Overclock.net banner

Einstein@Home project support thread

12K views 86 replies 17 participants last post by  tictoc 
#1 ·
This thread is specifically for Einstein@Home project support.
Einstein@Home Website

Project info:
Einstein@Home uses your computer's idle time to search for weak astrophysical signals from spinning neutron stars (also called pulsars) using data from the LIGO gravitational-wave detectors, the Arecibo radio telescope, and the Fermi gamma-ray satellite. Einstein@Home volunteers have already discovered more than three dozens new neutron stars, and we hope to find many more in the future. Our long-term goal is to make the first direct detections of gravitational-wave emission from spinning neutron stars. Gravitational waves were predicted by Albert Einstein almost a century ago, but have never been directly detected. Such observations would open up a new window on the universe, and usher in a new era in astronomy.

**This will be updated by the BOINC Editors from time-to-time.
 
#5 ·
http://einstein.phys.uwm.edu/forum_thread.php?id=10707
Quote:
Due to the excellent work of our French volunteer Christophe Choquet we finally have a working OpenCL version of the Gravitational Wave search ("S6CasA") application. Thank you Christophe!

This App version is currently considered 'Beta' and being tested on Einstein@Home. To participate in the Beta test, you need to edit your Einstein@Home preferences, and set "Run beta/test application versions?" to "yes".

It is currently available for Windows (32 Bit) and Linux (64 Bit) only, and you should have a card which supports double precision FP in hardware.
It's probably going to be similar to Milkyway@Home where the top GPUs are all Tahiti/Cayman/Cypress , since it's Open-cl and double precision.
 
#6 ·
I think BRP4G (GPU Arecibo) has no more WUs to send for now. I didn't get any WUs whatsoever the past day.

http://einstein.phys.uwm.edu/server_status.html

Tasks to send
S6CasA 20,484
FGRP3 3,560
BRP4 11,270
BRP5 2,944
BRP4G 0

BRP4 work generator e1#0 einstein4 Disabled
BRP4 work generator e1#1 einstein4 Disabled
BRP4 work generator e3#0 einstein4 Disabled
BRP4 work generator e3#1 einstein4 Disabled
BRP4 work generator e6#0 einstein4 Not Running
BRP4 work generator e6#1 einstein4 Disabled
BRP4G work generator e3#0 einstein4 Not Running
BRP4G work generator e3#1 einstein4 Disabled
BRP5 work generator einstein4 Not Running
BRP5 work generator einstein4 Not Running

and this thread post May 6
http://einstein.phys.uwm.edu/forum_thread.php?id=10422
Quote:
About 2d left for BRP4G (87 beams at ~42 beams/d).

Better don't rely on BRP4G work only. If you selected specific applications to run, make sure to have BRP5 (Parkes Perseus Arm Survey) enabled as well.

BM
 
#7 ·
It looks like this is related to the recent earthquake in the area? I found a few articles related to the issue. The first shows the extent of the damage to the telescope:

http://www.planetary.org/blogs/guest-blogs/2014/0409-arecibo-observatory-earthquake-repairs.html

This one reports the telescope is back in action as of March 13th.

http://www.universetoday.com/110328/arecibo-observatory-back-in-action-following-earthquake-damage/

Hopefully they will start getting more WU's soon. Fortunately, I still seem to be getting WU's from other applications. I havent noticed any slowdowns recently, although I have only been running BOINC for 10 days now.
 
#9 ·
Question for those who know a lot more..... I'm currently running "GPU utilization factor of BRP apps" of 0.5, but on a regular basis it's trying to load two work units at once onto my Intel HD 4600 (aka IGP). Anyone know how to set things up so that I can still run two work units on the GTX 980, but have it set so it only ever tries to run a single one at once on the IGP? This problem has become more prevalent now that there's a 1.52 BRP beta app for Intel IGPs).

Thanks in advance!
 
