app_info.xml Help
18 Apr 2011 21:11:39 UTC
Topic 195759
(moderation:
I'm running Boinc on two virtually identical AMD 6-core machines with 16 GB and 8 GB DDR3 ram under Ubuntu 10.10. Each machine has 2 MSI Twin-Frozr Fermi cards.
Both machines are now having problems running 2 E@H WU's simultaneously on each card. At first, it looked like everything was going good. I was returning good, validated WU's. Then they started getting rated as "inconclusive" and now as bad WU's.
If I delete the app-into.xml so that each Fermi only runs 1 E@H at a time, it's all good but I feel like I'm wasting the Fermi's then. FYI, MW@H and it's app_info.xml work great.
I'm not a programmer. Could someone take a look and tell me what I'm doing wrong?
I'm wondering if problem is with the line:
314572800.000000
since the Fermis each have 767MB of ram not the amount shown. But MW@H doesn't have a problem and, until yesterday, neither did E@H.
Thanks for any help that you can provide.
Regards,
Steve
einstein_S5GC1HF
Global Correlations S5 HF search #1
einsteinbinary_BRP3
Binary Radio Pulsar Search
einsteinbinary_BRP3_1.08_i686-pc-linux-gnu__BRP3cuda32fullCPU
einstein_S5GC1HF_1.07_i686-pc-linux-gnu__SSE2
einstein_S5R6_1.01_graphics_i686-pc-linux-gnu
einsteinbinary_BRP3_1.00_graphics_i686-pc-linux-gnu
einsteinbinary_BRP3_1.06_i686-pc-linux-gnu__BRP3cuda32fullCPU
libcudart32_32_16.so
libcufft32_32_16.so
EULA.txt
db.dev.lin.826dd08f
dbhs.dev.lin.826dd08f
einstein_S5GC1HF
107
i686-pc-linux-gnu
1.000000
1.000000
SSE2
einstein_S5GC1HF_1.07_i686-pc-linux-gnu__SSE2
einstein_S5R6_1.01_graphics_i686-pc-linux-gnu
graphics_app
einsteinbinary_BRP3
108
i686-pc-linux-gnu
1.000000
1.000000
6.13.0
einsteinbinary_BRP3_1.08_i686-pc-linux-gnu__BRP3cuda32fullCPU
einsteinbinary_BRP3_1.00_graphics_i686-pc-linux-gnu
graphics_app
einsteinbinary_BRP3
106
i686-pc-linux-gnu
1.000000
1.000000
BRP3cuda32fullCPU
6.13.0
einsteinbinary_BRP3_1.06_i686-pc-linux-gnu__BRP3cuda32fullCPU
einsteinbinary_BRP3_1.00_graphics_i686-pc-linux-gnu
graphics_app
libcudart32_32_16.so
libcudart.so.3
libcufft32_32_16.so
libcufft.so.3
EULA.txt
EULA.txt
db.dev.lin.826dd08f
db.dev
dbhs.dev.lin.826dd08f
dbhs.dev
CUDA
0.500000
314572800.000000
app_info.xml Help
)
In this result, I found the following line:[pre]------> Used in total: 565 MB (203 MB free / 768 MB total) -> Used by this application: 216 MB[/pre]So, it seems that 768 MB are not always enough to run two tasks simultaneously.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
Darn. Thanks for the
)
Darn. Thanks for the feedback.
Regards,
Steve
One other question. In
)
One other question.
In E@H's config.xml I see what looks like a command line to block any cuda work. Is there a line I could add that would let me force E@H into using a particular GPU in a multi-GPU system?
I hate to waste my Fermi's on E@H if it doesn't need double-precision AND can't process 2 wu's simultaneously on the same card. However, one of my boxes has some open slots and I may be able to use one of my old GPU cards that E@H would be happy with.
If that's possible, I want to make sure that it wouldn't use the Fermi's anymore.
Thanks for any help that you can provide.
Regards,
Steve
RE: In E@H's config.xml I
)
?????
I don't know of a config.xml on the client side. If you are speaking of cc_config.xml, that's not project specific. So, there's no way to enter Einstein specific options there.
The only (easy) way I see for you is when you can place the old cards in a system without any Fermi cards. Then you can place that machine in a separate venue, in which you enable the use of CUDA devices in the Einstein@Home preferences (and of course disable it in the default venue).
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
Gundolf, Thanks for the
)
Gundolf,
Thanks for the info. Yes, it was the cc_config.xml that I was referring to.
I was afraid that I might have to relegate the E@H cuda app to it's own machine. The reason why I just built this second machine and put the Fermi's in it instead of being spread out in multiple machines was to cut down on the power draw of extra boxes. If I add back another machine, it means that if I want to use the laser printer or plasma TV, I have to shut one of them off! :)
Ah, the things we do for science!
Thanks again,
Regards,
Steve
RE: RE: In E@H's
)
Thinking out loud here while on vacation and not using my regular pc, isn't there a line in one of the app_info or cc_config files that tells it to use gpu 0 and not to use gpu 1 for a certain project? Or vice versa, meaning you could turn the Fermi off for a certain project while still leaving it in the machine.
RE: ...while on
)
Better go back on vacation, before I warn Mrs Mikey that you're refusing to relax again. ;-)
No, not for a certain project. As Gundolf said, one doesn't tell BOINC through cc_config.xml what to do per project. The file is the core client configuration file, not the project configuration file.
What you're thinking about is the option to disable certain GPUs, like 1, which will ignore the second ATI device in the system (device counting starts from zero, 0), or 1 for its Nvidia counterpart.
Then you could run with two clients on one system, using the --allow_multiple_clients flag on the BOINC executable, and tell one of them to be attached to Einstein only and use GPU 1 (0) and the other to be attached to Collatz and to use GPU 2 (1).
I was just wondering, if you
)
I was just wondering, if you start up 2 instances of the Boinc Manager simultaneously, are you just seeing two views of the same WU's being processed or is each instance of Boinc Manager running it's own separate set of WU's? If it was, that could get one around the whole app_info.xml issue for running multiple WU's simultaneously on the GPUs.
I know, I know, it can't be that simple. :)
Thanks,
Steve
Well it is almost... 1st.
)
Well it is almost...
1st. BOINC is told to ignore GPU_1
2nd. BOINC is told to ignore GPU_0
Then all you need is to get both to work via app_info.
Not that I've got this far as yet as I run 2 x GTX 460 OC's at 2 WU's per GPU.!
HTH
dunx
P.S. Found it ! http://vyper.kafit.se/wp/index.php/2011/02/04/running-different-nvidia-architectures-most-optimal-at-setihome/
RE: I was just wondering,
)
No, BOINC Manager is mostly only a GUI that allows you to easily command the client, and as such, starting it up multiple times will not start multiple clients. As a matter of fact, you can start BM multiple times, if you have multiple computers on a network that you want to monitor. Then you start a BM for each networked BOINC client.
When you start 2 BOINC Managers, all you see is all the information on your one client, twice. That is, with the 6.10 client, it's no longer possible to do so with 6.12, which will automatically start up the 'network looking BOINC' once you try to open a second BOINC Manager on the same computer.