Overclock.net - An Overclocking Community

Overclock.net - An Overclocking Community (https://www.overclock.net/forum/)
-   Overclock.net BOINC Team (https://www.overclock.net/forum/365-overclock-net-boinc-team/)
-   -   Projects of the Month for December 2017 (https://www.overclock.net/forum/365-overclock-net-boinc-team/1643184-projects-month-december-2017-a.html)

tictoc 12-01-2017 12:07 AM

December Projects of the Month

[email protected]

Start: 22 - Goal: 21 - End: 22 |  It has been more than three years since we had LHC as a POTM.

 

SRBase

Start: 36 - Goal: 30 - End: 26 |  It is a bit of feast or famine for tasks, but if the queues are kept full we can probably move up a number of spots. :thumb:

 

[email protected]

Start: 22 - Goal: 21 - End: 22 |  Will be a heck of a month if we can move up to 21, but we can give it a go. :D

 

 

A bit of a different mix for the last month of the year. :) 


emoga 12-02-2017 10:27 AM

I'll switch over to SRBase. Still got some badges to grind. thumb.gif

tictoc 12-02-2017 10:57 AM

I have been having issues with Einstein.  Not sure what the deal is, but I am getting driver crashes on my 290s and my 1070s.  No issues on my 480 which is running the same OS and drivers as my 290s and 1070s. :headscrat

 

After killing the stuck tasks, they go into uninterruptable sleep and I have to shutdown and reboot in order to clear the stuck tasks.  I haven't had time to see what is going on, but hopefully I can get it sorted out over the weekend.


emoga 12-02-2017 02:39 PM

Looks like my Ryzen 1800X has terrible performance compared to my slower intel 2670 v0 processors when crunching SRBase. thinking.gif

I scanned through the top computers and the first Ryzen that appears is at #190 place...

tictoc 12-02-2017 07:25 PM

Quote:
Originally Posted by emoga View Post

Looks like my Ryzen 1800X has terrible performance compared to my slower intel 2670 v0 processors when crunching SRBase. thinking.gif

I scanned through the top computers and the first Ryzen that appears is at #190 place...

 

There is really no way that a 2670 should be 20% faster than an 1800X.  The only thing I could find was a post from April where rebirther said that it is not optimized for Ryzen yet.

 

I just loaded it up in Linux to see if it is a Windows issue, or if Ryzen is just knee-capped on this project

*Edit* I show the same results in Linux.  My R7 1700 is about 30% slower than my 2670v2

 

You might try and start a thread in the forums, since rebirther is active and is obviously optimizing the project with the use of AVX, AVX2, and FMA3.


AlphaC 12-02-2017 08:26 PM

LHC seems to be mostly ATLAS units (i.e. using VirtualBox) right now.

mmonnin 12-02-2017 10:32 PM

All I've gotten the past few weeks have been Atlas. There was a post about it somewhere on their forums. Got 4 threads of 1950x (ends up utilizing about 2.5-3) and about 1 threads worth of a 3570k. 9.8gb of memory usage on the 1950x and 5.3gb on 3570k.

Finished my Moo tasks and moved over the RX 580 to [email protected]

tictoc 12-05-2017 08:16 AM

Just fired up six 4-core Atlas tasks on my 2P.  Running six 4-core tasks uses roughly 35GB of memory on my 2P.  I did have to use an app_config with a much higher memory allocation than anything I saw on the forums, in order for the tasks to truly use 4 cores per task. 

 

Maybe they just take a while to start and I could have left the memory set at a lower value, but for the first thirty minutes each of the six tasks I was running were only using a single thread.  Once I upped the memory to 40GB in the app_config, the tasks immediately started using 4 threads each.


mmonnin 12-05-2017 09:56 AM

Quote:
Originally Posted by tictoc View Post

Just fired up six 4-core Atlas tasks on my 2P.  Running six 4-core tasks uses roughly 35GB of memory on my 2P.  I did have to use an app_config with a much higher memory allocation than anything I saw on the forums, in order for the tasks to truly use 4 cores per task. 

Maybe they just take a while to start and I could have left the memory set at a lower value, but for the first thirty minutes each of the six tasks I was running were only using a single thread.  Once I upped the memory to 40GB in the app_config, the tasks immediately started using 4 threads each.

They either take awhile to get up to higher CPU usage or it takes awhile to write nearly 10gb of data. They go from less than 1 core of CPU time to something like 2.5 cores for me when set to 4. I run a couple of CPU tasks on my 2nd client for GPUs to keep the CPU busy at all times.

mmonnin 12-05-2017 05:59 PM

Was browsing LHC forums and found a thread about this topic. Looks like 4, 5 and 3 threads are the most efficient.

https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=4479#32732


tictoc 12-05-2017 07:29 PM

I saw the same thing when I was digging around to try and find out why each of the tasks I was running were only using about 20% of a thread, instead of the 400% I expected to see.

