11/10-16/10 The Moon*Shine Challenge-Einstein

Alez · Post by **Alez** » Tue Sep 16, 2014 4:00 pm

http://boincstats.com/en/stats/challenge/team/chat/582

Name The Moon*Shine Challenge
Status Upcoming
Project Einstein@Home
Issued by Sicituradastra.
Start time 2014-10-11 04:30 UTC
End time 2014-10-16 04:30 UTC
Late entrants allowed? Yes
Number of teams participating Â» 12
Number of users participating 0

Post by **scole of TSBT** » Tue Sep 30, 2014 2:24 am

Two GPU challenges at the same time. I think we're AMD GPU heavy so I hope this is one we're going to attack.

Post by **scole of TSBT** » Thu Oct 09, 2014 2:12 am

Janos, we need a good lead-off hitter. You in?

Post by **scole of TSBT** » Fri Oct 10, 2014 2:33 pm

I've added this entry to my hosts file...
192.168.1.241 einstein.phys.uwm.edu
...it won't downlod any new WUs. but it still uploads completed WUs.

What am I missing? I'm pretty sure it's not a hardware installation problem this time.

Silver · Post by **Silver** » Fri Oct 10, 2014 6:20 pm

Are you trying to block uploading units?
Is that ip you quoted not an internal address in your network to get to your router?
I think 129.89.61.70 is the ip you need for the einstein website

Post by **scole of TSBT** » Fri Oct 10, 2014 7:07 pm

I'm trying to block uploading. That's my IP network, but nothing at that address. I rebooted one system and that seemed to work.

Alez · Post by **Alez** » Fri Oct 10, 2014 7:19 pm

Challenge starts in less than 2 hours, not worth bunkering on Einstein anyway in my opinion as you will have so many pendings anyway. Too many people on Einstein hoard units with ridiculous caches like mad preppers in the off chance that the server isn't available for a few hours.

Post by **scole of TSBT** » Fri Oct 10, 2014 7:41 pm

Challenge site says 9 hours 48 minutes...

Alez · Post by **Alez** » Sat Oct 11, 2014 12:34 am

Your right, GPUGrid starts first. Still don't think it's worth bunkering though

Post by **scole of TSBT** » Sat Oct 11, 2014 1:04 am

So what is the best optimization setup for Einstein@home? Looking for AMD and nvidia GPU advice too. Time to put a GTX 780 to work.

Alez · Post by **Alez** » Sat Oct 11, 2014 1:22 am

It was discussed a bit on this thread http://www.dunadd.co.uk/seti/forum/view ... =17&t=1750 Some apps pay better than others and running more apps than 1 on a GPU is dependent on the individual GPU involved. There is no way to alter the app. Overclocking is , as always, an option but again mileage varies with individual systems. Need to figure out which apps pays best , currently, and run those and then try different numbers of apps per GPU on your own system and see if it makes a difference. I remember my 660ti's were better at 2 apps, my 650 at 1 and all my ATI's were better at 1.

Janos (retired) · Post by **Janos (retired)** » Sat Oct 11, 2014 11:01 am

Well the Einstein WU's keep crashing my computer and I don't have the time to babysit. I will bow out and revert to Milkyway.

Alez · Post by **Alez** » Sat Oct 11, 2014 1:42 pm

Not sure what is going on . I have more than 10 various GPU's on this and I am making more credit per hour on the Enigma challenge

Anyone figured out the best units time / credit rating yet ?
Janos, do you know why they are crashing ? Have they upped the memory requirements or any clue. Maybe your crashes and my crawling in the slow lane are related.

Post by **scole of TSBT** » Sat Oct 11, 2014 7:00 pm

On my duel 7970 rig, I'm running 2 WUs per GPU with the app_config.xml below.
The Binary Radio Pulsar Search (Arecibo, GPU) v1.39 (BRP4G-opencl-ati) app pays 1000 credits for avg 2200 sec crunch time.
The Binary Radio Pulsar Search (Perseus Arm Survey) v1.39 (BRP5-opencl-ati) app pays 3333 credits for avg 6000 sec crunch time. Which pays a little better, but not that much.
I just eyeballed what I thought the avg run time was per WU, so not real scientific.

Running 3 x 7970, a R9 270, a GTX 780 and a GTX 460 and logged over 200k on this on this challenge so far.

