DC Vault 2

I didn't get much of a response, so I'm switching the majority of my CPU resources to a mix of Docking and eOn. Docking should be wrapping up here soon. There is no way I'm hitting the 1 million point mark, but last I looked, I was ahead of Jathanis! :D
 
I've switched to eOn for now, as they take around 8 minutes a work-unit on my E5-2695V2.
 
Still Continuing GIMPS and muon dpad processing, I see dmaine and tjmagneto are also processing some muon now. :)
 
I still have 2 dedicated and 2 shared cores on GIMPS as well as 2 cores dedicated to SOB. I still have resources elsewhere, but most of my CPU resources are shifting to eOn. GPU's are still spread out in various places. I still need to move a couple GPU's physically around to simplify things...
 
Gilthanis said:
I still have 2 dedicated and 2 shared cores on GIMPS as well as 2 cores dedicated to SOB. I still have resources elsewhere, but most of my CPU resources are shifting to eOn. GPU's are still spread out in various places. I still need to move a couple GPU's physically around to simplify things...
Click to expand...

I have not had real good luck with eon make sure you keep an eye on it. I have had times it will run for days straight then the next thing you know it will get a batch of bad WU's and it will just quit with a bunch of errored WU's waiting to send. I have tried it a few times with the same results each time. I end up with an unacceptable amount of down time when running it. Unfortunately I am not real good at baby sitting work no do I tolerate it very well. In my opinion the Eon developer needs to fix his product. ;)
 
Gilthanis said:
I didn't get much of a response, so I'm switching the majority of my CPU resources to a mix of Docking and eOn. Docking should be wrapping up here soon. There is no way I'm hitting the 1 million point mark, but last I looked, I was ahead of Jathanis! :D
Click to expand...

Yeah, not sure what dirty underhanded trick you pulled to keep pulling in 5-10K ppd in WU - I've had it as my main project and POGS set to 0 as backup, and I only pull 1-2 WU a day at best! :D I'm just hoping to get to a nice even 650K before it ends... For the time being, for CPUs, I'm going to aim for the 2.5M mark in POGS, then probably move on to Yoyo, unless we decide on a group push somewhere. Trying to decide if I want to get DistrRTgen up to 10M or even 25M, or keep pushing on GPUGrid... But laziness says it'll probably stay GPUGrid!

On a side note, my 10 year old daughter told me yesterday that she was sad Docking is ending - because she likes to watch the screensaver while she's doing her homework, lol...
 
I'm gonna shut down GPUGrid here for a couple of days. Temperatures here are hitting 90F and more for the next few days. The 780ti I'm using is in the house and it's getting pretty warm in here.
I guess that the good news is I'm adding a few more cores from the 980x toward muon.
 
Eon PSA: Now its been a while since I have ran eon but formerly it performed MUCH better on Linux. So to anyone doing some crunching over there you might compare WU times/credit to see if this is still the case.

I remember spinning up Linux VMs on a Windows host and still doubling the PPD with the same CPU power. Hopefully they have fixed the windows client by now. Crunch on.
 
Kai said:
I've switched to eOn for now, as they take around 8 minutes a work-unit on my E5-2695V2.
Click to expand...

metallicafan said:
Eon PSA: Now its been a while since I have ran eon but formerly it performed MUCH better on Linux. So to anyone doing some crunching over there you might compare WU times/credit to see if this is still the case.

I remember spinning up Linux VMs on a Windows host and still doubling the PPD with the same CPU power. Hopefully they have fixed the windows client by now. Crunch on.
Click to expand...

I believe you are right, I am asuming Kai's 2695's are in windows and they should be faster than my 4650's I run Linux and I do a Eon WU in around 4 minuets with them so just about twice as fast as Kai is doing them, either that or Kai has a setup problem. :eek:
 
I'm helping tjmagneto on Enigma at the minute, looks like some quick progress can be made there. My two Linux hosts are on eOn but they're nothing special.
All over the place with work at the minute so last night was the first chance I had to do anything DC for a while. Looks like I need to upgrade my MJ-12 node tonight as well, maybe that'll help its recent problems.
 
Grandpa_01 said:
I believe you are right, I am asuming Kai's 2695's are in windows and they should be faster than my 4650's I run Linux and I do a Eon WU in around 4 minuets with them so just about twice as fast as Kai is doing them, either that or Kai has a setup problem. :eek:
Click to expand...

Well poop! I'll monitor temps and everything else, but they've been a stable 8 minutes, not for nothing. Not surprising that linux is better. I'll try some VM's to see if there is a difference too. Though my 2695's are at the stock 2.4, and if you have spicy 4650's, eOn might be sensitive to clock speed (you know you love pushing those 4650's!). Hmmm.
 
