XANSONS Project Update

[JugNut]

For anyone interested I whipped up an app_config for controlling the MT app.  It's almost identical to Multithreaded(MT) ones posted before by kashi & Dingo for other projects,  but with a slightly different plan class and different app name of course.  Tested and working...

Change Nthreads & avg_ncpu's  to change number of thread/cores you would like the app to use.

<app_config>
<app>
  <name>xansons_cpu</name>
  <max_concurrent>1</max_concurrent>
</app>
<app_version>
  <app_name>xansons_cpu</app_name>
  <plan_class>mt_windows</plan_class>
  <cmdline>--nthreads 8</cmdline>
  <avg_ncpus>8.000000</avg_ncpus>
</app_version>
</app_config>


For GPU's. 
To change the amount of running GPU apps just change this section.   <gpu_usage>1</gpu_usage> Change it as per the line below.
1 = 1,  0.5 = 2,  0.33 = 3,   0.25 = 4 ect.  EG changing it to say 0.5 will get you 2 work units running in parallel on the same GPU at the same time.

<app_config>
<app>
<name>xansons_gpu</name>
<gpu_versions>
<gpu_usage>1</gpu_usage> 
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>


For the more adventurous this is my full app_config that I started using tonight.  Since I have both an Nvidia & an ATI card in the same box I must use an app_config written like this to control each card separately.  On this test box the Nvidia card runs much faster than the ATI card with the GPU app so i'm running one instance on the ATI card and two instance(0.5) at a time on Nvidia card.  Will try 3(0.33) soon.

<app_config>

<app>
  <name>xansons_cpu</name>
  <max_concurrent>1</max_concurrent>
</app>
<app_version>
  <app_name>xansons_cpu</app_name>
  <plan_class>mt_windows</plan_class>
  <cmdline>--nthreads 8</cmdline>
  <avg_ncpus>8.000000</avg_ncpus>
</app_version>

<app_version>
    <app_name>xansons_gpu</app_name>
    <plan_class>cuda65_windows</plan_class>
    <avg_ncpus>1.0</avg_ncpus>
    <ngpus>0.5</ngpus>
  </app_version>

  <app_version>
     <app_name>xansons_gpu</app_name>
     <plan_class>opencl_ati_102_windows</plan_class>
     <avg_ncpus>1.0</avg_ncpus>
     <ngpus>1.0</ngpus>
</app_version>

</app_config>

For anyone who's not familiar with app_config's this is a brief how to. Still not sure? Just ask.
There's no need to shut down boinc to do this..

Say if you would like to control the Xansons multi threaded CPU app,  then just copy out the first section of this post starting at(and including) the line <app_config> down to(and including) the line </app_config>  And everything in between of course.

Then fire up notepad or any other text only editor and paste it in. Then while still in notepad click the file menu & choose "save as" then navigate to the folder C:\ProgramData\BOINC\projects\xansons4cod.com_xansons4cod\ and save the resulting text file there naming it app_config.xml
Alternatively you could save the app_config.xml file anywhere then cut & paste it to the above mentioned Xansons folder.

Then just go back to boinc manager & click on the toolbar menu called "Options" then just click on "read config files" and your done.  It will either take effect within the next 10 sec's or depending on a few thing it may kick in on the next work unit.  If you haven't noticed any changed by then, then somethings gone wrong.  Some of the usual things that can go wrong are....  you've placed the app_config in the wrong directory or there's been a syntax error.  In that case the most usual things are that when you cut & pasted there was a letter or word that was missed.  Or LOL the silly bugger(in this case me) who posted the app_config did the same thing and there's a word or letter missing...  Again just ask. 
But it will either work or it will not work so don't worry about buggering something up.

Beginner tips:  Make sure when you save your app_config.xml that it's extension ends in .xml and NOT for example something like this... app_config.xml.txt  This happens most often with people that still have their windows in a default state.   By default windows comes with the file extensions hidden,   similarly also if you can't find C:\ProgramData\  folder then it's hidden too,  again googles your friend or just post and ask for help.

PS Credit is still quite poor but I got my fingers crossed it will improve in the future

PSS LOL..  I could have really saved myself some major time & effort if i'd of just searched the Xansons forums first, instead of spending hours re-inventing the wheel all over again. Sheesh!

[Dingo]

Good info Jugnut, are you running two tasks on any of your GPU's.  I only have a AMD HD7790 and when running one task GPU-Z says it is running at 100% but it is an old card nowadays.

