SETI MB CUDA for Linux

[sunu]

riofl, it's a long standing bug. I have it for years 

[riofl]

ahh ok... ive seen it for years but it only has started irritating me recently... ok ill be gentle with it

[Pepi]

Please how to solve this problem

boinc@pepi-desktop:~/BOINC/projects/setiathome.berkeley.edu$ ldd setiathome
   linux-vdso.so.1 =>  (0x00007fffe9dff000)
   libcuda.so.1 => /usr/lib/libcuda.so.1 (0x00007ff5e168b000)
   libcudart.so.2 => not found
   libcufft.so.2 => not found
   libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x00007ff5e137e000)
   libm.so.6 => /lib/libm.so.6 (0x00007ff5e10f9000)
   libpthread.so.0 => /lib/libpthread.so.0 (0x00007ff5e0edd000)
   libc.so.6 => /lib/libc.so.6 (0x00007ff5e0b6b000)
   libz.so.1 => /usr/lib/libz.so.1 (0x00007ff5e0953000)
   libdl.so.2 => /lib/libdl.so.2 (0x00007ff5e074f000)
   /lib64/ld-linux-x86-64.so.2 (0x00007ff5e1b27000)
   libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x00007ff5e0537000)

as you can see all so are all right but cannot find libcudart.so.2 and libcufft.so.2.
I am sure that those files are in boinc project directory. where else those files must be?

Also : what to put in this conf?
3)  Edit accordingly your ld.so.conf or the corresponding ld-something file of your distro with the above location of the cuda libs.

Now my linux machine crunch but it is slow

Thanks  for any help

[sunu]

In directory /etc/ld.so.conf.d create a file named cudalibs.conf and inside it put the path to your cuda libraries, for example:

/home/pepi/BOINC/projects/setiathome.berkeley.edu

Then run ldconfig and you should be ok.

[riofl]

i realize this isnt the place for this but i thought one of you guys may know the answer and save me finding a forum somewhere to ask..  maybe i need to call a psu mfgr?

if i find i need to run a dedicated power supply just for the GPU cards, how does startup/shutdown work? what gets powered on first? mobo then gpus? then mobo off then gpus? or should i have them on a common switch to turn both on / off at once? can the 2 motherboard/psu 'sense' contacts that automate the power on/off be paralleled for both psus to hook to the same 2 contacts?

[sunu]

You can use one of these http://www.performance-pcs.com/catalog/index.php?main_page=product_info&cPath=34_186&products_id=21193

If you need it only for the gpu you can use also a "dedicated" PSU like http://thermaltakeusa.com/Product.aspx?C=1156&ID=1544 . It comes with the above cable extension built in. Review of this PSU in http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story&reid=122 .

[riofl]

cool... thanks! i wasnt sure it was possible to tie the 2 sense connectors together but obviously it is. simplifies things a lot!

will look thru these links..

[riofl]

is there a cuda 3.0 seti app available yet? i installed 3.0 toolkit but the existing app still uses 2.2

[sunu]

No.

[Pepi]

is there a cuda 3.0 seti app available yet? i installed 3.0 toolkit but the existing app still uses 2.2

YES LINUX x64 CUDA 3.0

http://calbe.dw70.de/mb/viewtopic.php?f=9&t=120

[riofl]

is there a cuda 3.0 seti app available yet? i installed 3.0 toolkit but the existing app still uses 2.2

YES LINUX x64 CUDA 3.0

http://calbe.dw70.de/mb/viewtopic.php?f=9&t=120

thank you! i notice they still discard vlars rather than move them over to cpu responsibility... i guess that is beyond the scope of a seti app though.

[Pepi]

Reshedule tool for Win work great: why dont make it also for Linux? That will be great help!

[riofl]

there is a rescheduling tool for linux. in the form of a script.. the earlier version of the windows one before it became an exe file. works well although all it does is assign all vlars  and vhars to the cpu and depending on how the conditional is set everything else to the gpus.. also there is a script which is in one of the seti forums that acts as a workunit counting reporting tool.. kindof a companion to the rescheduling script. the script is in the first few pages of the beginnings of the windows rescheduler forum. it is a pearl script. easy to customize. for me i put all vlars on the cpus and everything else including vhar on the gpus.

[riofl]

question on proper procedure....

i am still dealing vith that huge dump of 6600 workunits i received a while back. it is down now to only VLARS and i can easily see that many of them will not get completed on time (its a shame gpus cant process them properly).

is the proper procedure to "abort" many of those tasks and report them immediately so they can be reassigned  leaving myself enough close dates that i kîow can be completed properly and then just get more units to make up my cache?

changing back to boinc 6.10.17 cured that wu get insanity, btw.

[sunu]

Yes, abort them. There is really no other way.