#10 ·
Since I don't run both igpu and gpu units on the same machine I can't verify if this works, but using the 'plan_class' is a way of separating workunits with the same application name.

Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda32-nv301</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>
 <project_max_concurrent>4</project_max_concurrent>
</app_config>
 
#11 ·
Quote:
Originally Posted by emoga View Post

Since I don't run both igpu and gpu units on the same machine I can't verify if this works, but using the 'plan_class' is a way of separating workunits with the same application name.

Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda32-nv301</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>
 <project_max_concurrent>4</project_max_concurrent>
</app_config>
Hmmmm, so with that app_config I get this message in the log now...

Code:

Code:
JagerWulfe

20      Einstein@Home   13-Jul-16 06:18:36      Entry in app_config.xml for app 'einsteinbinary_BRP6', plan class 'BRP6-Beta-cuda32-nv301' doesn't match any app versions
Interesting part is that it's taking just over 4 hours (4h20m or so) to do two work units, versus the 2h18m my GTX 980 takes to do two. They are giving 1k points each though, which is 7.9 seconds per point if you factor in two tasks, versus 10.92 seconds per point for the basic Areceibo work units (that give 62.5 points when completed and validated)
 
#12 ·
This should work without any error logs.

Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
        </app_version>
           <app_version>
       <app_name>einsteinbinary_BRP4G</app_name>
       <plan_class>BRP4G-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
        </app_version>
</app_config>
I only run BRP6's on the igpus but throw some arecibo's in on my gpu. If you aren't planning on doing arecibo then just delete the last app in the code.... or use this

Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
        </app_version>
</app_config>
I actually do 4 BRP6 tasks at a time on my 970 with 0.25 cpus each. Seems to get the most PPD. Hope this helped.
 
#13 ·
Quote:
Originally Posted by emoga View Post

This should work without any error logs.
Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
        </app_version>
           <app_version>
       <app_name>einsteinbinary_BRP4G</app_name>
       <plan_class>BRP4G-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
        </app_version>
</app_config>
I only run BRP6's on the igpus but throw some arecibo's in on my gpu. If you aren't planning on doing arecibo then just delete the last app in the code.... or use this

Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.5</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
        </app_version>
</app_config>
I actually do 4 BRP6 tasks at a time on my 970 with 0.25 cpus each. Seems to get the most PPD. Hope this helped.
Seems to do the trick! As for running four at a time, I'll give that a shot and see how it goes. I was doing two before because three seemed to be a tad slower, but why not give it a try ^_^

Eventual code I wound up going with was slightly modified to what you wrote up.... Basically just a few extra intel_gpu entries due to having the work units.

Code:

Code:
<app_config>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-cuda32-nv301</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.25</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-cuda55</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>0.25</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP6</app_name>
       <plan_class>BRP6-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>
   <app_version>
       <app_name>einsteinbinary_BRP4G</app_name>
       <plan_class>BRP4G-Beta-opencl-intel_gpu</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>   
   <app_version>
       <app_name>einsteinbinary_BRP4</app_name>
       <plan_class>opencl-intel_gpu-new</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>   
 <project_max_concurrent>4</project_max_concurrent>
</app_config>
 
#14 ·
How do you find out what the app version vs plan classes are for each task?

Most of that doesn't work for me. I still have 3 errors:

Code:

Code:
7/18/2016 9:08:36 PM | Einstein@Home | Entry in app_config.xml for app 'einsteinbinary_BRP6', plan class 'BRP6-Beta-opencl-intel_gpu' doesn't match any app versions
7/18/2016 9:08:36 PM | Einstein@Home | Entry in app_config.xml for app 'einsteinbinary_BRP4G', plan class 'BRP4G-Beta-opencl-intel_gpu' doesn't match any app versions
7/18/2016 9:08:36 PM | Einstein@Home | Entry in app_config.xml for app 'einsteinbinary_BRP4', plan class 'opencl-intel_gpu-new' doesn't match any app versions
4 BRP6-Beta-cuda55 tasks are running at once but only a single BRP4G-Beta-cuda32-nv301 was running at once.

