BOINC

Just a reminder that the next Primgrid challenge starts on this Wednesday for anybody wanting to run in it. This is a short one only thee days long so getting started on time is going to be important on this one. ;)

Maybe I can do a little better job of starting on time this go around rather than 14 hrs. late. :D
And for those of you with Intel rigs HT off is the way to go. ;)

http://www.primegrid.com/
 
I have not ran any work on it yet, but have added Atlas@home to the all inclusive DC list. It is yet another CERN project. It as well requires Virtualbox to be installed. I also set up our team there.

http://atlasathome.cern.ch/index.php

You also need to run it from a 64bit OS.

http://atlasathome.cern.ch/forum_thread.php?id=15#49

I am holding off on adding their 4th project because it is invite only right now and I haven't attempted to get a code. If anyone else bothers, PM me and I will set the team up. http://lhcbathome.cern.ch/Beauty/
 
Last edited:
Grandpa_01 said:
Just a reminder that the next Primgrid challenge starts on this Wednesday for anybody wanting to run in it. This is a short one only thee days long so getting started on time is going to be important on this one. ;)

Maybe I can do a little better job of starting on time this go around rather than 14 hrs. late. :D
And for those of you with Intel rigs HT off is the way to go. ;)

http://www.primegrid.com/
Click to expand...

Is this one of those projects that crunch faster if your CPUs have AVX? I'll defer if that's the case as my AMD 4P rig was a slug during the last challenge.
 
tjmagneto said:
Is this one of those projects that crunch faster if your CPUs have AVX? I'll defer if that's the case as my AMD 4P rig was a slug during the last challenge.
Click to expand...

Yes it is TJ my amd boxes are takeing about 10 hrs per WU on the AMD 6276 box vs around 3 on the 4650 boxes. Primegrid really favors Intel AVX when it comes to preformance.
 
I don't know whom all is interested but Milkyway has a challenge going from 6/27-7/1. I have it listed in the challenge thread. Since PrimeGrid overlaps, this would be a good place to run those double precision GPU's. I will be out of state during a good portion of this, so don't know how much I will accomplish on either challenge. http://hardforum.com/showthread.php?t=1822815
 
I set up two BOINC systems with GPUGRID on GPU, WCG on CPU (no GPUGRID CPU app) :
- first system dual core + 1 Kepler GPU : 100% CPU use (2 WCG tasks crunched for 1 GPUGRID task), no problem
- second system dual core + 2 Kepler GPU : only 60-65% CPU use, 1 WCG task crunched for 2 GPUGRID tasks... of course I would like 2 WCG tasks so CPU is fully used

On second system, GPUGRID reports wrongly in boinc manager it is using 0,691 CPU core per GPU, when real use is more 0,12 CPU core per GPU, and I suspect it's based on this information that boinc manager runs only one task because it thinks it has only 0,6-0,7 core available for WCG.

Any thoughts ?
 
The BOINC manager won't change the amount of CPU usage it reports the app is using. In fact that isn't what it is actually using. However, it is an estimate of what is expected to be needed. Since GPU apps are severely hindered by access to CPU when needed, it will try to reserve that amount to feed it. Most people will set aside an entire thread/core per GPU unless it is a low end card. GPUGrid will typically use features of your card that most other projects do not. It also has a habbit of pushing parts of it harder. Don't expect it to always be pushed to extreme though because each batch may have its own requirements.

If you have a strong desire to force the CPU usage to be less available for the cards, you can do that with config files. But you would be hindering your potential GPU output. I would just leave it as it is.
 
That is great news AgrFan. I can't wait for phase 2. Probably my favorite project/sub-project ever.
 
I have been away from the forum for several years and just installed BOINC. I only have a 2.3Ghz I7 in my Macbook Pro. Nonetheless, all of my points are going to hard ocp.
 
molson3530 said:
I have been away from the forum for several years and just installed BOINC. I only have a 2.3Ghz I7 in my Macbook Pro. Nonetheless, all of my points are going to hard ocp.
Click to expand...

Great to have you back with us! Look around, especially at the DC Vault thread if you want to join us on monthly challenges, or just pick your favorite DC Project and gain yourself some personal goals - either way, it's all good!
 
molson3530, I recommend also attaching QCN to your Macbook Pro. QCN can use the accelerometer in the hard drive and the work units are non-CPU intensive. Therefore they can run alongside other projects and use a nominal amount of resources. Our team doesn't have a lot of users there yet. You can also use Lenovo/IBM Thinkpads, HP laptops with the 3d Hard Drive software, and various Android devices. Otherwise the project requires a usb sensor.
 
They finally did a stats export and I am now over 1 million points there. Time to detach.
 
Ben Sagel a project admin/moderator at LHC Six Track has posted that they intend on merging the LHC BOINC projects some time in the future. I think this is important due to the fact that Six Track (the original LHC project) has had so much trouble putting out work.

http://lhcathomeclassic.cern.ch/sixtrack/forum_thread.php?id=3848#26608

I wouldn't hold my breath on a time frame though. There are currently 4 LHC BOINC projects.
 
Last edited:
New site features: https://secure.worldcommunitygrid.org/about_us/viewNewsArticle.do?articleId=366

7 Jul 2014

Summary
New website features give you more information about the people and connections that make World Community Grid possible.


World Community Grid is all about people who donate resources, coordinate their efforts, undertake research and help change the world for the better. So as part of the ongoing improvements to our website, we've just rolled out two features to help you learn more about the people and connections that make World Community Grid possible.