[JugNut]

Yes mate i'm running 4 at once on the big guy(1080 ti) most of the WU's are tiny and were completing in mere seconds. Running 4 at a time has slowed em' down enough that now there is only the occasional stutter on the odd occasion they all complete at once. My AMD card runs much slower with these xansons GPU app and there seems only a small advantage if any to running more than 1 at a time on the AMD cards.  But it's early days anything could happen.

I'm having a troublesome problem on a system of mine that has 2 GTX 970's.  On that PC the xansons cuda app only crunch's on one GPU.  If I look in boinc manager it appears as both WU's are crunching normally.  They also appear that they have both been properly assigned by boinc to GPU-0 & GPU-1.  But in reality the 2 WU's are crunching on just one GPU and that's without the use of any app_configs,  just as default.  This can be seen by running afterburner or some other GPU monitoring program,  you'll see just one GPU has full load the other has little to none.

After finding out what was wrong I then tried all manner of configs but so far nothing has worked. It's very much like what happened on POEM@home a few years back,  it took them ages to fix that problem then.  I hope it wont be the case this time.

So if anyone gets two identical Nvidia GPU's to work properly please let me know.  I mean it could be a problem on just this PC but I kinda doubt it as if I stop xansons and then start 2 Einstein tasks they run properly. If it's just my PC it's one heck of an weird problem that only affects xansons apps.  But anythings possible..

Also the credit is not quite as bad as I first thought either,  so there's that.  Plus it's always good having fresh projects to crunch especially ones that have GPU apps too.

Crunch on 


PS: I reported the problem i'm having on the xansons forum,  so we'll see what happens?

[JugNut]

The glitch in the xansons GPU app is now fixed.  The latest release app v1.02 fixes the bug that apparently all Multi Nvidia card users were having but few even noticed.  Unless you have a GPU monitoring app open all the time most would never even realize that only one GPU was actively crunching all running tasks,  no matter how many GPU's they had.

Phew i'm glad thats over,  the admin had me almost convinced it was my problem. 

[kashi]

Glad it's fixed now.

I had lots of trouble with various projects years ago when running multiple GPUs in the one box. Had very similar symptoms at one stage; one GPU core idle and the other running 2 tasks. Plus also sometimes phantom tasks running. It all needed lots of fiddling with dummy plugs and other workarounds but still gave trouble on some projects.

If you are one one of the few with that particular mix of hardware at a project it's sometimes hard to get an acknowledgement of a problem. If you could deal with someone like Cluster Physik or Slicker it was OK because they knew their way around GPU applications, but some of the other project admins were struggling with how GPU crunching worked.

Although that was at a time when GPU crunching was relatively new to BOINC and even the ATI CAL/Brook and Nvidia CUDA software was at an early stage of its evolution.

Luckily looks like Vlad knows his onions so it's great he was able to fix it.

Also GPU related, I just restarted GPUGrid and wondered why I couldn't get any tasks. Turns out a new application uses CUDA 8 and requires a recent Nvidia driver. Updated driver and am now cooking with gas. So that was easy once I checked the forums and worked it out. Had forgotten to add SWAN_SYNC so did that for my 2nd task this morning. Really should use Linux with GPUGrid but VM won't do, need to actually boot into it for GPU support, so keep putting it off.

[JugNut]

Hey kashi how are you?
I've been wondering when you'd chime in mate?  Of course it's only been a few days since xansons restarted but your usually the first one in the team with boots on the ground when it comes to trying new apps and especially when it comes to trying new GPU apps,  so I was starting to get a little worried when you didn't chime in straight away.  After all one of the most interesting things about boinc is trying new app's out and figuring out how they work work and seeing what can be done to optimise them further.  As we know anyone can attach to a project and walk away,  but what fun would that be.  Although I know i'm preaching to the choir about that eh kashi? LOL
Anyway have you had a chance to have a crack at some yet?

In many ways these xansons GPU tasks are similar to milky way tasks with all the hundreds of little work units being crunched in a day. It utilises the GPU pretty well though but because of all the stopping & starting there's a fair bit a wasted time that can be mopped up by running a few at a time with an app_config.  Even if there is little to no gain from better utilisation eliminating all the wasted time should give you something for nothing anyway. 
Although I found it quite difficult to judge how much gain was being made because of the variation in task sizes. While there all comparatively small there is enough variation in their run time to make it difficult to judge.

