DC Vault 2

I renewed the discussion in the DC-Vault about adding ibercivis. I had emailed the project admin to see if he would give permission, and he does. So, if anyone would like to give their input over at the Vault, please do so. If you do not have a way to log in (because they block you from creating an account), just let me know and I will forward it to their forums for you.

ibercivis is an umbrella project that would more than likely go into the Misc. category.

http://www.team-ninja.com/vbulletin/showthread.php?t=45061
 
Rusty has added me to the Vault's admin group. So, if team members have issues, concerns, or suggestions, they should really let me know. This way I have a better idea of things that need addressed or clarified. I am very verbal and tend to have my own opinions regardless of whether it helps or hinders the team. I have a feeling we will be very competitive regardless.
 
And another +40% gain for intel bringing the total gain up to 350%. only +33% on AMD.

Hmm, now the gfortran -O3 version is up to a 60% gain, might be enough for Kevin to update the project version now...

https://dl.dropboxusercontent.com/u/123993183/fit_sed_x86_64-pc-linux-gnu_340_intel.zip

Nick, this work you're doing is amazing. Although I'm not participating in this particular project, I'm happy to see someone step-up with some code and optimization contributions. The project would be stupid to not accept your patches. :D
 
Nick, this work you're doing is amazing. Although I'm not participating in this particular project, I'm happy to see someone step-up with some code and optimization contributions. The project would be stupid to not accept your patches. :D

I think it isn't as much as accepting patches as having the proper licensing to do them. They will more than likely want to replicate the procedure to make sure there wasn't tampering for malicious intent. If they don't have access to the tools, they may not be able to accept the changes. Anyways, that is outside my range of experience. I'm hoping your work pans out. This is awesome that you can achieve that much performance gain.
 
Rusty has added me to the Vault's admin group. So, if team members have issues, concerns, or suggestions, they should really let me know. This way I have a better idea of things that need addressed or clarified. I am very verbal and tend to have my own opinions regardless of whether it helps or hinders the team. I have a feeling we will be very competitive regardless.

Excellent news!

And I would also like to echo the above, really great work NickOfTime :)
 
I notice we jumped 1 position (to 12) in POGS and gained 20 points in DC-Vault. We jumped two positions in Rosetta (to 122) and gained ~2 points. I've been concentrating on Rosetta and SIMAP since it seems we can move up multiple spots per day. Can someone remind me how the points per position works? Did we get 20 points on POGS because it's harder to move up or because it's position 12 instead of position 122?

Thanks.
 
POGS is worth more because there are less teams participating in it via the Vault.

Points are based on total teams.

http://www.team-ninja.com/vbulletin/showpost.php?p=327549&postcount=6

The scoring calculation to determine the team positions within the vault ranking system works as follows:

Overall score = Sum of all category scores
Category scores = Sum of all projects scores in a category
Project score = 10000 - ((Position - 1) * (10000 / Total number of teams in the project))

At first glance this may seem odd to place more "value" on a less popular project. However, here are other things to consider:
1. 1st place is always just worth 10,000 points. If you want to stay competitive in the Vault, you need to decide if you can afford to divert resources to the smaller projects to be able to claim bragging rights. Think of this like our [H]ard DC'er of the Month/Year awards.
2. Just because there are a lot of teams, doesn't mean you can't rac up a ton of Vault points pretty easily. You don't have to be first in any of the projects to be first over all in the Vault. This allows for people to have selection to support multiple projects. It also helps teams that are divided in interests to stay competitive.
3. This is a side benefit to project for being in the Vault. It gives small projects a little extra resources they may not have been given otherwise.
4. If more teams join the Vault, the points per position will actually decrease in value. Granted you still get a total of 10k for 1st place, the difference between 1st and 2nd place is minute.

If you take a look at the non-BOINC non-FAH projects, you will see up until recently there has been very little activity from our members in any of the other projects. There are only a couple who continue to run those clients. Those projects were probably only supported by our team simply because of the Vault and most of the pushes in those projects have stopped or only fire up on the rare occasion. The pushes stopped because the "easy" points were completed and the resources to move on just aren't "worth it". Especially if the whole team isn't pushing them.
 
Last edited:
Thanks Gilthanis and KMac. I remember reading some of this a while ago and lost track of it plus I didn't understand it before getting involved. I've added RNA to my project list so we can hopefully get some traction there and it's about 20 points per position.
 
Keep in mind that RNA is skimp on available work units and each one can eat up to 3GB of Ram. They aren't short running either. However, I do encourage running it if you can.
 
Keep in mind that RNA is skimp on available work units and each one can eat up to 3GB of Ram. They aren't short running either. However, I do encourage running it if you can.

Phew, that's good to know. Luckily the majority of my hosts have 2-4GB RAM per core so I should be okay.
 
Also, supposedly the work units will auto extend the deadline if they detect that progress is still being made. Kinda cool. I should also mention that completed work units using the virtual box wrapper, will show up as error while computing many times and other times will show up as completed. This is natural and you should still get credit. BOINC wasn't originally designed with the wrapper in mind, so the servers validate a little different on those work units. It can be confusing at times because you can still have erroring work units that don't get credit.
 
