a time.
ID: 674 · Rating: 0 · rate: Rate + / Rate - Report as offensive
Profile Tim Turner
Avatar

Send message
Joined: 1 May 09
Posts: 570
Credit: 184,322
RAC: 0
Message 675 - Posted: 24 Jun 2009, 1:21:04 UTC - in response to Message 674.  

both!
Tim Turner
Public Relations Admin
Secunia PSI: http://secunia.com/vulnerability_scanning/personal/
If you need help via voice or Convo; PM me and i will give you details on where i will be; Teamspeak, Yahoo Messenger, or Skype.
ID: 675 · Rating: 0 · rate: Rate + / Rate - Report as offensive
1 · 2 · 3 · Next

Message boards : Number crunching : GPU GROMACS


©2017 All rights reserved | Design by Digital BioPharm Ltd