New member introduction

[Kryten451]

Hello,
I am rejoining the SETI project after a brief try 10 years ago. I am re-joining because I want to do something worthwhile and always found SETI the most interesting project. I also visited the PR radio antenna . So now I am building a new system that will run 24 hrs for the most part, though I am primarily targeting overnight processing for SETI, until I can figure out how to work and run SETI during the day. I have most of the components to build a new moderate-high performance level Intel based system, using either 1366 or 1156 socket. Bill of materials aquired so far:
2- cases available for either mATX or ATX (PCD Qv2e or Scythe Fenris Wolf)
1- Seasonic 850 supply
1- Noctuna NH-D14 air cooler (14 inch fan model)
2- BFG 9800 GT (overclocked), CUDA capable
2- WD Green drives for raid0
1- OS, Win 7 64 Home premium

All I need at this point is a motherboard, CPU and memory. Any suggestions on a particular setup that would work particularly well for SETI overnight processing. Am pretty open minded to either Intel socket/architecture. I will overclock to high 3.X, for 4 with whatever I get (i5 or i7).

Any tips are appreciated..

 

[Duner]

Welcome back.

You'll find BOINC quite different than the classic SETI you ran 10 year ago.
My only suggestion is to upgrade the video card for improved SETI performance. The GTX260 216sp is a nice bang for your buck CUDA card right now.

 

[Kryten451]

Thanks for suggestion re: upgrading to GTX260 216sp.. These boards look in very short supply. Maybe if I wait until Fermi is generally available, if the price isn't too dear.

 

[Careface]

back mate

As Duner pointer out, a heap has changed in the last 10 years with regards to SETI. If you're interested in the science behind it, there's some good links for AstroPulse, but I can't seem to find decent links for the current MultiBeam (the "normal" SETI@home application). The BOINC platform has fundamentally changed the way that SETI (and other distributed computing programs) grants credit for work done.. SETI classic did the whole "user has done X work units", now we have a credit system loosely based on FLOP count (which dramatically varies for different work units) which in theory more accurately represents how much "science" is being done... personally, it took a while to get used to it, but I now prefer it makes it easier to see performance increases/decreases which, given these forums, is somewhat important.. anyway, I'm ranting!

Hmm GTX216s are in short supply? Weird indeed. Sadly, I'm sure Fermi will be incredibly expensive when they first come out, but I guess only time will tell.

If you're looking for a guide to installing SETI so that it'll be fully optimised (can't really remember if optimised clients were around 10years ago - I just ran the screensaver in 1999/2000 personally) and will run so you don't notice it, check out the Install sticky.

If you have any problems, just let us know Looking forward to crunching along side you .

Also, I <3 Red Dwarf lol Kryten is awesome

EDIT: Also, 99.9999% of users don't notice SETI running when their pc is in use, but if you do/are uncomfortable with running it other than overnight, let us know so we can show you how to get it running when you want it... I really should add that to the sticky at some stage, too

 

[Kryten451]

Thanks Careface, I will check out those links. Regarding 216p supply; the only New GTX 260 216sp I can find is a Palit brand, which I am unfamiliar with. http://www.amazon.com/Palit-Sonic-216SP-448bit-NE3X262SFHD94/dp/B002VJJO6Y
Not a bad price at all but thats the extent of new offerings. I'll keep my eyes peeled. I am hoping I can run CUDA/Seti on a PAIR of cards.

Because this is looking to take some time getting a new system together, I will probably start with my current system running a single 9800GT OC just to get my feet wet with Seti@home CUDA implementation. Thanks

Hey, another Dwarfer !

 

[Duner]

I just installed BOINC/CUDA-GPU-ONLY (unchecked the CPU selections) on Vista 64, and it downloaded a new SETI job, but CUDA processing (progress) appears quite slow in comparison to the CPU jobs running earlier when I first installed BOINC

now it stopped for no reason with status=Waiting to run (0.04 CPUs + 1.00 NVIDIA GPUs)

Again, in comparison, CPU tasks were progressing very quickly.. maybe my 9800GT OC is too feeble in comparison to the 3.2 Quad... although it is running Very Quietly now on the GPU only, vs. on the CPU only

I took that from the other thread to keep this one your own. I think your problem is different than SameOldFatMan's.

Short answer, do the following and your problem should be resolved.

1) go to your messages tab in BOINC. At the very start you should see your 9800GT and est. Gflops (estimated Gigaflops). If you don't see it, stop Boinc and restart it, and the message should be near the top.
2) stop BOINC
3) find your app_info.xml file and open it in notepad
4) find the lines
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
5) Immediately below that add the line
<flops>xxx</flops>
where xxx = your est. Gigaflops * 0.5 * 1000000000 (I'm assuming you added the cuda 2.3dlls from the lunatics website, if you're not sure what I'm talking about, let me know)
e.g. 70 Gflops would be 35000000000 flops.
*edit* I just reinstalled BOINC and it seems to grossly miscalculate est GFLOPS now, just put 35000000000 for your 9800GT and it should be close.
6) your app_info.xml file should loook something like.


<app_info>
<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft.dll</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3-1-1a_upx.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
<flops>35000000000</flops>
<plan_class>cuda</plan_class>
<avg_ncpus>0.040000</avg_ncpus>
<max_ncpus>0.040000</max_ncpus>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cufft.dll</file_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-1-1a_upx.dll</file_name>
</file_ref>
</app_version>
</app_info>


7) save the file and restart BOINC

MEDIUM ANSWER
Boinc still has problems estimating when a task will finish so it stops a CUDA task erroneously and starts another. When you have too many CUDA tasks "Waiting to run", your GPU will crash. BOINC then uses the CPU to crunch a GPU unit, which is very inefficient and that's why you're seeing the GPU tasks take so long. It's actually a crashed GPU and a CPU emulating that GPU task and finishing it.
Manually entering the flops value allows BOINC to properly estimate workunit end times so you don't get a tonne of "waiting to run" units.

LONG ANSWER
I don't know, it's a known error on the SETI forums and new versions of boinc were supposed to fix it. Obviously not.

Also, you're most efficient crunching combo is probably 3 CPU cores and 1 GPU core. Though if you just want to get your feet wet with CUDA, that 9800GT should put out some good numbers.

 

[Kryten451]

The troubles I reported earlier disappeared, except the GPU running "slowly".

So I went ahead and added the gigaflops value to the xml file as instructed.

The GPU still churns out the tasks about the same speed, but BOINC perhaps downloaded a whole bunch more work than it used to.. 50 in advance.. not sure if this is expected or not, i'll try removing the xml tag, observe, and re-add it to see...

Bottom line, I think my expectations on GPU performance was unrealistic, overblown for my 9800 GT... But its been sluggin away and makin points for the last week..

Nevertheless.....
I ordered a shiny new GTX 260 216sp , and i7 920 to help with the SETI tasks, and I will look at devoting 3 CPU cores @ a reasonable % of total CPU... Thanks