Jathanis said:
Yeah, not sure what dirty underhanded trick you pulled to keep pulling in 5-10K ppd in WU - I've had it as my main project and POGS set to 0 as backup, and I only pull 1-2 WU a day at best! :D I'm just hoping to get to a nice even 650K before it ends... For the time being, for CPUs, I'm going to aim for the 2.5M mark in POGS, then probably move on to Yoyo, unless we decide on a group push somewhere. Trying to decide if I want to get DistrRTgen up to 10M or even 25M, or keep pushing on GPUGrid... But laziness says it'll probably stay GPUGrid!

On a side note, my 10 year old daughter told me yesterday that she was sad Docking is ending - because she likes to watch the screensaver while she's doing her homework, lol...
Click to expand...

The trick was simple. I ran VM's on my file server setting them to only request work from Docking, QOS, and WUProp. That way they would cache up whatever work they could. Then the file server crunched whatever I felt like throwing on it with all cores. That way when I was out of docking, full processing went to the main system. When there was Docking work units, the processing got shared with all the VM's. The VM's work may have been slower, but it got me work. In your setup, if you already had work, you weren't asking for more. So you limited the opportunity to request work by having that backup project on your client. It isn't a pretty setup, but it gets you what you need without idle cores/threads. You can use that technique to get BETA work at WCG as well... :cool:

Edit: Also, if you continue with the setup you had, it would be best to have a quick return work unit like eOn as your backup project so that you will request work more often...
 
Last edited:
Thanks to NickOfTime for writing up a guide for PSP and Muon1. They have been linked to the all inclusive DC list. :)

We still need guides for RC5 and OGR-28 if anyone has the time and interest.
 
I'm still playing around with those, trying to figure them out come felt. So hopefully I can write something soon, unless someone else already has something in the works.
 
Grandpa_01 said:
I believe you are right, I am asuming Kai's 2695's are in windows and they should be faster than my 4650's I run Linux and I do a Eon WU in around 4 minuets with them so just about twice as fast as Kai is doing them, either that or Kai has a setup problem. :eek:
Click to expand...

Huzzah! You're right! I was getting 8-11 minutes in Windows 7 and it looks that even in a VM I'm getting 3 minutes!
 
Maybe a little sooner, sure, looks like another place on top of those should be fairly soon after too, so a fairly quick and easy 4 places. We're #5 in RAC right now too :)
 
2 290x setup and running gimps, I think I may have doubled my ghz per day and might catch up to brilong :)
 
brilong said:
Thanks, you all did great too! It was my first time participating in a challenge of any sort and I enjoyed it. :D
Click to expand...

And if you are like me it will not be your last, speaking of challanges, the next on is on the 20th it is an 8 day challange :) http://www.primegrid.com/ and we stand a real chance of placing in the top 2 or 3 of the challange series for the year if we win enough of these. :cool:
 
Grandpa_01 said:
And if you are like me it will not be your last, speaking of challanges, the next on is on the 20th it is an 8 day challange :) http://www.primegrid.com/ and we stand a real chance of placing in the top 2 or 3 of the challange series for the year if we win enough of these. :cool:
Click to expand...

I've disabled hyper-threading on two of my I4P's which I'm using to test PRP. I imagine they'll do a lot better for the next PrimeGrid challenge. :)
 
brilong said:
I've disabled hyper-threading on two of my I4P's which I'm using to test PRP. I imagine they'll do a lot better for the next PrimeGrid challenge. :)
Click to expand...

What is the ms per bit that you are getting?
I'm seeing 46ms on my 6234, and 32ms on i7-4770s with ht on...
 
Yep all of the primegrid work I have tested so far does better with HT disabled on the I4P's but the % varies PSP work has a rather large difference between HT Enabled and disabled iirc it was 3x faster with HT disabled.
 
We took a spot at Majestic-12 today. That was a nice 50 pointer. Good job guys. :)

The next position will be quite a ways off with our current production. We certainly need more people with unlimited data to hop on board. :D

Looking at Free-DC, we will be over taken by 3 teams before we take our next position at our current rate. However, I'm not sure how up to the moment Free-DC's stats for Majestic-12 is. http://stats.free-dc.org/stats.php?page=team&proj=maj&team=147
 
Last edited:
http://pogs.theskynet.org/pogs/forum_thread.php?id=436#2224

TheSkyNet POGS has been chosen as the marathon project in the BOINC pentathlon! To celebrate, we've got some great prizes to give away - including some free Lego. More details to come on how these prizes will be given out to pentathlon participants. Thanks for being part of theSkyNet!

