Moo Runtimes

[Fire$torm]

I was looking at my completed tasks on Moo's site and discovered some HUGE variation in run times. From a low of 2,300+ sec. to over 13,000+ sec. And the amount of time varies by date.

Nexus (My Desktop):
ID#- 297 (http://moowrap.net/show_host_detail.php?hostid=297)
List: Completed Tasks (http://moowrap.net/results.php?userid=1391)

Clocks
CPU: 2.8Ghz (Stock)
ATI 5830: GPU = 950Mhz / MEM = 900Mhz

Projects:
WUProp
FreeHal
MooWrap
PG-PPS (Sieve)
YoYo (Now set to NNT)
Aqua

Any ideas?

Note: I just up'd the 5830 clocks to 1000/1000 (Which MW did not like).

Edit: I put Odin (w/2x 4850) on Moo a short time ago and it completed its first wu in 4,141.21 sec.

4850 clocks: GPU = 700Mhz / MEM = 650

2nd Edit: WUProp Data on ATI Hd 5830 here. (http://wuprop.boinc-af.org/results/gpu.py?fabricant=ATI&type=Evergreen+%28HD+5xxx%29+series&modele=Radeon+HD+5830&plateforme=win&tri=projet&sort=asc)

[dan]

Do you have one free core that is doing nothing. I've seen the same when I have all my cores working on a CPU task. So on my i7 I have 75% as the setting for "On multi-core systems use at most". With MOO it then runs 100% of the cpu.

Dan

[Bryan]

F$, there are several things with Moo.

1. The time shown in the "Run Time" may or may not be accurate. On single GPU systems I've found it to be close on dual GPU it may or may not be depending on the system.
2. To get accuracte run time you have to open the wu record "Task" and subtract start time from end time.
3. On a single GPU you need to reserve a thread for Moo. On a dual GPU you need to reserve a FULL CORE.
4. You are running your wu on "core 0". On all my ATIs systems it is MUCH MUCH better to run on "Core 3". You set that in project preferences.

You can run a benchmark that will show you how it will perform on each of your cores. Then use that to select the correct "core" in prefernces.

1, Stop BOINC
2. Call up the command prompt.
3. Type in the path name of the executable "dnetc)1.02_windows_intelx86_ati14.exe -bench". So it will be something like C:\program data\boinc\project\moo wrap\dnetc.... . exe -bench
4. Hit enter.

The program will run and show you the speed of each core on your processor (relative only to Moo). In Project Prefernces that is the one you want to use. Makes a BIG BIG BIG difference in credits.

If you do a search on the Moo forums using the executable name you should be able to find the procedure.

[Fire$torm]

First, thanks for the feedback.

Since the system in question is my desktop I have BM set to use at most 75% of processors. I tried running the -bench switch with dnetc_1.02_windows_intelx86__ati14.exe but no go. Then I tried the switch on dnetc518-win32-x86-stream.exe, Success!!!


dnetc v2.9109-518-GTR-10092921 for ATI Stream on Win32 (WindowsNT 6.1).
Please provide the *entire* version descriptor when submitting bug reports.
The distributed.net bug report pages are at http://bugs.distributed.net/

[Jul 06 16:39:06 UTC] RC5-72: using core #0 (IL 4-pipe c).
[Jul 06 16:39:11 UTC] RC5-72: Benchmark for core #0 (IL 4-pipe c)
0.00:00:02.77 [1,600,863,234 keys/sec]
[Jul 06 16:39:12 UTC] RC5-72: using core #1 (IL 4-pipe c alt).
[Jul 06 16:39:19 UTC] RC5-72: Benchmark for core #1 (IL 4-pipe c alt)
0.00:00:04.75 [920,051,200 keys/sec]
[Jul 06 16:39:19 UTC] RC5-72: using core #2 (IL 4-pipe 2 threads).
[Jul 06 16:39:25 UTC] RC5-72: Benchmark for core #2 (IL 4-pipe 2 threads)
0.00:00:03.38 [1,332,476,140 keys/sec]
[Jul 06 16:39:25 UTC] RC5-72: using core #3 (IL 4-pipe cs-1).
[Jul 06 16:39:30 UTC] RC5-72: Benchmark for core #3 (IL 4-pipe cs-1)
0.00:00:02.77 [1,613,737,911 keys/sec]
[Jul 06 16:39:30 UTC] RC5-72 benchmark summary :
Default core : #0 (IL 4-pipe c)
Fastest core : #3 (IL 4-pipe cs-1)
[Jul 06 16:39:30 UTC] Core #3 is marginally faster than the default core.
Testing variability might lead to pick one or the other.