<app_config>
<project_max_concurrent>8</project_max_concurrent>
<app>
<name>hsgamma_FGRP3</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP4G
</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP5
</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

Alez · Post by **Alez** » Sat Oct 11, 2014 7:16 pm

Without bragging, I am running more crunching power than you have on the competition and I have got 60 k and 300 odd pending.

Not running a config or anything and straight 1 unit per GPU. Going to have to check round the machines again. Something must be wrong or I am being very unlucky.
I'm way down in 41st and TSBT in 10th

we should be 7th at least

Post by **scole of TSBT** » Sat Oct 11, 2014 10:15 pm

Are your run times in range or is all your credit tied up in pending? My GTX 780 runs the Arecibo, GPU WUs in around 2700 secs while running 2 at a time and Perseus Arm Survey GPU WUs in around 8000 sec while running 2 at a time.

Alez · Post by **Alez** » Sat Oct 11, 2014 11:30 pm

Hard to tell as I have so many different types of cards scattered around, but even my 7970's seem to be struggling to put in decent numbers. What are you getting for your 7970's ? I have a feeling that the BRP5 units are messing me up, but not sure at all. Going to check them again.

On a happier note BU is back up and running.

Post by **scole of TSBT** » Sat Oct 11, 2014 11:51 pm

The Binary Radio Pulsar Search (Arecibo, GPU) v1.39 (BRP4G-opencl-ati) ~2200 sec per WU with 2 WU running.
The Binary Radio Pulsar Search (Perseus Arm Survey) (BRP5-opencl-ati) ~6000 sec per WU with 2 WU running.

Silver · Post by **Silver** » Sun Oct 12, 2014 12:25 am

I think I've got some config files sorted out and now appear to be crunching Einstein again.
Hopefully I'll be able to help the cause now!

Alez · Post by **Alez** » Sun Oct 12, 2014 12:37 am

I'm taking double that on my 7970 with only a single unit running. Going to have to figure this out a bit. Might be project reset etc. Have rebooted first, see where it goes from there. Doubt it will be tonight though, knackered.

Post by **scole of TSBT** » Sun Oct 12, 2014 12:45 am

Only other things I can think of is driver version, 13.12 on one, 14.4 on the other. I'm leaving a little CPU wiggle room too, less than 1 core I think.

Alez · Post by **Alez** » Sun Oct 12, 2014 2:51 am

Drivers are fine. 13.12 is on the comp with the 6990 and 6970. Need that as otherwise milkyway won't run on it. 14.4 has always worked fine before so don't see any point in updating. All systems with GPU's are running with at least 1 free core. Seems that Einstein has changed how it ran before. Stay with the comp, I will figure it out.
So far waiting for the times to come through, but up to 4 units per GPU on the 7970's and no drop in performance and load still under 90%. Previously ran better with 1 or 2 apps per GPU. Looks like there is much more headroom now.
We will crack this comp. Don't really see the GPUGrid comp as worth the effort. Far more teams here and moon would be disappointing if we don't put in our usual punching above our weight effort :)
I've almost managed to double my output and slowly moving up. Hopefully more to come

Post by **scole of TSBT** » Sun Oct 12, 2014 2:09 pm

Rob, your 7990 is getting fat and lazy. How about giving it an Einstein workout?

Alez · Post by **Alez** » Sun Oct 12, 2014 2:35 pm

Scole, what driver are you using for the 7970 ? Think I will also try updating my nVidia's to the latest version. I notice that the GPU usage is up and down like a yoyo. Might also try pulling the 650 out as maybe I am simply using up too much bandwidth on the x16 slots having 4 cards. Don't know. Got an increase in productivity but still not correct. An awful lot of pendings to which ain't helping much. Open to other suggestions guys

Post by **scole of TSBT** » Sun Oct 12, 2014 2:51 pm

CCC 13.12 on one, 14.4 on the other. Nvidia 344.11, but going to drop back to 337.88. Binary Radio Pulsar Search (Perseus Arm Survey) WUs jumped from 5000 secs to 8000 secs after upgrading to 344.11.

Alez · Post by **Alez** » Sun Oct 12, 2014 3:20 pm

Right will try and change the nvidia driver. Strange that I have the same drivers on the 7970's as you with a free core etc. but seem to be running slower. The brp5 units especially seem to be very variable in times. Really pi$$ing me off. Shed time before the last BTCC race of the season

Post by **scole of TSBT** » Sun Oct 12, 2014 3:32 pm

I'm sure you've checked, but no harm asking. Are the 7970s are clocked correctly?

