Let us know how they're working here.

---

Ok, I got twelve MilkyWay@Home N-Body Simulation v0.03 workunits for the x86_64 platform. The application file size is approx. 4.5 MB and it is a 32bit executable. Runtime was very short, between 0.5 and 2.5 minutes on a 3,2 GHz Phenom II X4. No screensaver / graphics included apparently.

Strange thing: No upload seen, but finished without error acc. to BOINC manager. Maybe very small result files?

All twelve are now listed as pending with status 'Completed, validation inconclusive'.

EDIT: Two results are now validated! Fine.

Computer ID=152604
List of Tasks

That's it for now. Comments?

Ok, I got twelve MilkyWay@Home N-Body Simulation v0.03 workunits for the x86_64 platform. The application file size is approx. 4.5 MB and it is a 32bit executable. Runtime was very short, between 0.5 and 2.5 minutes on a 3,2 GHz Phenom II X4. No screensaver / graphics included apparently.

Strange thing: No upload seen, but finished without error acc. to BOINC manager. Maybe very small result files?

All twelve are now listed as pending with status 'Completed, validation inconclusive'.

EDIT: Two results are now validated! Fine.

Computer ID=152604
List of Tasks

That's it for now. Comments?

There's no upload because we're passing the fitness back to the server in the standard error, which gets put right into the database. This saves the server having to download and read a file for every result, which should really reduce the server load.

The validation inconclusive is for the same reason as the separation code, we got your result and are waiting for it to be verified against another result before we insert it into our population of valid results.

The runtimes are short for now just because we want to test how well the search can find the correct solution, and see if any issues come up for any of the different binaries (without people having to lose credit for running something for a couple hours only for it to result in a bug and them not getting credit).

Once we're a bit satisfied as to how things are running and the current search has come up with a good solution, we'll start to ramp up the workunit times.

---

Thanks for the reply, Travis. And understood. :-)

Two workunits with errors here and also on the other hosts, that got them so far:

de_nbody_test_10_9674_1282101611
de_nbody_test_10_10314_1282103078

Btw, if anyone has a working app_info.xml which excludes the milkyway 0.19 workunits from being downloaded, it would be appreciated, if you would share it here. ;-)

try this one - its for 64bit though

its opp apps for both cpu and gpu

if what you are after is gPU only then set that in you account preferences or delete out the cpu refernce block below - but boinc will show errors if you do

<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>astronomy_0.21_x64_SSE3.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway</app_name>
<version_num>21</version_num>
<file_ref>
<file_name>astronomy_0.21_x64_SSE3.exe</file_name>
<main_program/>
</file_ref>
</app_version>

<app_version>
<app_name>milkyway</app_name>
<version_num>20</version_num>
<file_ref>
<file_name>astronomy_0.21_x64_SSE3.exe</file_name>
<main_program/>
</file_ref>
</app_version>

<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_0.24_windows_intelx86__cuda23.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart.dll</name>
<executable/>
</file_info>
<file_info>
<name>cutil32.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>milkyway</app_name>
<version_num>24</version_num>
<plan_class>cuda23</plan_class>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.040000</max_ncpus>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>milkyway_0.24_windows_intelx86__cuda23.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cutil32.dll</file_name>
</file_ref> 
</app_version>

</app_info>

thx a lot

I'd like to have only CPU and only the new nbody sim not the former app.

I see that the older CPU app - 0.21 SSE3 in your case, stock 0.19 in mine - does not need any other file references than to the executable itself.

My first try was as follows, but it failed to do the job:

<app_info>
 <app>
 <name>milkyway</name>
 </app>
 <file_info>
  <name>milkyway_nbody_0.03_windows_x86_64.exe</name>
  <executable/>
 </file_info>
 <app_version>
  <app_name>milkyway</app_name>
  <version_num>3</version_num>
  <file_ref>
   <file_name>milkyway_nbody_0.03_windows_x86_64.exe</file_name>
   <main_program/>
  </file_ref>
 </app_version>
</app_info>

I'm not sure about the content of the <name> and <app_name> tags and that of the <version_num> tags.

And the first download included some more files

orphan_test_4096_bodies.js
histogram_4096_bodies
p-12-3s-free.txt
p-14-3s-free.txt

Is it necessary to include a reference to some of those as well?

Two workunits with errors here and also on the other hosts, that got them so far:

de_nbody_test_10_9674_1282101611
de_nbody_test_10_10314_1282103078

Btw, if anyone has a working app_info.xml which excludes the milkyway 0.19 workunits from being downloaded, it would be appreciated, if you would share it here. ;-)

Hello Alexander, i tried a few different app_info.xml's myself and i've come up with this one:

<app_info>
<app>
<name>milkyway_nbody</name>
</app>
<file_info>
<name>milkyway_nbody_0.03_windows_x86_64.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway_nbody</app_name>
<version_num>3</version_num>
<file_ref>
<file_name>milkyway_nbody_0.03_windows_x86_64.exe</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>

Let me know if this works for you

* I have a bug to report as well, even the X86-64 version runs in 32bit mode on my machine.