I changed the project preferences on the Moo site so now its wait and see....

F$

[Bryan]

See what happens when you listen to an old guy! :)) I thought it was the ati executable but it has been along time since I did that. I'm glad you figured it out!

[Fire$torm]

See what happens when you listen to an old guy! :)) I thought it was the ati executable but it has been along time since I did that. I'm glad you figured it out!

Hahaha Very funny :P At least you got me pointed in the right direction, so thanks.

Just to let you know, a couple of people posted the same suggestion on the Moo forum. So your not alone :)

Good news. It worked. The elapsed time shown in BOINC Manager dropped to under 40 minutes. :-bd

I should have realized something was off much earlier because the 5830 temps were around 59C before the change. Now it sits between 68~70C. Of course now my room is slightly warmer..........

Also, I had to drop the 5830 clocks back to its previous settings. Windows was acting a little funny and when I tried playing World of Tanks I started getting VPU recovery errors. :(
I may try updating CCC but I am leery of what AMD has been putting out lately.

Anyhow, thanks again for the help.

F$

[Crazybob]

Is this the reason for the differences in credit? I have 4 machines that get on average 1,600 credit per wu. I have 1 machine that gets 6,000 per wu? I checked and I'm not alone as far as this goes. If I could get them all to get the 6,000, I could increase my output by 375%. Any ideas? This would really help the team to figure it out. Run times don't seem to be it. GPU 4800, 5800, doesn't seem to matter either.

[Bryan]

CB, they have different sized wu depending on the speed of your GPU ... tiny, small, normal, and huge. Cards like the 58/69 series get the huge wu that pay 6k. Nvidia and slower ATI GPUs get the 1600, and I'm guessing CPUs get the "tiny" ones and I have no idea what they pay. If you look at the "Server Status" you will see what I'm referring to.

[Crazybob]

Thanks for the info Bryan. I made my observations actually from looking at some of your machines running the 5800 cards and getting a mixture of both 6,000 and 1,600 credit work units. Didn't realize that they were doing that. Oh well, I can wish can't I?:)

[DrPop]

Huh....could this be why my credit is all over the place? I've got a single 5870 on it, and some days I get really good credit, and other days I'd be better off crunching Collatz! :mad:

[Crazybob]

Must be. I have 2 machines that are getting about 80,000/day. My son's machine with the 5870 has an RAC of 225,000. Not that I mind, but would like to have the rest of them putting up that much.:(

[Bryan]

Jed there has to be something else wrong. The 58XX will always be issued the HUGE wu that pay 6k. Even if it was issuing smaller wu your credits wouldn't drop that much because you would complete them much faster.

What you want to check is the GPU loading on your card. If you reserve a thread for the GPU the loading should stay between 95-98%. If you are running 100% CPU crunching and Moo then you will take a major hit on the GPU output. A 5870 running at 950 MHz will produce somewhere around 380k per day on Moo. At 850 MHz it does something like 350k ... I can't give you the exact figures anymore because I'm also running the cudas on Moo.

What I do is use the program "ProLasso" to reserve a thread for Moo (affinity 7 on an I7) and set the priority to HIGH. Then for the Nvidia I set it to run on threads 0-6 also at HIGH priority. On the CPU tasks I drop the "use CPU %" to 90 which then only schedule 7 CPU tasks. Using ProLasso I set the priority to "Above Normal" and the affinity to 0-6. Doing this clears a thread that is only used by Moo.

On the machines with 5970s I have to reserve 2 threads (1 full core) to Moo. Running the 5970 or 2 5870s Moo will use 25% of the processor by itself. If you don't reserve this then you will take a major hit on GPU output.

[zombie67]

Also, be sure test out the various "Cores" to see which is best for your GPU. Don't let it auto-select which core it thinks is best. That failed for most of my machines. Once you find out which one is best, you set it using your project preferences page. Setting it on your machine at the command line works for only that one task, and then it will get reset to your project pref setting (default is -1). If you have different GPUs in different machines, then set up multiple "locations" in your project preferences, and then associate each machine accordingly.