I'll have to try it each way over a couple of days or more to be sure of what gain is involved.  Of course it's probably like all GPU tasks in that it will be heavily GPU dependant,  so everyone will have to test it for themselves to see how well it's working for them.  But that's the fun part isn't it? Squeezing out the last drop of power. LOL

Xansons CPU app is another multi-threaded jobbie but it works OK.  Really the CPU work units are small enough that they don't need to be multi threaded but oh well? it is what it is.  Maybe he'll change that in the future?  He seems a pretty clever cookie, so he might?

Lots of good crunching at the moment, right now I got my poker in few to many fires and will soon have to make a choice of what to crunch.

I really like these Citizen science Grid apps, for anyone else they are well worth a try if your keen on the science involved. Kashi & me have been carving them up for a while now, and boy!! don't they give out massively huge credit's?  Not complaining mind you,  quite the opposite :)
I just checked and yesterday I received 846K,  but today I got over 1.3 million credits for Citizen Science Grid.
A million credit a day from a cpu only project is unheard of,  at least for anyone without 1000+ cores at their disposal and it's even more unheard of for me.
LOL It's kinda like BU but for CPU work. But the good thing is this work has some actual real life value to it   

What about SWAN_SYNC,  do you think  makes any real difference these days? Earlier on it did seem to make a big difference but in latter years GPUgrids app's or maybe newer drivers mean't I saw little if any benefits from having it installed or not.

Anyway gotta go. 

Crunch em' if you got em'

[kashi]

Yes just about to give Xansons a little go. Had to do at least a token amount of GPUGrid first as I've been neglecting it and apparently they've been heavily shedding contributors. Possibly the very long GPU tasks which means slower cards no longer get the full bonus credit puts more people off than they realise. My GTX 970 barely makes it as it takes about 20 hours to complete the longer tasks. If I had a slower card and couldn't complete within the 24 hour period it would likely prevent me from contributing there as it would imply my GPU contribution was processed too slowly to be fully useful/appreciated. Actually GPU tasks running for 20 hours or longer is a bit much for many regardless of the credit rate.

Not sure if SWAN_SYNC still works but after adding it the first task of the same ADRIA 539,100 credit type completed 3.7% faster. So looks like it may still make a small difference.

Haha, yep CSG on CPU is still giving higher credit than many GPU applications. Travis responded a few days ago to the person complaining about it being too high and asked whether or not it should be reduced. Nobody has responded to him yet, so it's still make hay while the sun shines.

On the subject of credit, recently did a few more Sourcefinder Duchamp tasks and oh boy did they pay low, even lower than previously. CreditNew? Probably. Hopefully eventually Sam will become better informed and realise that Duchamp credit rate is currently very poor. Maybe Kevin from POGS could whisper in his ear and get him to set it to be about the same as POGS. That would help attract more contributors which they may need if they soon start to analyse oodles of ASKAP observation data. Hmm maybe not, just had a look and Kevin hasn't posted at POGS since February, hope he's OK.

[kashi]

Haha, mozzed myself at Xansons by mentioning GPUGrid tasks running for so long. Going to the other extreme, Xanson tasks are so short and cache so small I keep running out. Looks like server can't keep up, it's like ancient history of MilkyWay server overload and task shortage when they also had ridiculously short tasks. Gave it a go as promised, will leave it running overnight and see if it keeps mucking up. At this stage looks like I may be heading back to GPUGrid soon.

The tasks run without problems at least. Like you I'm running 2 concurrently on my GTX 970 to avoid excessive thermal cycling. Thanks for app_config to do so.

Double jeopardy in mozz stakes, also mentioned Citizen Science Grid and they ran out of work too. Went to switch to DENIS but alas their server has gone kaput. Grr, this must be infectious, often happens on weekends too when project admins aren't available to fix things. Have settled on POGS on dual box and about to try Universe on Skylake. Edit: Aaaargh, Universe out of work too, this is getting ridiculous.

Hmm, excuse double post, it's a bit like talking to yourself, innit?  

[Dataman]

On the subject of credit, recently did a few more Sourcefinder Duchamp tasks and oh boy did they pay low, even lower than previously. CreditNew? Probably. Hopefully eventually Sam will become better informed and realise that Duchamp credit rate is currently very poor. Maybe Kevin from POGS could whisper in his ear and get him to set it to be about the same as POGS. That would help attract more contributors which they may need if they soon start to analyse oodles of ASKAP observation data. Hmm maybe not, just had a look and Kevin hasn't posted at POGS since February, hope he's OK.