They didn't officially announce the release from what I was looking at a few minutes ago. The last update I saw indicating this was what I had already posted estimating end of the week.
 
Just posted

I've now started some GPU work units. Unfortunately I can already see some errors.

Please make sure you run the new work units with poemcl 2.00. This should be done automatically; however I can see some failed work units using deprecated versions, so please check your config files if you have edited them manually.

You may also have to check the settings of installed anti virus software and allow poemcl to use your GPU.

If there are other errors, please report them here.


To the latest questions:
The new POEM Version is also running old tasks, but has new force field implementations which are used for the recently started simulations. All of our running projects are concerning Protein Optimization.

We are working on a release for the Mac, but it will still only support CPU jobs.

http://boinc.fzk.de/poem/forum_thread.php?id=1034&postid=9449#9449
 
So far they appear to run better on Linux than Windows at least as far as Nvidia goes and they do use a full core to feed them.

Edit
The Linux task showed a computational error in the stats after completeing. :(

Edit
I have now finshed 1 Linux and 2 Windows task all errored, I would not recomend running these yet
 
Last edited:
So, after about a month of testing SOB, it took ~2 weeks to compete each work unit on a i5-3470 at clock speeds. So, over the last month it has completed 8 work units. The client seems very stable too. When I get a chance, I will make a guide.

GIMPS has only been running part time on my laptop that has a T1400 CPU in it. That one is taking much longer to process and it doesn't help that I don't always remember to turn the client back on sometimes. They really need an option for it to only pause for a certain amount of time rather than an all or nothing start/stop.

We also need to start discussing what we want to push next month. POEM is still working out the kinks, so I would suggest playing that one by ear.

If we focused as a team, we could probably go from 20th place to the top 10 at NFS. I will also point out that with our current POGS production, it would take ~20 days to jump to the next spot. However, I don't think the whole team is working on it right now.
 
Last edited:
POGS will take the next step in around 7 or 8 days, I think you are looking at the RAC which is actually less than 1/2 of what our actual daily production is. I believe I am currently producing 1.6M daily and we are 12M away from the next overtake. So we should gain 1 more before the end of the push.

I did move the 3 - G34's over to Asteroids to give a boost there. NFS does look pretty good for push next month to me the first 6 takeovers look like they would come pretty fast.

What I would really like to see is a guide for installing and running the Prime Sierpinski Problem - PRP, I could really make some vault points there in a relatively short amount of time, if I could figure out how to get it installed and running in Linux Ubuntu 12.04 :)
 
That is a good suggestion Grandpa for a guide, but I'm still testing three other clients right now. I'm thinking of also putting up a guide for theSkynets Source Finder client. I will also point out that to get chunks of work for PSP, you will have to email them according to their forums post.
This thread has been locked as of now. If you need to reserve a private range for manual testing please PM me.
http://www.mersenneforum.org/showthread.php?t=2667
So, I'm not sure how much of a pain it will be to get work.

Perhaps one of our team mates that have already ran the client could chime in. Sieve http://www.psp-project.de/statdetail/sieveteam29.html PRP http://www.psp-project.de/statdetail/prpteam29.html

Here is their installation instructions for those interested: http://www.ldausch.de/guides/prp.html
 
Last edited:
Got a little bit of reorganisation done today and pointed one machine over at NFS to get a handle on how it goes. Most of my machines are crunching over at Leiden right now to give Gilthanis a hand, see if we might be able to pull one or two places in that before I refocus for April.
 
Also at Primegrid in the Month of April there is the Shakespeare's Birthday challenge, it is only a 3 day event, and I figured since we already have 1720 challange points form the last one I will run in the rest of the challanges to see how we can do, (If I can remember to) who knows we may do pretty good there.

http://www.primegrid.com/challenge/challenge.php
 
I mentioned Foumla BOINC over a year ago. Not sure exactly when. However, I noticed that they don't have us fairly listed since we use 2 different team names across the BOINC world. I emailed them once to see if they could adjust it so that we were fairly represented and never got a response back. So, I never put much interest in it.
 

I saw that too. I have a slow lowly laptop playing with that client right now. I plan on doing a Guide soon in case others are interested. It isn't really that hard to set up. The work units are large though. It is actually the same client as SOB, but the config file is a little different.

edit:
Well, Everybody lost 19 points on Gimps...
Not entirely true. I'm guessing someone did a very large push. Or a team got added that was already a high ranking team.
 
Last edited:
No kidding. ComputerraRU got almost 29M points added in a single day according to this:
http://digik-oz.nl/DcStats.aspx?project=GIMPS

:eek:

Looking further into it, I'm guessing possible bug or error.

http://www.mersenneforum.org/showthread.php?t=15772

Whatever it was it happened between 5am and 6am UTC (1am-2am EDT) around 10 hours ago.

