DC Vault 2

jondi said:
I don't know by what stats the DC Vault is going by but I've noticed both yours and Nickoftime's Muon1 yoyo work credit is not showing up here.
Click to expand...

I am actually waiting on a response on this from both Rusty and BOK. I would prefer that the stats not allow for double dipping, but if they do, every team should be doing it. I will update as I hear back.
 
jondi said:
I don't know by what stats the DC Vault is going by but I've noticed both yours and Nickoftime's Muon1 yoyo work credit is not showing up here.
Click to expand...

Wait - how am I 9th on the team? I didn't know I had even run this before! I wonder if this was one of my many command line set and forget programs...
 
jondi said:
I don't know by what stats the DC Vault is going by but I've noticed both yours and Nickoftime's Muon1 yoyo work credit is not showing up here.
Click to expand...

BOK said that he was just using the csv file provided to him, so he wasn't separating anything. So, I am now waiting to hear back from Rusty on how the DC-Vault sees the data. I will be glad if they do indeed report the stats separate because that is how it should be. With the info from BOK, I'm guessing the merging of the stats is just for their local stats listing at MUON1 and not necessarily at the Vault.
 
We took another spot at POGS and DPC have ramped up to half of our/his production according to the POGS website.

I also suggested renewing the potential of adding Primeboinca to the Vault. It was brought up September 2010 but didn't really have any response yay or nay. The project seems to run stable for me, so I am not sure why it wouldn't be included. If you have feeling one way or another, please take the time to drop in the thread and express your wishes or views. http://www.team-ninja.com/vbulletin/showthread.php?t=47951

NickOfTime, nice work you are doing there. Keep it up!
 
NickOfTime said:
http://www.stephenbrooks.org/muon1/stats/?allteams=1

digik-oz.nl isn't include the boinc-wrapper stats, So at muon1 site itself we have 11m more points, and I think we are contributing 500k per day at the moment...
Click to expand...

Yeah...but the MUON1 rankings don't effect our score per se. It is what is exported and counted by the DC-Vault stats. So, they may be separating them from the YoYo points as they truly should. I'm still waiting on an actual confirmation. So, The digik-oz.nl sight does represent our standing because those points are under YoYo.
 
Well, I am setup to run it either way, so I'll switch as the yoyo tasks complete to the manual version...
 
For simplicity, I have added links to NickOfTime's optimization files under the all inclusive DC list thread.
 
Ok, Updated the pogs sse2/generic version again, the -fast compile option was overwriting my -xSSE2 option. Changed to -O3 -static -axAVX,SSE4.2,SSSE3,SSE3,SSE2 and tested (Compiled on AMD, ran on intel).
(Hmm, since it's a zip file you will have to grant eXecute permission after extracting, chmod +x filename )
 
Last edited:
NickOfTime said:
Ok, Updated the pogs sse2/generic version again, the -fast compile option was overwriting my -xSSE2 option. Changed to -O3 -static -axAVX,SSE4.2,SSSE3,SSE3,SSE2 and tested (Compiled on AMD, ran on intel).
(Hmm, since it's a zip file you will have to grant eXecute permission after extracting, chmod +x filename )
Click to expand...

Ummm where is the file is it the last link you left. I need to try it on the G34's :)
 
Grandpa...I am also linking his files on the all inclusive DC list for easy reference.
 
NickOfTime

Congrats Red Rider sse2 early results at least a 100% increase in production :) and it looks like it is going to actually be better than that, I have only been running it for a short time 20 min. at this point, I will post after it has been running for a couple of hrs.

EDIT
looks like a 105% increase in production went from 2:45:00 for the longest WU's to 1:19:00 :cool:

Nice work NickOfTime ;)
 
Last edited:
Good, it looks like the sse2 version is generic enough to work everywhere. Working on another optimization that should give another 3x over my last version (when no dust is being used in the filters, which I haven't seen in my wu's yet)
 
Last edited:
I should clarify overall the AMD's are very close to the Intels clock for clock core for core now the 105% is gain over the last version you posted which I forgot to update the gain overall around a 250% to 300% increase I believe.
 
Ok, pogs files have been updated again (v3) x3 more speed :eek::eek::eek: , Longest runtime on my AMDs/Intels is 35m..

I switched from muon to pogs for a bit to validate :)
 
Last edited:
Hmm, 7 inconclusive wu and 150 validated, 555 pending validation. 4.5% error rate , I'll monitor today and try another version with more precise math and see how the wu error rate and runtime compare.