I have been thinking much the same. Unfortunately, many scientists just treat crunchers as if they are grad students ... AKA "slaves". They might be brilliant in their field but complete imbeciles when it comes to marketing.   We are "volunteers" not indentured servants!

Crunchers are leaving GPUGrid in droves AND the admin's seem to have a deaf ear to the comments of their most loyal supporters. Go figure? I, for one, am not going to tie up a GPU for the better part of a day to run a wu.

I am a bit peeved at the lack of communication from POGS as well. I have not run any Sourcefinder yet but hope to do some when weather permits. Still too hot here to run much so I plug away at Asteroids, Collatz and NFS. Speaking of Asteroids, I am listening to a lecture series on the search for exoplanets and a whole lecture is devoted to shape and spin of planetary objects. Ties in nicely to what they are doing at the Asteroid project.

https://www.audible.com/pd/Science-Technology/The-Search-for-Exoplanets-What-Astronomers-Know-Audiobook/B0159LVSNC

Well, enough griping for a Saturday morning. Cheers and crunch on.

[kashi]

Well he increased the maximum cache to 50 which seems to be working nicely to prevent running out of work, so that's much better.

Also changed the application to use less CPU. Don't think that resulted in any increase in speed, well none I noticed anyway, but it encouraged me to change my app_config.xml to use only half a CPU per GPU task instead of a full CPU. This allows an extra CPU task to run. In other words changed cpu_usage parameter from 1 to 0.5. Like this:

<app_config>
<app>
<name>xansons_gpu</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

Looks like it's granting about 400k per day for my GTX 970. Not as generous as some others but better than my earlier estimates. So that's good too.

Will do a few more days here, a week or so at GPUGrid and then cease GPU crunching until next AA.

Tariff is now 31.372 c/kWh for all usage, which is 29.6% higher than the previous marginal rate. I know nobody's chosen to mention this subject until now but it's not going away. Geez, I'm rapidly falling further behind the 8 ball here, now I need a Tesla Powerwall too and haven't even got any solar panels to connect to it.
"We'll all be rooned," said Hanrahan, "if this 'ere corruption don't cease."

[Dataman]

Well he increased the maximum cache to 50 which seems to be working nicely to prevent running out of work, so that's much better.

Also changed the application to use less CPU. Don't think that resulted in any increase in speed, well none I noticed anyway, but it encouraged me to change my app_config.xml to use only half a CPU per GPU task instead of a full CPU. This allows an extra CPU task to run. In other words changed cpu_usage parameter from 1 to 0.5. Like this:

<app_config>
<app>
<name>xansons_gpu</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

Looks like it's granting about 400k per day for my GTX 970. Not as generous as some others but better than my earlier estimates. So that's good too.

Will do a few more days here, a week or so at GPUGrid and then cease GPU crunching until next AA.

Tariff is now 31.372 c/kWh for all usage, which is 29.6% higher than the previous marginal rate. I know nobody's chosen to mention this subject until now but it's not going away. Geez, I'm rapidly falling further behind the 8 ball here, now I need a Tesla Powerwall too and haven't even got any solar panels to connect to it.
"We'll all be rooned," said Hanrahan, "if this 'ere corruption don't cease."

Good on you kashi. 
Put my whole farm on it today despite it being a Russian project. What the hell ...  

[tazzduke]

Hooty Tooty Y'all

Gave it a run, got me a million and now gone back to Seti,

Yep the prices of power have gone up here as well, once I track down another 980, I think two of those switching between Seti and Einstein will do me fine.  Me 960 and 950, who knows if I will still have them.

Times change, yet the more things change, the more they stay the same lo.

Cheers

[tazzduke]

Greetings All

As per the following post, project is wrapped up.

http://xansons4cod.com/xansons4cod/forum_thread.php?id=91

Glad I got to at least 5 million on it lol.

Cheers

[JugNut]

Thanks for the info tazzduke.
I'm sad to hear that as I really enjoyed crunching Xansons and was planning on giving them another wack in the near future.
As far as admins go Vlad was one of the better ones, very responsive and pro active and was always ready to help or to add new features to the site if asked.  Which unlike many newer project that once they are up and running completely ignore their crunchers.  I gladly crunch for this admin again in the future if he ever had another boinc project. 

Bon Voyage Xansons 

[Daniel]

XANSONS for COD has published a paper.

You can read it here ...
https://www.degruyter.com/view/j/eng.2018.8.issue-1/eng-2018-0014/eng-2018-0014.xml