[DrPop]

Wow, there is a lot more to setting up this project than I realized. On the 5870, I was getting over 300K / day, and then all of a sudden it dropped to like 50K. The last two days it was just over 100K. So, either Primaboinca is taking all the CPU resources (I don’t have any cores reserved), or maybe the core setting got off somehow? The weird thing is that I was getting the good credits for a long time, and then all of a sudden it just went way down. I can't think of anything that was changed. CPU is PhenomII x6 1100t and there is also a GTX 460 in there crunching Primegrid at the same time the 5870 is crunching Moo! - maybe something got into conflict?

[zombie67]

Wow, there is a lot more to setting up this project than I realized. On the 5870, I was getting over 300K / day, and then all of a sudden it dropped to like 50K. The last two days it was just over 100K. So, either Primaboinca is taking all the CPU resources (I don’t have any cores reserved), or maybe the core setting got off somehow? The weird thing is that I was getting the good credits for a long time, and then all of a sudden it just went way down. I can't think of anything that was changed. CPU is PhenomII x6 1100t and there is also a GTX 460 in there crunching Primegrid at the same time the 5870 is crunching Moo! - maybe something got into conflict?

Oh yeah. A couple more things:

1) With faster cards (58xx+), you may want to reserve threads. With my dual 5870 (remember it runs a single task across all GPUs), I had to reserve a full two threads on my 1st gen i7). No difference with my Core 2 machines and a 4870, if I reserved a thread (in this case that is a whole core) or not.

2) Like I said, if you don't define which "core" you use, the app decides for you. It it not always constant. It seems to change randomly. That could account for your changes in credits.

And to be extra clear, when Moo/DNETC talks about "core", they are not talking about CPU cores. They are talking about which sub-app to use for crunching a particular task.

[Fire$torm]

And to be extra clear, when Moo/DNETC talks about "core", they are not talking about CPU cores. They are talking about which sub-app to use for crunching a particular task.

Ah Ha!! You are the first person That I recall making that statement. So thanks! That clears things right up.


Jed there has to be something else wrong. The 58XX will always be issued the HUGE wu that pay 6k. Even if it was issuing smaller wu your credits wouldn't drop that much because you would complete them much faster.

What you want to check is the GPU loading on your card. If you reserve a thread for the GPU the loading should stay between 95-98%. If you are running 100% CPU crunching and Moo then you will take a major hit on the GPU output. A 5870 running at 950 MHz will produce somewhere around 380k per day on Moo. At 850 MHz it does something like 350k ... I can't give you the exact figures anymore because I'm also running the cudas on Moo.

What I do is use the program "ProLasso" to reserve a thread for Moo (affinity 7 on an I7) and set the priority to HIGH. Then for the Nvidia I set it to run on threads 0-6 also at HIGH priority. On the CPU tasks I drop the "use CPU %" to 90 which then only schedule 7 CPU tasks. Using ProLasso I set the priority to "Above Normal" and the affinity to 0-6. Doing this clears a thread that is only used by Moo.

On the machines with 5970s I have to reserve 2 threads (1 full core) to Moo. Running the 5970 or 2 5870s Moo will use 25% of the processor by itself. If you don't reserve this then you will take a major hit on GPU output.

Thanks for the suggestions. I know you and Maxwell have discussed ProLasso before but I've been hesitant to use it on my desktop. Already have a ton of crap, err processes, running in the background. Maybe in the near future.....

[spingadus]

So, I just put a 6970 in my i7 920 rig. I'm running both Moo and dnetc as well as wcg on all 8 threads. Can someone tell me if my stats look ok?

http://moowrap.net/results.php?hostid=1939&offset=0&show_names=0&state=3&appid=

http://dnetc.net/results.php?hostid=20642


Also, if I free up 1 thread, by setting boinc manager to use only 87.5% of the processors and find the best "core" to use by running the moo app with the -bench flag, how do I know if Moo will use the correct(free) thread? Does boinc know that the 1 free thread is to be used by moo and that it should be the core that I selected based on the -bench test? Am I making sense?