Post by **Megacruncher** » Sun Oct 12, 2014 5:41 pm

Sorry guys but I'm not in a position to help with this.

Post by **scole of TSBT** » Sun Oct 12, 2014 7:11 pm

scole250 wrote:CCC 13.12 on one, 14.4 on the other. Nvidia 344.11, but going to drop back to 337.88. Binary Radio Pulsar Search (Perseus Arm Survey) WUs jumped from 5000 secs to 8000 secs after upgrading to 344.11.

I double checked this and I must have still been half asleep. 344.11 does not appear to have caused a performance problem. The 5000 sec WUs were when only 1 WU was running. 8000 with 2 at a time.

Alez · Post by **Alez** » Sun Oct 12, 2014 11:35 pm

Never the less I went to 337.88 and my GPU's stopped yoyoing on the usage and are now sitting steady at 95% usage, so it's definitely better for me.

The 7970's are actually slightly overclocked. More digging to be done on this issue.

Post by **robmacagain** » Mon Oct 13, 2014 8:35 am

Is there optimization for this as i am taking 5,850 seconds per task on BRPS, and my GPU's are sitting at only 51%, Alez is saying his/yours are 95%???

My reasons for not doing this normally is the amount of time it takes to get your credit, i'm an instant gratification type

.

Post by **scole of TSBT** » Mon Oct 13, 2014 10:27 am

Should have prefaced with, to get the best optimization, you need to run 2 WUs per GPU. In your case 4 total. It will increase individual WU run time to ~7000 sec, but you're doing 2 WU in 7000 sec instead of 1 WU in 5000 sec. There's a way to control that from your einstien@home website preferences, but I use an app_config,xml. I'd go with an app_config like this...

<app_config>
<project_max_concurrent>4</project_max_concurrent>
<app>
<name>hsgamma_FGRP3</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP4G
</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP5
</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

Post by **robmacagain** » Mon Oct 13, 2014 12:33 pm

Done this >

<app_config>
<project_max_concurrent>4</project_max_concurrent>
<app>
<name>hsgamma_FGRP3</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP4G
</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
<app>
<name>einsteinbinary_BRP5
</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

Its still only doing 1 each?

Stopped and restarted boinc, didn't work so rebooted the pc and still didn't work???

Alez · Post by **Alez** » Mon Oct 13, 2014 1:46 pm

Rob, I'm using it and it does work. You need to copy the text and paste into a notepad file. Save as app_config.xml ( type all files ) and save in the einstein project folder under c:/ProgramData/Boinc/Projects
A common error is that the file get saved as app_config.xml.txt
Also it needs to be notepad. Windows will try and use an xml editor, but these files are not true xml.

Stop Boinc and restart. Check the output on restart. It should state found app_config file for einstein. The running units should also change to running 0.5 (app) on 0.5 ( device number )

You can also do it through your project preferences by editing einstein@home preferences and changing gpu utilization factor to 0.5 though this will cause you to run 2 apps per GPU but will not change how many CPU cores to support so each app will still grab 0.9 CPU core. Doesn't matter if you are not running cpu tasks. If you want to run 3 per device then 0.33, 4 per device 0.25.
If you do it this way after editing your preferences then issue an update from your client and it should work.

Post by **robmacagain** » Mon Oct 13, 2014 2:23 pm

HI, i use XML Notepad to create app_config files so there defo XML and save them to the project folder, but it still hasn't worked, I have now done it with project preferences and as soon as it has completed my next BU unit i will restart it again and report back.

Post by **scole of TSBT** » Mon Oct 13, 2014 2:25 pm

There's something else going on with Rob's system too. It's taking over 5000 sec for the Binary Radio Pulsar Search (Arecibo, GPU) WUs. Should be able to do those in ~2200 secs or less, running 2 WU at a time. What Catalyst Control Center version you running? The best numbers I'm getting are from the one running 14.4. Had one running 13.12 and had horrible times. Just got that one on 14.4 and waiting to see if that helps. Could be something else but they are almost identical systems.

Post by **robmacagain** » Mon Oct 13, 2014 2:42 pm

Am using 14.10 it updated itself t'other day.