Meet the team
Get to know key members of the World Community Grid team. We came to IBM from a wide variety of backgrounds, but we all found places on the team because of our commitment to technology, philanthropy and community.

See who you've introduced to World Community Grid
Get recognized for the work you do to help build our community. You now have a personalized invitation link to invite others to join World Community Grid. When friends sign up using your link, you'll get recognized for recruiting them on your My Contribution page.

Check out the new features, and invite your friends to join today.
Click to expand...
 
Gilthanis said:
molson3530, I recommend also attaching QCN to your Macbook Pro. QCN can use the accelerometer in the hard drive and the work units are non-CPU intensive. Therefore they can run alongside other projects and use a nominal amount of resources. Our team doesn't have a lot of users there yet. You can also use Lenovo/IBM Thinkpads, HP laptops with the 3d Hard Drive software, and various Android devices. Otherwise the project requires a usb sensor.
Click to expand...

Thanks, I will look into that. I need to take my T60 out of my sig as it is currently down.
 
I added Universe@home to the all inclusive DC list. Odds are that it will be added to BOINCStats soon.
 
Another cryptocoin tied to BOINC is about to be launched. There is no info on if you need to switch teams yet or not, but I couldn't imagine them not monopolizing on all of it by following the other 3 that did it. I only mention it here as a heads up to potential new team to keep on the radar. These crypto teams tend to steal a lot of members and then just mow through the ranks really fast.

https://bitcointalk.org/index.php?topic=688284.0
 
Gilthanis said:
The BOINC manager won't change the amount of CPU usage it reports the app is using. In fact that isn't what it is actually using. However, it is an estimate of what is expected to be needed. Since GPU apps are severely hindered by access to CPU when needed, it will try to reserve that amount to feed it. Most people will set aside an entire thread/core per GPU unless it is a low end card. GPUGrid will typically use features of your card that most other projects do not. It also has a habbit of pushing parts of it harder. Don't expect it to always be pushed to extreme though because each batch may have its own requirements.

If you have a strong desire to force the CPU usage to be less available for the cards, you can do that with config files. But you would be hindering your potential GPU output. I would just leave it as it is.
Click to expand...

I have been out of town, back now. Thanks Gilthanis, I followed your advice and scored some 15M GPUGrid points in 2 weeks :).
 
I pulled this from their news

Mega Prime Search Now Running on BOINC
We have moved the PRPNet MEGA project over to the BOINC server. To run PPS-Mega, just select the PPS-Mega (LLR) app in your PrimeGrid Preferences.

The PPS-Mega project is part of our Proth Prime Search (PPS) project. There's now three ranges of PPS tests:

PPSE: Currently around n=1.3M and about 400K digits
PPS: Currently around n=2.1M and about 650K digits
Mega: 3.322M < n < 3.6M and just over 1 million digits

The PPS-Mega tasks take about an hour on the fastest computers (Intel Haswell CPUs). This project is ideal for finding mega primes. In all of history, only 97 mega primes have ever been found -- but PrrimeGrid has already found 17 so far this year.

Will you discover the next mega prime?
Click to expand...
 
I figured I would post this here even though I added the info to the All Inclusive DC List. GPU Grid currently is limiting the number of GPU's you can use per host to a maximum of 8. This doesn't affect many people, but does affect at least one of our team members. They are looking at fixing the issue, but it probably wont be soon.

http://www.gpugrid.net/forum_thread.php?id=3783&nowrap=true#37320

Edit: They also said in another thread that more CPU work was coming, but this batch would not support ARM processors.
 
Interesting thing I just noticed. An R9 270X is marginally faster than a 780 Ti at Milkyway@home at both OpenCL tasks. Fascinating considering the double precision GFLOPS is actually more for the 780Ti at 493 Gflops (according to TechPowerup) vs around 193 Gflops for the 270X. Interesting. The other measured speed of 210 Gflops for the 780 Ti could be more accurate though.
 
Yeah, AMD + DP is where its at for Milkyway I think.
 
I just noticed GPUGRID team released a new CPU only app for testing, that is multithreaded.

Developer's comment :
This application scales linearly for small N - I'm estimating 4-8 cores on most machines.
Click to expand...

Will need to test with more cores :D

Original thread
 
Interesting note, QCN seems to be able to utilize the accelerometers in SSD's. I've got it attached to two desktops, and the work units are working splendidly.
 
Nicolas_orleans said:
Well, a Kraken or derivative boosting GPUGRID's GROMACS would be fantastic for sure. Anyone happens to have news from tear (I have not seen posts from him lately) ?
Click to expand...

He is around he just posted in a thread the other day an he is on irc quite often.
 
Have they said what version of GROMACS they are using? I know some of the newer versions have some nice speedups and AVX (not sure of the gains). Scaling, as we know from F@h, will depend on the size of the simulation.
 
Kai said:
Interesting note, QCN seems to be able to utilize the accelerometers in SSD's. I've got it attached to two desktops, and the work units are working splendidly.
Click to expand...

Kai, are you sure they are running work units or is it just the screen saver apps. Even systems without working accelerometers can still download work units...they just don't do anything but allow for a screen saver. I haven't read anywhere on their forums of supporting SSD's in general yet. The good news is that you get credit for run time of the app on WUProp even if you just run the screen saver app. I also don't see you on the team...
 
Last edited:
You must log in or register to reply here.
Back
Top