Ok, Added to zipfile fit_sed_x86_64-pc-linux-gnu_340 which is the gfortran (project like) compile with precise math...
http://pogs.theskynet.org/pogs/hosts_user.php?userid=13755
All wu's submitted after apr 7 13:30utc on my two ubuntu 12core vm will be that version...

Hmm, looks like 48m longest runtime for gfortran version...
 
Last edited:
DPC is now doing about 200k more than us at POGS for the April stampede. We might be able to hold them off at this pace, but if they keep ramping, they may over take us this month.
 
We took that last easy place in Leiden so I'm switching over my rigs to POGS now.

How do I go about using the optimised POGS whatsits? I have one Linux host and the rest are Win7 X64.
 
Windows machines
The files should go under the projects folder in your ProgramData\BOINC\ directory.
You should also write an app_info file so that you can restart your computer and not worry about the project over writing your files.

Linux
I don't recall exactly.
 
I can provide/update the windows version of pogs compiled with mingw32 later today, there is an app-info file in the Linux version already for those who boot their machine frequently...
 
Last edited:
NickOfTime said:
I can provide/update the windows version of pogs compiled with mingw32 later today, there is an app-info file in the Linux version already for those who boot there machine frequently...
Click to expand...

Now get to work on a version to make ARM super fast. :D
 
Hmm, well the openCL port also had cpu only section that they were using to compile for raspberrry pi... I could give you updated cpp source that you could compile yourself on your ARM...
 
Last edited:
NickOfTime said:
Hmm, well the openCL port also had cpu only section that they were using to compile for raspberrry pi... I could give you update cpp source that you could compile yourself on your ARM...
Click to expand...

I will pass. I only have a couple that can run it and they aren't going to be mass producers. I was more or less just kidding about the ARM apps. :)
 
Not for nothing, but openCL for tegra 4 and K1? Imagine using those Kepler and soon maxwell cores on arm! Would that be possible? I'll buy you a K1 development kit if you are interested in developing BOINC apps for it...
 
pogs validation update, Looks like Faster Ifortran version error rate is normalizing better :)
Looking at AMD vm work units....
Ifortran 5 inconclusive, 253 validated. Error rate fallen to 2% from 4.5%
gfortran 1 inconclusive, 36 validated. Error rate 2.7%, still a bit early to compare...


Hmm, Now at 1.5% for each..
 
Last edited:
Kai, that is probably still to be seen on whether the GPU cores within ARM will be usable. I would venture that they would have to develop a GPU app wouldn't they?
 
Kai said:
Not for nothing, but openCL for tegra 4 and K1? Imagine using those Kepler and soon maxwell cores on arm! Would that be possible? I'll buy you a K1 development kit if you are interested in developing BOINC apps for it...
Click to expand...

Hmm, let me finish playing with the openCL version for my keplers (660ti/690) first, as I am debugging it, I keep finding optimizations that can be backported to the fortran version, resulting in a cpu version that is running as fast as my previous (n-1) opencl version... fortran is around 70s per step now and opencl version was 40s per step , but I haven't applied the last NoDust change to it yet...
 
Gilthanis said:
Kai, that is probably still to be seen on whether the GPU cores within ARM will be usable. I would venture that they would have to develop a GPU app wouldn't they?
Click to expand...

They should be usable. The K1 developer platform comes with all the necessary software to utilize the GPU. They're marketing it as a CUDA platform for such things. I'd imagine it would take a Linux arm port with CUDA code to use those cores. Something far beyond my expertise. But NickofTime, if you'll develop it, I'll buy you a K1 developer board. Also I'll send you an AMD card if you're interested in developing for that as we'll. you're an extremely talented and valued member of the [H]orde. Thank you for all you do!
 
I'm trying NoT windows x64 stuff right now with an i2p. Will see how it goes.
 
You should just be able to extract the files into the pogs folder.
On Windows x64 it is by default:
C:\ProgramData\BOINC\projects\pogs.theskynet.org_pogs
 
Gilthanis said:
You should just be able to extract the files into the pogs folder.
On Windows x64 it is by default:
C:\ProgramData\BOINC\projects\pogs.theskynet.org_pogs
Click to expand...

I did and the new files do not override the old files. Yes, I did quit the application. I suspect you have to replace files and rename them.
 
You must log in or register to reply here.
Back
Top