* I have a bug to report as well, even the X86-64 version runs in 32bit mode on my machine.

There isn't a 64 bit binary yet. Windows is a pain, and I need to do some work to get an actual 64 bit version. It will be 64 bit eventually; it's quite a bit faster for this.

* I have a bug to report as well, even the X86-64 version runs in 32bit mode on my machine.

There isn't a 64 bit binary yet. Windows is a pain, and I need to do some work to get an actual 64 bit version. It will be 64 bit eventually; it's quite a bit faster for this.

Hm, then why do we have two apps for Windos now?
It's a bit misleading if one is called x86_64 when it's in fact a 32-bit app... ;-)

---

Lovely greetings, Cori

Hello Alexander, i tried a few different app_info.xml's myself and i've come up with this one:

<app_info>
<app>
<name>milkyway_nbody</name>
</app>
<file_info>
<name>milkyway_nbody_0.03_windows_x86_64.exe</name>
<executable/>
</file_info>

<app_version>
<app_name>milkyway_nbody</app_name>
<version_num>3</version_num>
<file_ref>
<file_name>milkyway_nbody_0.03_windows_x86_64.exe</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>

Let me know if this works for you

Hi Arjan,
did you try this with yout ati-machine? If yes, did your gpu continue to crunch?

I tried it right now and all my gpu-wu's failed immediate. Since the last code-update (HD58xx-series) there is no longer a app_info.xml included. So I copied the last distributed one to the work-directory, changed the names, copied your code into it and restarted boinc. Crashed in the moment of starting it.

Alexander

There isn't a 64 bit binary yet. Windows is a pain, and I need to do some work to get an actual 64 bit version. It will be 64 bit eventually; it's quite a bit faster for this.

Hm, then why do we have two apps for Windos now?
It's a bit misleading if one is called x86_64 when it's in fact a 32-bit app... ;-)

Well, that's not too unusual. Quite a number of projects send 32bit apps to the x86_64 system, for example Rosetta, LHC, Docking, RCN, Spinhenge...

I wouldn't bother if I get a 32bit app as something_x86.exe or something_x86_64.exe. ;-)

However, the possibility to have a real 64bit app later with some performance gain, is promising. :-)

Hello Alexander, i tried a few different app_info.xml's myself and i've come up with this one:

(...)

I tried it right now and all my gpu-wu's failed immediate. Since the last code-update (HD58xx-series) there is no longer a app_info.xml included. So I copied the last distributed one to the work-directory, changed the names, copied your code into it and restarted boinc. Crashed in the moment of starting it.

Alexander,

if you want to crunch on your ATI and on the CPU you have to build an app_info.xml which includes both sections. Take a look an the xml which was posted by Tim Norton here. It was, however, for the 'old' opt. CPU app and for a NVIDIA GPU, so yours would look different. You have to combine Arjans CPU part and another part for the ATI. Can't help you on the last, cause I have only a single-precision ATI GPU, but I'm sure others can.

P.S.: Arjan, thanks so far, I'm coming back after I've tested it. Have to empty my cache first a bit, have tons of SIMAP and some QuantumFire burning. :-)

EDIT:

So I copied the last distributed one to the work-directory, changed the names, copied your code into it and restarted boinc.

Uhmm.. ok, I should read better, sorry, my bad. Maybe you should try first to get only the GPU part working, might be easier to find an error that way.

EDIT:

Hello Arjan,

your app_info.xml works like a charm! Great, thanks a lot.

No i didn't try this on my machine with the ATI card but rather on my laptop, i will try tomorrow to get it working on my core i7 with the ATI card as well.

Still some errors from time to time:

de_nbody_test_10_14620_1282140181
de_nbody_test_10_14642_1282140181
de_nbody_test_10_14717_1282140181
de_nbody_test_10_28566_1282168828
de_nbody_test_10_28643_1282168828

I noticed that the runtimes of the error results are quite similar on my box, all around 932 seconds. And I got none pending or valid which ran longer. Some kind of timeout?