[Fire$torm]

So, I just put a 6970 in my i7 920 rig. I'm running both Moo and dnetc as well as wcg on all 8 threads. Can someone tell me if my stats look ok?

http://moowrap.net/results.php?hostid=1939&offset=0&show_names=0&state=3&appid=

http://dnetc.net/results.php?hostid=20642


Also, if I free up 1 thread, by setting boinc manager to use only 87.5% of the processors and find the best "core" to use by running the moo app with the -bench flag, how do I know if Moo will use the correct(free) thread? Does boinc know that the 1 free thread is to be used by moo and that it should be the core that I selected based on the -bench test? Am I making sense?

Your run times seem good but as I do not have a 6970 I cannot say for sure. Also you didn't mention if you are OC'ing the card. BTW, if you haven't already you should install MSI's Afterburner (AB) utility. Even if you do not OC, AB is the best way to control fan speed and monitor GPU temp and load. You can configure AB to display temp, Core speed and load readings in the Windows System Tray.

Note: In my experience with AB, version 2.0.0 works best for OC when unlocked but has a minor bug that does not correctly display GPU Core speed in system tray. This bug is corrected in version 2.1.0 but the procedure for unlocking AB does not work properly for me.

For controlling GPU load Bryan & Maxwell are fond of a utility called Process Lasso (Link (http://www.bitsum.com/prolasso.php)). take a look at the posts linked below.

http://www.setiusa.us/showthread.php?1513-Dual-ATI-on-Moo-What-are-your-loads&p=16610&viewfull=1#post16610

http://www.setiusa.us/showthread.php?1589-Looks-like-Sic-is-backing-off...&p=17602&viewfull=1#post17602

http://www.setiusa.us/showthread.php?1304-Sluggish-Response-Times&p=13025&viewfull=1#post13025


Edit:
Afterburner 2.0.0 ---> http://www.mediafire.com/?qhptap1gxcarqpk
Afterburner 2.1.0 ---> http://event.msi.com/vga/afterburner/download.htm

[spingadus]

Wonderful! Thanks for the info Fire$torm :)

You also answered a question I had about tools to use for AMD cards, since I've always used nvidia in the last 5 years. I had thought AB was for nvidia, good to hear I can use it.

[Fire$torm]

You are welcome.

One thing I did not expound on, unlocking Afterburner. Unlocking refers to a setting that allows AB to OC a GPU past ATI/AMD/nVidia reference specs. Most say the OC limit imposed is due to concerns over warranty of the product.
Anyhow, if you are interested in unlocking Afterburner then....

For Afterburner 2.0.0 see the Overclock.net How-To (Link (http://www.overclock.net/ati-drivers-overclocking-software/677696-how-unlock-msi-afterburner.html)).

For Afterburner 2.1.0 see the Guru3D How-To (Link (http://forums.guru3d.com/showthread.php?t=337897)) and OCXtreme (Link (http://wp.ocxtreme.org/?p=26)).
If you have File Permission Issues with AB files refer to this link ---> http://forums.guru3d.com/showpost.php?p=3890692&postcount=4

[Slicker]

You are welcome.

One thing I did not expound on, unlocking Afterburner. Unlocking refers to a setting that allows AB to OC a GPU past ATI/AMD/nVidia reference specs. Most say the OC limit imposed is due to concerns over warranty of the product.
Anyhow, if you are interested in unlocking Afterburner then....

For Afterburner 2.0.0 see the Overclock.net How-To (Link (http://www.overclock.net/ati-drivers-overclocking-software/677696-how-unlock-msi-afterburner.html)).

For Afterburner 2.1.0 see the Guru3D How-To (Link (http://forums.guru3d.com/showthread.php?t=337897)) and OCXtreme (Link (http://wp.ocxtreme.org/?p=26)).
If you have File Permission Issues with AB files refer to this link ---> http://forums.guru3d.com/showpost.php?p=3890692&postcount=4

Does Afterburner work with non-MSI cards?

[spingadus]

I forgot to mention that I'm not going to OC anything at the moment due to heat and electrical bill $$. Just normal crunching.

[Cruncher Pete]

Does Afterburner work with non-MSI cards?

Yes...

[Fire$torm]

Does Afterburner work with non-MSI cards?


Yes...

Affirmative on both non MSI - ATI/AMD & nVidia cards.