-theSkyNet
____________
Cheers
Alex
theSkyNet.org webMaster
ICRAR
Click to expand...
 
NickOfTime said:
What is the ms per bit that you are getting?
I'm seeing 46ms on my 6234, and 32ms on i7-4770s with ht on...
Click to expand...

I'm running 32 threads and each of my clients/windows shows a different ms per bit. Here are some sample excerpts:

Code: 
222113*2^16102541+1, bit: 2880000 / 16102558 [17.88%].  Time per bit: 38.672 ms.

222113*2^16102581+1, bit: 3560000 / 16102598 [22.10%].  Time per bit: 22.606 ms.

168451*2^16102728+1, bit: 4080000 / 16102745 [25.33%].  Time per bit: 24.548 ms.

222113*2^16102821+1, bit: 3040000 / 16102838 [18.87%].  Time per bit: 29.238 ms.

222113*2^16102901+1, bit: 4340000 / 16102918 [26.95%].  Time per bit: 14.592 ms.

This is from a 4x E5-4650 host running stock clock (all-core turbo at 3100MHz).
 
Gilthanis said:
We took a spot at Majestic-12 today. That was a nice 50 pointer. Good job guys. :)

The next position will be quite a ways off with our current production. We certainly need more people with unlimited data to hop on board. :D

Looking at Free-DC, we will be over taken by 3 teams before we take our next position at our current rate. However, I'm not sure how up to the moment Free-DC's stats for Majestic-12 is. http://stats.free-dc.org/stats.php?page=team&proj=maj&team=147
Click to expand...

Nice, 50 points is not to be sniffed at! Looks like the new client is more stable than the older one, it's been running solidly for longer than before without a restart and, although it's hard to tell as it does naturally fluctuate, looks like my average download is slightly up, from low 3.x to around 3.8mbps.

Still need to get around to setting a node or two up at home, things have been hectic but this weekend is set aside for computer things - mainly trying to clear out stuff I don't need to raise some cash (mainly for my car) but also to do some maintenance and swap some things around. Not many of the GPU projects have been interesting me so I'm contemplating downsizing my GPUs (not that I have all that many) and seeing if I can bring on some more CPUs. Having said that I've seen a GTX590 for a very good price which I'm sorely tempted by...
 
http://pogs.theskynet.org/pogs/forum_thread.php?id=413#2536

Kevin finally made a statement...

The problem is quite simple. Some of the custom clients are returning the wrong answers. This is due to two factors: firstly problems with the compiler optimisation on some platforms and secondly people modifying the code but not knowing how SED fitting works. 99% of the time they give the right answer, but the ones I've investigated when there is a mismatch are giving the wrong answer.

If you find one tell me about it and I'll stop it processing.

One of our volunteers has been working on optimising the code for me (as I'm flat out on the SKA work). I'll be testing the CPU and GPU code against 250,000 pixels across a large range of red-shifts to make sure the results are the same.
____________
Regards
Kevin
-----
International Centre for Radio Astronomy Research
Click to expand...

He also didn't point out whose optimized clients that he detected being bad either. I was hoping for a little more confirmation on whether he thought the one NickOfTime had was problematic or not. By the way, how's that coming? :D
 
Anyone else notice anything odd in these photos.....






And no that is NOT a 750 Ti.....
 
Gilthanis said:
http://pogs.theskynet.org/pogs/forum_thread.php?id=413#2536

Kevin finally made a statement...

He also didn't point out whose optimized clients that he detected being bad either. I was hoping for a little more confirmation on whether he thought the one NickOfTime had was problematic or not. By the way, how's that coming? :D
Click to expand...

I got distracted last week building another boxen :)

Well, I did have one compiled version that was using fast math and was letting the validator reject the ones it didn't like...
But for the Faster No-dust model, I suspect he is just checking the results file and not seeing an exact match against the older version, and he has not looked at/tried the fortran code yet...

For comparision that last release fortran version with safe math was running at 5x with no dust filters, and I had provided Kevin with a 10x version.

but the c++ and GPU port , I am currently working on switching it from working on a single observation at a time, 24 times. To working on all 24 observations at a time.

Still have to finish the new method of writing the output file... but CPU version no dust filters is running about 30x and I can get that a bit faster. The GPU version is still single step at the moment, but was processing with dust filters 15x on a gtx660ti.
 
The nForce 750a is the onboard. It has an 8400 and a 9600 GT as physical cards. I'm guessing something with the SLI is allowing it to run the CUDA some how. Very odd....
 
DooKey, I think your CPID is screwed up again...take a look at your stats....
 
You must log in or register to reply here.
Back
Top