 

@mmonnin is correct, and it is just a matter of waiting for the tasks to ramp up to full usage.  Screenshot below is six 4-thread tasks running.  One of the tasks had only been running for about 10 minutes.  It is not using any CPU, while the other 5 tasks are using 400+%.

 

 

*Edit* After waiting a few more minutes, the new task is now at full usage. :D


mmonnin 12-06-2017 03:44 AM

Anyone know if its just some pre-calculations, VM setup, loading the large dataset to mem or something? Like wheres the bottleneck to get to full cores. I just have an el cheapo 120gb SSD on mine that might take a bit to load 9.8gb.

CasperGS 12-06-2017 04:07 AM

How do you change user name in [email protected]??

mmonnin 12-06-2017 04:59 AM

Quote:
Originally Posted by CasperGS View Post

How do you change user name in [email protected]??

Account ->Preferences->Community

See if that changes your BOINC ID or just the one on the forum.

https://einsteinathome.org/account/prefs/community

Not the only one to ask:
https://einsteinathome.org/content/how-change-my-account-name

CasperGS 12-06-2017 05:04 AM

Quote:
Originally Posted by mmonnin View Post

Account ->Preferences->Community

See if that changes your BOINC ID or just the one on the forum.

https://einsteinathome.org/account/prefs/community

Not the only one to ask:
https://einsteinathome.org/content/how-change-my-account-name

Seems like everything is outdated, unable to locate ability to change username or set gpu use only on website or application (ver 7.8.3) or im stupid and cant figure it out.

CasperGS 12-06-2017 05:07 AM

Scratch that, I figured it out for username, but cant get it to run gpu only

tictoc 12-06-2017 05:56 AM

Quote:
Originally Posted by mmonnin View Post

Anyone know if its just some pre-calculations, VM setup, loading the large dataset to mem or something? Like wheres the bottleneck to get to full cores. I just have an el cheapo 120gb SSD on mine that might take a bit to load 9.8gb.

 

I haven't had a chance to log what's going in and out of the VM when the ATLAS tasks start, but according to this post it is downloading the dataset for the task.  https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=4440&postid=32442

 

Once that is complete then the WU launches with a single process

Quote:
The slides also explain the large initialisation time - running a single process for as long as possible in order to share as much memory as possible between the processes.

Link to post on ATLAS forums: https://lhcathome.cern.ch/lhcathome/forum_thread.php?id=4165&postid=29433

 

After reading through all of that and a bunch of other posts regarding ATLAS efficiency, I am probably going to change my settings to run 12*2-core tasks, rather than the 6*4-core tasks I am currently running.  Memory usage is not an issue on my 2P, but I'm not sure how it will affect disk usage.  Running 6*4-core tasks, ATLAS is using 18GB of space on my hard drive, so even if that doubles, I should have plenty of available space. 


mmonnin 12-06-2017 07:15 AM

Quote:
Originally Posted by CasperGS View Post

Scratch that, I figured it out for username, but cant get it to run gpu only

There is an option to run CPU versions of GPU tasks. Is that checked? Separate location?

CasperGS 12-06-2017 04:18 PM

Quote:
Originally Posted by mmonnin View Post

There is an option to run CPU versions of GPU tasks. Is that checked? Separate location?

I ended up changing it to 1% cpu usage, it wouldnt let me do 0%, that fixed my issue, ty

tictoc 12-13-2017 07:19 PM

POTM Update

[email protected]

Start: 22 - Currently: 23 - End:  |  CERN was only doing about 45k per day when I set the goal, so I didn't see them being able to catch us.:)  We are only a few days away from passing our original target the Dutch Power Cows. :thumb:

 

SRBase

Start: 36 - Currently: 30 - End:  |  Goal achieved. :thumbsups  We also have a shot at moving into the top 25 before the end of the month.

 

[email protected]

Start: 22 - Currently: 22 - End:  |  We are creeping up on Team Russia, and with the boost from @lanofsong the gap is getting narrower.

 

 

Great start to the month. :cheers: 


tictoc 12-14-2017 10:27 AM

I added a pair of 290s to Einstein last night. The 290s are about 100 seconds faster than my 480. Running two tasks concurrently the 290s average about 860 seconds per task. My 480 averages about 960 seconds per task. No OC on the 290s, and both systems are running the same OS and drivers. thinking.gif

mmonnin 12-14-2017 10:42 AM

My 580 is like 11-12min I think? But that's in Win7 where AMD cards go crazy slow with more than a task at a time. Looks idle quite a bit due to the task inefficiency. I need to find that thread that goes through the issue. GPU performance in [email protected] should really be measured up to 90%, after that CPU calculations are done which can skew run times.

