View Full Version : SETI@Home CUDA GPU client released
maddog39
12-17-08, 02:15 PM
Just noticed this on the front page of the SETI@Home site, looks like it was just posted.
http://setiathome.berkeley.edu/cuda.php
[Edit]
Dated December 18, 2008 on the front page news article.
EarthDog
12-17-08, 02:18 PM
Will this crunch SETI faster like it does F@H?
Looks like they finally got it working right. Too bad I only have a 7950gt. :(
muddocktor
12-17-08, 02:45 PM
Will this run with the optimized apps installed? And does this run as a supplement to the regular work units (4 regular wu's and a CUDA wu on my QX9650 system)?
maddog39
12-17-08, 02:56 PM
Will this run with the optimized apps installed? And does this run as a supplement to the regular work units (4 regular wu's and a CUDA wu on my QX9650 system)?
No unfortunately, I just tried that. Although its extremely fast. Its almost done with a 2:30 long workunit after 33 minutes. Also, I believe that its using the CPU and the GPU at the same time to work on a given workunit which is why it doesnt work.
EarthDog
12-17-08, 03:08 PM
You cant set affinty to the process like in F@H?
Voodoo Rufus
12-17-08, 04:09 PM
Nice. Once we see how it'll compare to our quads this could be a good bang/buck option for farming like FAH.
Shelnutt2
12-17-08, 05:04 PM
I'm just wondering does this include the linux clients? Does seti CUDA support work natively in linux?
toomnymods
12-17-08, 05:47 PM
hmmm can u run the optimized client with this?? will icrunch 4 units at a time or 5?
figures it was too good to be true.. the thing wont work with optimized ak stuff.. guess untill the optimizers come out i wont use the gpu
maddog39
12-17-08, 07:10 PM
I'm just wondering does this include the linux clients? Does seti CUDA support work natively in linux?
Haha dude, look at my sig, of course. Go download the .sh for BOINC 6.4.5, run it and then cd BOINC and ./run_manager. You need a recent nvidia driver and make sure that your BOINC system service is shut down (if you've installed from the repositories) before running the new one. As the new one will just connect to whatever service is already running (ie. the old one).
Rickdstik
12-18-08, 09:21 AM
That would be awesome ... except all my machines have either 79xx cards or 4870s in them :/
tricknasty
12-18-08, 10:11 AM
The performance of a GeForce GTX 280 GPU running SETI@ is nearly 2 times faster than the fastest consumer multicore CPU (3.2GHz Intel Core i7 965) and almost 10 times faster than an average dual core consumer CPU (2.66 GHz AMD Phenom 9950)ii. hrrm im gonna have to give this a try on my gtx280
http://www.nvidia.com/object/io_1229516081227.html
Evilsizer
12-18-08, 12:32 PM
o man!! o man!! what timing! think of a bunch of these running FAH on the 9400 it has.
http://www.ocforums.com/showthread.php?t=588758
i just wonder how many 9400's are needed to match one GTX280 in crunching power. im thinking for the same power used, a bunch of those may beat the GTX280 in work done per watt.
Haha dude, look at my sig, of course. Go download the .sh for BOINC 6.4.5, run it and then cd BOINC and ./run_manager. You need a recent nvidia driver and make sure that your BOINC system service is shut down (if you've installed from the repositories) before running the new one. As the new one will just connect to whatever service is already running (ie. the old one).
it's not working for me. BOINC successfully detects CUDA device but S@H not use it. on WinXP it's working ok. maybe becouse S@H linux client is older than windows? (6.03 vs 6.05)
maddog39
12-18-08, 02:10 PM
I'm just wondering does this include the linux clients? Does seti CUDA support work natively in linux?
it's not working for me. BOINC successfully detects CUDA device but S@H not use it. on WinXP it's working ok. maybe becouse S@H linux client is older than windows? (6.03 vs 6.05)
Im not really sure where those version numbers come from. I am using BOINC 6.4.5 with setiathome app version 5.28. Also, CUDA support is built into the nvidia drivers but you need a fairly recent version and I dont remember off hand what version exactly it is that you need. The version 177 driver that comes with Ubuntu Intrepid doesnt seem to cut it. I am working on installing 180.06 right now to see if theres a difference.
[Edit]
Irritatingly enough, after some searching apparently Linux is not supported. Which doesnt seem to make much sense as there is full support for it in the latest drivers. I have just tested and BOINC with the 180.11 nvidia driver from the jaunty repos will not use the GPU. Wow, major fail on there part...
http://setiathome.berkeley.edu/cuda_faq.php
there is an actuall list of Seti@Home applications: http://setiweb.ssl.berkeley.edu/beta/apps.php
now i'm waiting 4 linux version ;)
You think you could run it on this ? :p
http://www.nvidia.com/object/product_tesla_c870_us.html
toomnymods
12-18-08, 04:34 PM
dont feel bad it doesnt work on xp either..
dont feel bad it doesnt work on xp either..
hmmm... i'm tested new S@H on XP & latest NVidia drivers and it's working and computing really fast...
tricknasty
12-18-08, 08:40 PM
how do utilize the other cores of my system? the CUDA app is running right now but the other cores are idle.
toomnymods
12-18-08, 09:06 PM
hmmm... i'm tested new S@H on XP & latest NVidia drivers and it's working and computing really fast...
yea i was completing units like 1 every 2 mins on my gpu but they werent giving me any credit, did it for a few hours and i got zilch... until someone figures out how to make it work im not going to waste any more time on it
You think you could run it on this ? :p
http://www.nvidia.com/object/product_tesla_c870_us.html
Definitely, but if your thinking about buying one you should check out the c1060 and s1070. If anybody would like to donate I'll volunteer to build a test system :)
eaglescouter
12-30-08, 09:39 AM
Found on the SETI Project Forums:
It is possible to run 5 wu at a time on a quaddie with a gpu (ie 4 cpu + 1 gpu), but you need to have a cc_config.xml with a <ncpus>5</ncpus> statement in it. By default it will run as 3+1. There is a "feeder" app that passes the data to the gpu and BOINC treats it as a task too, even though it only requires a small percentage of cpu time, which is why most of us use a cc_config.xml. If you run this way your RAC should climb.
The ability to correctly run 4+1 without using cc_config is coming in a future BOINC version. Also the ability to run the cpu and gpu versions of the same app (for MB) is coming in a future version. Last news on that was "in a month or two".
vBulletin® v3.8.7, Copyright ©2000-2012, vBulletin Solutions, Inc.