Have done it through project pref's re booted too and it's still only doing 1 per gpu core and says "0.5 cpu's and 1 ATI GPU per wu

.

Post by **robmacagain** » Mon Oct 13, 2014 2:49 pm

Just out of curiosity i suspended Einstein and did a quick bash of MW and it did 12 units @ 4 per gpu and in 97 seconds per 8 so

.

Post by **scole of TSBT** » Mon Oct 13, 2014 3:14 pm

I'll connect up via a remote support session and take a look at the config if you want.

Post by **robmacagain** » Mon Oct 13, 2014 3:28 pm

If you like, MacLennan-PC Computer drive C:, gis a bit of warning though

.

Post by **robmacagain** » Mon Oct 13, 2014 3:52 pm

Am set up now.

Post by **scole of TSBT** » Mon Oct 13, 2014 3:53 pm

Go to http://showmypc.com/, click the green ShowMyPC button/link, let it download the .exe file and open/run it. When the program window appears, click the Show My PC button there. It will generate a number. Send it in a PM. Don't post here else anyone could jump on.

Alez · Post by **Alez** » Mon Oct 13, 2014 4:00 pm

Rob, you need to use the plain ascii txt editor notepad in windows. You can't use an xml editor as these leave tags in the file. The file is called app_config.xml but it is just a confusing naming convention. The file is actually plain text and must be so as boinccmd can't handle html tags.
No idea why they do it that way but that is the way it must be. The file must be plain ascii only.

Post by **robmacagain** » Mon Oct 13, 2014 4:00 pm

Done, 5 attempts to send though, still being rubbish this forum!!!

Alez · Post by **Alez** » Mon Oct 13, 2014 4:03 pm

robmacagain wrote:Done, 5 attempts to send though, still being rubbish this forum!!!

Don't think they have done the server migration yet at our host. Hopefully that will be fixed soon enough.

Post by **scole of TSBT** » Mon Oct 13, 2014 4:09 pm

Well I checked and Rob's app_config.xml is set correctly. I'm stumped.

Alez · Post by **Alez** » Mon Oct 13, 2014 4:16 pm

is use all GPU's set in the config file in boinc ? the 7990 is a dual card and I know that from my 6990 that sometimes confuses. Are there enough units downloaded to run ? When you set the app_config and restart boinc does it have an error in the startup list for the app_config, does it say 1 found and do the units running reflect 0.5cpu,0.5 gpu ?

Post by **robmacagain** » Mon Oct 13, 2014 4:20 pm

This >

<no_proxy></no_proxy>
</proxy_info>
<rec_half_life_days>10.000000</rec_half_life_days>
<report_results_immediately>0</report_results_immediately>
<run_apps_manually>0</run_apps_manually>
<save_stats_days>30</save_stats_days>
<skip_cpu_benchmarks>0</skip_cpu_benchmarks>
<simple_gui_only>0</simple_gui_only>
<start_delay>0.000000</start_delay>
<stderr_head>0</stderr_head>
<suppress_net_info>0</suppress_net_info>
<unsigned_apps_ok>0</unsigned_apps_ok>
<use_all_gpus>1</use_all_gpus>
<use_certs>0</use_certs>
<use_certs_only>0</use_certs_only>
<vbox_window>0</vbox_window>
</options>
</cc_config>

No errors, it's just being a c*** to be honest, restarted it re read config files re booted, scole looked at it, all is how it should be it just won't play the bas!!!

Oh aye there is plenty of units down too.

Alez · Post by **Alez** » Mon Oct 13, 2014 4:38 pm

Freaking boinc eh ? These things are sent to pi$$ us off and sometimes I wonder why we bother. It runs milkyway multiple units so should be no issues. I'm about to re-fire up one of my linux computers and put that on Einstein and BU. Guess that will be fun to.
I'm now as stumped as scole

Try deleting the app_config totally from the project folder and doing it again through the project. Issue a project update through boinc manager then shut boinc down and restart. Let it finish the unit it is working on and see if it picks up at 2 per card or suspend the units it is running and see if it starts 2 per card after that.The app_config in the folder overrides the project settings. Failing that try remaking the app_config and try again.

Post by **robmacagain** » Mon Oct 13, 2014 4:59 pm

Well after all the fooin about i have came to the conclusion, Donald Duck it

.