If there's anyone interested here is the nbody simulation on CPU combined with the ATI GPU program:

<app_info>
<app>
    <name>milkyway</name>
    <user_friendly_name>MilkyWay@Home</user_friendly_name>
</app>
<file_info>
    <name>milkyway_0.23_windows_x86_64__ati13ati.exe</name>
    <executable/>    
</file_info>
<file_info>
    <name>brook64a_ati.dll</name>    
    <executable/>    
</file_info>
<app_version>
    <app_name>milkyway</app_name>
    <version_num>23</version_num>
    <file_ref>
       <file_name>milkyway_0.23_windows_x86_64__ati13ati.exe</file_name>
       <main_program/>
    </file_ref>
    <file_ref>
       <file_name>brook64a_ati.dll</file_name>
       <open_name>brook64.dll</open_name>
<copy_file/>
    </file_ref>
    <platform>windows_x86_64</platform>
    <plan_class>ati13ati</plan_class>
    <avg_ncpus>0.050000</avg_ncpus>
    <max_ncpus>0.100000</max_ncpus>
    <coproc>
        <type>ATI</type>
        <count>1.000000</count>
    </coproc>
</app_version>

<app>
    <name>milkyway_nbody</name>
    <user_friendly_name>MilkyWay@Home nbody Simulation</user_friendly_name>
</app>
<file_info>
    <name>milkyway_nbody_0.03_windows_x86_64.exe</name>
    <executable/>
</file_info>
<app_version>
    <app_name>milkyway_nbody</app_name>
    <version_num>3</version_num>
    <file_ref>
        <file_name>milkyway_nbody_0.03_windows_x86_64.exe</file_name>
        <main_program/>
    </file_ref>
    <platform>windows_x86_64</platform>
</app_version>
</app_info>

GPU processing is slowed down a bit though, about 20% slower.

For comparison I started a test on an old system of mine. This is a dual socket Athlon XP running Windows 2000 SP4 (32bit of course). This CPU supports SSE but nothing higher. It appears the nbody simulation does not run at all on such a system.

There are two types of errors:

1. The application crashes almost in an instant and progress bar jumps to 100%. See here for example: de_nbody_test_10_31110_1282174646_0


2. The application runs for approx. 20 minutes, elapsed time is increasing, remaining time is decreasing, progress bar is not running at all, CPU load is full. See here for example: de_nbody_test_10_31094_1282174646_0

Don't get me wrong here. It is not a big deal, if you intend not to support such old hardware and/or operating system.

Regards

For comparison I started a test on an old system of mine. This is a dual socket Athlon XP running Windows 2000 SP4 (32bit of course). This CPU supports SSE but nothing higher. It appears the nbody simulation does not run at all on such a system.

There are two types of errors:

1. The application crashes almost in an instant and progress bar jumps to 100%. See here for example: de_nbody_test_10_31110_1282174646_0


2. The application runs for approx. 20 minutes, elapsed time is increasing, remaining time is decreasing, progress bar is not running at all, CPU load is full. See here for example: de_nbody_test_10_31094_1282174646_0

Don't get me wrong here. It is not a big deal, if you intend not to support such old hardware and/or operating system.

Regards

Right now the issues with the progress bar and the maximum time expired should be because the current total FLOPs estimate which BOINC bases the maximum allowed time on is really terrible. It also appears to crash (on Windows at least) after BOINC kills off the long running job. I'm trying to come up with a way to predict it better. The actual time it will take can vary greatly depending on the settings of the simulation. It kind of depends on the positions the bodies end up in at each timestep, so it's kind of hard to predict. If the bodies stay tightly clustered, the simulation tries to be more precise and tends towards O(n^2) behaviour. As more bodies become unbound from the dwarf galaxy model's core and get pulled into the tidal stream, the simulation will tend to speed up. As more bodies get farther apart, more approximations can be made and things run faster. Optimally the simulation should run in O(n log n).

Old hardware that doesn't have SSE2 won't be supported because of precision difficulties related the x87 FPU which I won't go into now. However, currently nothing is stopping it from being sent out to those really old systems (like the Athlon XP) although that should be fixed at some point.

If there's anyone interested here is the nbody simulation on CPU combined with the ATI GPU program:


<avg_ncpus>0.050000</avg_ncpus>
<max_ncpus>0.100000</max_ncpus>


GPU processing is slowed down a bit though, about 20% slower.

Hi Arjan,

many many thanks for your effort, I'm shure, many people will be happy with it!

Two things are left to say:
I compared your app_info with the one I tried to compose. The original one, which was deliverd with the older apps, had the line
<max_ncpus>0.100000</max_ncpus>
different. They used <max_ncpus>1.00000</max_ncpus>
I changed this and found that there is no speed difference now.

The other thing is: MW doesn't download the milkyway_nbody_0.03_windows_x86_64-exe. It permanently reports project delay and Wiederhole in xxxxx.

How do I get this file?

Regards,
Alexander

How do I get this file?

This should do it: Link

---

Starfire