5am UTC:
Today's Numbers
Teams 690 Users 121,059 CPUs 917,267 TFLOP/s 194.846 GHz-Days 97,423

6am UTC:
Today's Numbers
Teams 690 Users 121,059 CPUs 917,281 TFLOP/s 57,922.173 GHz-Days 28,961,086

There is nothing big on the recent cleared page back to 2:44 am UTC.
http://www.mersenne.org/report_recent_cleared/

It is another factoring bug. There is a new #1 factoring producer with 28.9 mil Ghz-days :) :
1 Smok_bmv 28904862.001 8847 123

I am only assuming they are discussing this issue. I couldn't find anything else potentially related.
 
Hmm I was crusing the net and found this. http://formula-boinc.org/index.py?lang=en&team=6928&year=2014 It appears we are doing fairly well in it. But I do not know if we even try to compete there or not. Did somebdy enter this or is evertbody entered by default. :confused:

I did some looking around and found my old post here: http://hardforum.com/showpost.php?p=1039243151&postcount=640 So, it has been 2 years since I contacted them. Maybe it is time for a follow up email.....
 
Looks like ComputerraRU lost his 28M points he is now right behind us.

26 1 HardOCP 165,504.39
27 26 ComputerraRU 165,370.06 -28,865,347.79 1,242.01 5,907.37
 
POGS will take the next step in around 7 or 8 days, I think you are looking at the RAC which is actually less than 1/2 of what our actual daily production is. I believe I am currently producing 1.6M daily and we are 12M away from the next overtake. So we should gain 1 more before the end of the push.

I did move the 3 - G34's over to Asteroids to give a boost there. NFS does look pretty good for push next month to me the first 6 takeovers look like they would come pretty fast.

What I would really like to see is a guide for installing and running the Prime Sierpinski Problem - PRP, I could really make some vault points there in a relatively short amount of time, if I could figure out how to get it installed and running in Linux Ubuntu 12.04 :)

Since SZTAKI runs out of work units pretty quick, we could possibly even do an SZTAKI/NFS push next month. If work units were available, it wouldn't be too difficult for our team to take 5 positions at SZTAKI. However, I have a feeling the work unit availability would probably be the weak spot.
 
Looks like ComputerraRU lost his 28M points he is now right behind us.

26 1 HardOCP 165,504.39
27 26 ComputerraRU 165,370.06 -28,865,347.79 1,242.01 5,907.37

Yeah, I checked the DC-Vault stats this morning and we got our 19 points back in GIMPS. Looks like GIMPS had a stats server error or something along those lines.

I have a GTX 660 doing Milkyway and trying to get POEM WUs to no avail. Was thinking it could help our cause with GIMPS (outside of BOINC) instead. Any ideas if GIMPS does well on Nvidia or better on AMD?
 
If it is anything like Primegrid it will do pretty good on Nvidia but I have not ran it on my Nvidia GPU's so I am not qualified to answer your question :D, As far as poem goes you do not want any of the current GPU work I have run a few of them to no avail, I had 100% failure rate. :rolleyes:
 
Most of my POEM work units worked on the first to complete, but then failed on all that followed for each card. The only exception was a box using a GT430 which crunched two successfully. I need to check that box to see if it still uses an app config to run multiple units on POEM or not...

PrimeGrid runs great on nVidia.

GIMPS, I am not sure about yet. I am only running CPU on the one test laptop. I believe the GPU work uses a different app than the standard Prime95 client.

I found this and figured it would be helpful. http://www.mersenneforum.org/gccs/GIMPS_GPU_Computing_Cheat_Sheet.pdf
 
Last edited:
Hey guys, long time lurker, first time poster :)

I crunch for OcUK and have been using NickofTime's optimised Linux exe for pogs (thanks) and see you now talking about GIMPS on the GPU which I have some experience with so thought I'd help out if possible.

There is actually a guide in this thread from phoenics, it's a bit dated in the sense that you don't need to download the separate dlls anymore and also since then the newer client uses very little CPU resource and one instance maxes out the GPU.

As per Gilthanis download links it's mfaktc for Nvidia, mfakto for AMD.

There is a comparison of GPU performance here
For some reason Nvidia 5** series cards out perform many of the newer Nvidia cards, they are comparitable in performance to AMD 79** cards.
 
jondi, thanks for the reminder. I remember reading that thread but for some reason I didn't recall it. Glad that you are using the pogs files. Any information or tips you can give on GIMPS would be much appreciated. We have a lot of FAH members looking for new projects to run and I have little experience outside of BOINC.
 
You're welcome Gilthanis.
I'm happy to advise if anyone gets stuck, one tip that springs to mind is when you download the TF work units it's best to change the exponent at the end otherwise you will constantly be feeding your client work, a WU like this for e.g. *F3,100404259,69,70 will take a minute or so to complete, if you change it to *F3,100404259,69,73 it will search through 69-70 then 70-71 then 71-72 then 72-73 each time taking twice as long as the previous and giving twice as many points (Ghz/days) as the previous, plus if you find a factor in one of the harder ranges you'll get a larger points bonus.
 
Back
Top