Edit: I found this page which links the checkboxes on the preferences page with the plan class. Where was the app_name come from? Just einsteinbinary_BRPxxx where xxx is 4, 6 or 4G?
https://einstein.phys.uwm.edu/apps.php
 
#15 ·
Quote:
Originally Posted by mmonnin View Post

How do you find out what the app version vs plan classes are for each task?

Most of that doesn't work for me. I still have 3 errors:

Code:

Code:
7/18/2016 9:08:36 PM | Einstein@Home | Entry in app_config.xml for app 'einsteinbinary_BRP6', plan class 'BRP6-Beta-opencl-intel_gpu' doesn't match any app versions
7/18/2016 9:08:36 PM | Einstein@Home | Entry in app_config.xml for app 'einsteinbinary_BRP4G', plan class 'BRP4G-Beta-opencl-intel_gpu' doesn't match any app versions
7/18/2016 9:08:36 PM | Einstein@Home | Entry in app_config.xml for app 'einsteinbinary_BRP4', plan class 'opencl-intel_gpu-new' doesn't match any app versions
4 BRP6-Beta-cuda55 tasks are running at once but only a single BRP4G-Beta-cuda32-nv301 was running at once.

Edit: I found this page which links the checkboxes on the preferences page with the plan class. Where was the app_name come from? Just einsteinbinary_BRPxxx where xxx is 4, 6 or 4G?
https://einstein.phys.uwm.edu/apps.php
Before I forget again, I have beta tasks enabled, so that's probably where the errors are coming from for the first two lines. As for figuring/finding out what app version versus plan classes were for each task... I looked at the individual tasks I had downloaded and then found which one was which in the Applications matched..... Then (especially for that last one, the opencl-intel_gpu-new one) I manually created entries for each and every possible type of work unit that my Intel HD 4600 could download..... Saved, restarted my client, and then after fifteen minutes and eight client restarts, I had weeded out the various ones that were throwing errors on my end.

Not the prettiest of ways to do it, but it locked things down so that over the past five days (well, minus today anyways, due to the Foldathon going on atm) my IGP only has one task running at any given time. Those Parkes PMPS XT work units sure take a loooooong time though (8h17.35s), especially compared to BRP (Arceibo) tasks for the IGP, which usually are around 10m11-15s. Then again, 1k points vs 62.5, which isn't as efficient for points. Huh.
 
#16 ·
So, curious question.... Are Kepler based cards faster than Maxwell for Einstein@Home when factoring in default/factory clocks? GTX 780 Classified at 1136MHz core vs GTX 980 running at 1254MHz core. I'm curious because so far it seems to be completing 1.57 PMPS XT tasks almost twenty minutes faster...
 
#17 ·
I fired up my 7850 on some of the newer Gamma-Ray pulsar binary search #1 applications and could not believe how horrible the run times were. It was like 6 hours to complete one, while on a 960 I was getting 1:15, in each case running 2 at a time. Some investigation revealed the issue, these tasks are using too much VRAM to run 2 at a time on a 1G GPU, if I run one at a time on the 7850 my run times are looking to be closer to 33 minutes at about 70% usage.

Here's usage from the 960, the numbers are similar for the 7850 in Windows, except 1/2 the memory has to be swapped out/in to swap tasks.

+-------------------------------------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|==============================================================================|
| 0 1260 G /usr/bin/X 76MiB |
| 0 2559 G cinnamon 22MiB |
| 0 17248 C ...17_x86_64-pc-linux-gnu__FGRPopencl-nvidia 781MiB |
| 0 17404 C ...17_x86_64-pc-linux-gnu__FGRPopencl-nvidia 781MiB |
+--------------------------------------------------------------------------------------------------------+

Here is the 7850 running 2x in Windows, so I guess it was swapping 530M of memory every time it tried to swap tasks. Running one task memory usage drops to 750M.



