TN-Grid

Jump to latest Follow Reply
B

Beyond

Ars Praefectus
5,236
  • #1
    New project that's been suggested for the DC Vault:

    TN-Grid

    http://gene.disi.unitn.it/test/index.php

    To join TN-Grid: create an account first. You need the invitation code: science@tn
    This project uses BOINC. If you're already running BOINC, select Add Project. If not, download BOINC.
    When prompted, enter: http://gene.disi.unitn.it/test/

    Goals of the project:

    http://gene.disi.unitn.it/test/genehome/en/joinus/5reasons.php

    Also some of the initial research has already been on E. coli, which is most likely aimed at usefulness for medical research:

    http://gene.disi.unitn.it/test/forum_thread.php?id=121

    http://gene.disi.unitn.it/test/forum_thread.php?id=115

    Don't forget join the Ars Technica team! I signed myself and Don H. up today. So far Sprecher Brewery is doing us proud!

    http://gene.disi.unitn.it/test/team_mem ... t&offset=0
     
    F

    frazamatazzle

    Ars Centurion
    205
  • #4
    Don't forget join the Ars Technica team! I signed myself and Don H. up today. So far Sprecher Brewery is doing us proud!
    Click to expand...
    I knew someday, after my disappointment with the bitcoin era of DC computing had passed, and once I stopped competing with Chipotle for low-level Free-DC Mega-Milestone achievements and settled my remaining boxes on this neato project, I knew, someday... it would end up in the Vault and the prophecy would come fulfill itself.
     
    D

    Dave Peachey

    Ars Praetorian
    433
  • #5
    [url=http://arstechnica.com/civis/viewtopic.php?p=32218253#p32218253:3quqtyzl said:
    Beyond[/url]":3quqtyzl]Welcome to the TN-Grid team (and Dave Peachey too)!
    Click to expand...
    I thank you :)

    As of October, I'm back and crunching (seriously) across a number of projects and was looking for some variety (in the sciences arena) so I decided to add my (currently single) boxen to this cause.
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #8
    [url=http://arstechnica.com/civis/viewtopic.php?p=32531837#p32531837:31mpztxf said:
    fractal[/url]":31mpztxf]Be sure to download the custom app from the forum ( http://gene.disi.unitn.it/test/forum_thread.php?id=135 ). Twice as fast. They also have apps for arm / android on the forum.
    Click to expand...
    Thanks much for posting this. It's from 2.55x to 2.95x faster on my various machines.

    Merry Christmas everyone!
     
    D

    Dave Peachey

    Ars Praetorian
    433
  • #9
    Hmm, this is an interesting development; thanks for the "heads up".

    On my slower machine I seem to have gone from around 2hr50 per WU to a little under an hour on the first few WUs crunched exclusively with the new app. I'll leave it running overnight and, assuming I don't have any errors, I'll put it on the faster machine in the morning.

    I hope the WU scheduler can keep up with the faster turnaround once everyone (who can) starts using it!

    Dave
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #11
    fractal said:
    Be sure to download the custom app from the forum ( http://gene.disi.unitn.it/test/forum_thread.php?id=135 ). Twice as fast. They also have apps for arm / android on the forum.
    Click to expand...
    Now there's another new version that's 2x as fast as the previous optimized app. So we've had about a 5x speed increase in a few weeks. DL it here:

    https://bitbucket.org/sirzooro/pc-boinc/downloads

    I tried the newest fma version and it crashed on a machine that worked with the old fma optimized app.
    The new sse2 version worked on all my various CPUs and was over twice as fast as the previous optimized app.

    Edit: Turns out that the new fma app was compiled with incorrect parameters. There's a new build that now works, BUT:

    The new sse2 app is faster than the fma app on my AMD FX-8320E, a reversal from the earlier optimized app.

    Newest optimized sse2: 51:08 to 53:32
    Newest optimized fma: 56:10 to 58:12

    The sse2 app is about 9% faster on this CPU, while the older fma app was faster. Strange...
     
    D

    Dave Peachey

    Ars Praetorian
    433
  • #12
    This is certainly a very dynamic project ... optimised app released > doubling of WU length > optimised app becomes new "official app" > new optimised app released! Now I wonder if the project admins will take this as opportunity to redouble the WU length? :D

    Anyway ... my slower quad core machine, which was averaging just under 2h05m with the latest version of the official app has now sped up to just under 57m. This is using the sse2 version which seems to be the faster one. I need to run down the cache on my bigger 8-core machine (which churns them out in just under 1h12m) before installing it there so I don't (potentially) waste too many WUs.