EG's 7990s were pretty damn fast running [email protected] so I would expect 290s to be pretty quick as well.

tictoc 12-14-2017 12:02 PM

Quote:
Originally Posted by mmonnin View Post

My 580 is like 11-12min I think? But that's in Win7 where AMD cards go crazy slow with more than a task at a time. Looks idle quite a bit due to the task inefficiency. I need to find that thread that goes through the issue. GPU performance in [email protected] should really be measured up to 90%, after that CPU calculations are done which can skew run times.

EG's 7990s were pretty damn fast running [email protected] so I would expect 290s to be pretty quick as well.

 

Is 11-12 minutes just running one task at a time?


mmonnin 12-14-2017 12:08 PM

Quote:
Originally Posted by tictoc View Post

Is 11-12 minutes just running one task at a time?

Yes, a single task.

tictoc 12-15-2017 07:47 AM

The errors and lockups that I was having on Einstein on my 1070s and 290s were actually due to unstable memory. 

 

I was running my 1333 sticks at 1866, and apparently running that speed was not quite stable.  I caught the errors on SRBase, so I dropped the speed, and now my 290s have been crunching Einstein without issue for the last day and a half.:D


Egilman 12-15-2017 08:31 PM

Code:
<app_config> 
    <app> 
        <name>hsgamma_FGRPB1G</name> 
            <gpu_versions> 
                <gpu_usage>0.33</gpu_usage>
                <cpu_usage>0.2</cpu_usage> 
            </gpu_versions> 
    </app> 
</app_config> 

this is the app_config I was running mine on.

yes 3x, start out at stock clocks as long as they are stable.....
Then adjust upwards as long as they remain stable..........
And from what I remember, doesn't need a lot of memory....

This put my machines in the top ten on the project in the dailies.

tictoc 12-18-2017 06:47 PM

Quote:
Originally Posted by tictoc View Post
 

The errors and lockups that I was having on Einstein on my 1070s and 290s were actually due to unstable memory. 

 

I was running my 1333 sticks at 1866, and apparently running that speed was not quite stable.  I caught the errors on SRBase, so I dropped the speed, and now my 290s have been crunching Einstein without issue for the last day and a half.:D

 

Quoting myself. :D

 

Evidently that was not the issue.  Not sure what it is, and I don't have time to mess with it right now, but still seeing odd lock-ups and driver crashes when I run Einstein.  I'm going to put a 480 back on Einstein, and the 290s are going back to MilkyWay for now.


tictoc 12-20-2017 08:15 PM

I'm currently clearing out my LHC queue.  The constant traffic back and forth to the LHC servers murders my lowly DSL connection.  I will probably keep one rig crunching one or two tasks, but running 10+ tasks at a time, is not something that I can keep doing without throtlling the heck out of the traffic to the LHC servers. 

 

If I am going to keep running LHC, I will probably need to set up a local cache server to minimize all of the http overhead that LHC has.


emoga 12-20-2017 09:53 PM

@tictoc I switched to sixtrack when atlas was having its uploading issues. It's much smoother on the network, as well as the hard drives.

That's the thing with LHC...it's one of the most important distributed computing projects out there (imo)...but it also demands the most ram, bandwidth and patience from its users. rolleyes.gif

tictoc 12-20-2017 10:03 PM

I' m running 15 sixtrack tasks now.  I agree completely on the merits of the project. :)

 

The ATLAS tasks were running without too many issues, but when the ATLAS server went offline I started running the Theory and LHCb tasks.  I might go back to running just ATLAS and sixtrack.


mmonnin 12-21-2017 03:33 AM

Watch out for LHCb. Many people have issues with them running for 10min, condor not being able to communicate and then the task aborts. One computer does that for me, another is fine.

tictoc 01-01-2018 06:37 AM

POTM Final Total

[email protected]

Start: 22 - Goal: 21 - End: 22 |  We did pass the team we were originally gunning for, so I'm declaring this goal achieved. :D

 

SRBase

Start: 36 - Goal: 30 - End: 26 |  We nearly made it in to the top 25.  Great job team. :thumb:

 

[email protected]

Start: 22 - Currently: 22 - End: 22 |  This was a pretty impossible goal. We were making up some ground before Team Russia saw us coming and turned up the juice, which is a win for the project. :D

 

 

Good job everyone. :cheers: 

 

Working on the January POTM, and they will be up shortly.


mmonnin 01-01-2018 06:56 AM

Yeah the CERN team fired back up and zipped right past us. Seemed to be a lot of server space and upload issues when they came back.

tictoc 01-01-2018 07:10 AM

For sure.  I had a heck of a time getting any ATLAS work after they ramped up their production.



All times are GMT -7. The time now is 10:03 PM.

Powered by vBulletin® Copyright ©2000 - 2019, Jelsoft Enterprises Ltd.

User Alert System provided by Advanced User Tagging (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.
vBulletin Security provided by vBSecurity (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.

vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2019 DragonByte Technologies Ltd.