Of course with a RX-470 in the same system I would still like to run multiples, but I don't see anyway to set GPUs in the same system to run different numbers of tasks.
 
#18 ·
Quote:
Originally Posted by bfromcolo View Post

but I don't see anyway to set GPUs in the same system to run different numbers of tasks.
App_config with plan class?

Code:

Code:
<app_config>
  <app_version>
       <app_name>hsgamma_FGRPB1G</app_name>
       <plan_class>FGRPopencl-Beta-ati</plan_class>
       <avg_ncpus>1</avg_ncpus>
       <ngpus>1</ngpus>
  </app_version> 
<app_version>
       <app_name>hsgamma_FGRPB1G</app_name>
       <plan_class>FGRPopencl-Beta-nvidia</plan_class>
       <avg_ncpus>1</avg_ncpus>
       <ngpus>0.3</ngpus>
  </app_version>  
</app_config>
Not sure I read that right, you want to use the 7850 with the 960? or the 7850 and a 470? I haven't finished my morning coffee yet.
kookoo.gif
 
#19 ·
Quote:
Originally Posted by emoga View Post

App_config with plan class?

Code:

Code:
<app_version>
       <app_name>hsgamma_FGRPB1G</app_name>
       <plan_class>FGRPopencl-Beta-nvidia</plan_class>
       <avg_ncpus>1</avg_ncpus>
       <ngpus>0.3</ngpus>
  </app_version>  
  <app_version>
       <app_name>hsgamma_FGRPB1G</app_name>
       <plan_class>FGRPopencl-Beta-ati</plan_class>
       <avg_ncpus>0.5</avg_ncpus>
       <ngpus>1</ngpus>
</app_config>
Not sure I read that right, you want to use the 7850 with the 970? or the 7850 and a 470? I haven't finished my morning coffee yet.
kookoo.gif
I have a RX470 and HD7850 in the same Win 10 box. The way it is set up now if I wanted to run both, I could only run one task on each.
 
#23 ·
Quote:
Originally Posted by emoga View Post

Worst case would be to open a VM and use an exclude gpu tag in the cc_config for the 470. Sucks I know.
I have never managed to pass through a GPU to a VM, but that is probably just me using the wrong tools.

I could probably put this in my HTPC and pull the GTX-950 out of there, although if the 7850 runs much hotter gaming I might have issues given the limited airflow in my cabinet. The 7850 is still a decent performer on a lot of stuff, and FP64 is not as cut down as newer cards.

Just thought I would let people know about the memory requirements of the newer GPU tasks. Running 3x tasks on a 2G card would encounter similar issues.
 
#24 ·
Its interesting as I look into these new GPU tasks, I am not convinced there is any benefit to running multiple tasks per card, in fact it may hurt.

On my RX470, running 2x I got run times of 3863 seconds, running 1x I get 1341 seconds. So based on my sample size of 3 tasks it looks better to no longer run multiple tasks per GPU. Just to be clear I am testing:

Gamma-ray pulsar binary search #1 on GPU 1.17 (FGRP-opencl-ati) on Windows 10.

I will run a few more tasks and verify the result, and I guess I will go mess with a linux system and a nvidia card as well.
 
#26 ·
Quote:
Originally Posted by emoga View Post

Are you sure they're the 1.17's? 1.18's have been out for a while and have noticeable speed improvements....or they may cripple your card even further
wink.gif
Yes I am getting only 1.17 in both Windows and Linux. Maybe I need to set a test flag or something. Guess I will stop wasting my time on this then and wait for the 1.18s to appear. From reading in the forum the 1.18s are faster and people still benefit running multiples as long as they have sufficient vram.

Edit - set the test flag and getting 1.18 units now, faster, but still using about 750M each of vram. So I can run only one on my 7850, but it looks to be 90% utilized with one.
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top