    I had thought about temporarily stopping TN-Grid when I hit the 1m (in favour of a burst of POGS WUs) but I think I'll just back off slightly instead.
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #13
    Beyond said:
    Edit: Turns out that the new fma app was compiled with incorrect parameters. There's a new build that now works, BUT:

    The new sse2 app is faster than the fma app on my AMD FX-8320E, a reversal from the earlier optimized app.

    Newest optimized sse2: 51:08 to 53:32
    Newest optimized fma: 56:10 to 58:12

    The sse2 app is about 9% faster on this CPU, while the older fma app was faster. Strange...
    Click to expand...
    Another update. There's a new fma+sse2 app (listed as TN-Grid.windows-x86-64-fma-v1.1.zip (windows) at https://bitbucket.org/sirzooro/pc-boinc/downloads).

    It's running around 11% faster than the sse2 version on my four AMD X8 CPUs. Be aware that this new fma app only works on recent CPUs. There's also a Linux build but I haven't tried it.
     
    D

    Dave Peachey

    Ars Praetorian
    433
  • #14
    And yet another application update ... the most recent "optimised" application has now become the latest official version (v0.11) of the application. After a minor teething incident (being unable to download new WUs), it appear to be working properly.

    As per usual, the advice is to clear out your cache of WUs crunching on the optimised version before closing BOINC completely, deleting the app_info.xml file and pc.exe executable then restarting BOINC and allowing it to download the new official application and some fresh WUs.

    Dave
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #16
    Whizbang said:
    Having a heck of a time getting any work on the machines I added here, they request new tasks and event log just says the request was completed, got 0 new tasks. No error, no out of work, just nothing. Machines that were attached several weeks ago seem to work fine.
    Click to expand...
    Posted a question about it on the forum, no answer yet. I did notice that sometimes the server says no jobs to send. I guess for now at least we need to set a 2nd project. I added Universe@Home as I noticed that a couple of my boxes were low on work. There's also a lot of potential upside for us at Universe. A heads up for Universe WUs, they very occasionally get stuck so if you see any that aren't progressing, abort them. The other side of the coin is that credit adds up fast, for what that's worth.
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #22
    It hasn't happened to me on TN-Grid but have had this problem on a number of other projects. Now I just exclude the whole BOINC directory. If you look through the forums you'll find that the issue has cropped up a lot on various virus/spyware scanners. If the executable isn't in their database and it's running all the time like a distributed application does, sometimes it gets flagged just because. Notifying the offending scanner company usually gets things fixed sooner or later.
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #23
    First of all could a moderator change the name of this thread to: TN-Grid - Perpetual Discussion Thread

    BTW, it's a big pain not being able to edit the OP as things change.

    That said: if you're running a first generation Ryzen on this project, it appears that here's an error in the SSE2 code so you'll want to run the AVX version.
    Easily done by putting this in an app_info.xml in the BOINC TN-Grid project directory and shutting down and restarting BOINC.

    Here's what needs to be in the app_info.xml:

    <app_info>
    <app>
    <name>gene_pcim</name>
    <user_friendly_name>gene@home v1.11</user_friendly_name>
    </app>
    <file>
    <name>gene_pcim_v1.11_win64__avx.exe</name>
    <executable/>
    </file>
    <app_version>
    <app_name>gene_pcim</app_name>
    <version_num>111</version_num>
    <platform>windows_x86_64</platform>
    <avg_ncpus>1.000000</avg_ncpus>
    <max_ncpus>1.000000</max_ncpus>
    <plan_class>avx</plan_class>
    <api_version>7.3.0</api_version>
    <file_ref>
    <file_name>gene_pcim_v1.11_win64__avx.exe</file_name>
    <main_program/>
    </file_ref>
    </app_version>
    </app_info>

    You'll also need to check to see if gene_pcim_v1.11_win64__avx.exe is in the directory. This is for Windows 64. If you need the executable I can send it to you or you can just run the project without the app_info for a few minutes and it should appear.
    If you need instructions for Linux I can point you to the right place.
     
    B

    Beyond

    Ars Praefectus
    5,236
  • #25
    Baenwort said:
    Are there still recommended optimized applications or has that been folded into the main?
    Click to expand...
    Sorry it took so long to reply, just got back home yesterday. It's automatic unless you're running a 1st generation Ryzen. Due to a problem in the SSE app code you'll want to run only the AVX version by using the app_info.xml and instructions just above.
     
    You must log in or register to reply here.
    Jump to latest Follow Reply
    Jump to latest Follow Reply