CPU & GPU processing of SETI@home v7 7.00 and cuda50


log in

Advanced search

Message boards : Number crunching : CPU & GPU processing of SETI@home v7 7.00 and cuda50

Author Message
ph.gsc.sevilla.larry
Send message
Joined: 13 Mar 12
Posts: 3
Credit: 280,672
RAC: 1,031
Philippines
Message 1408278 - Posted: 27 Aug 2013, 8:21:45 UTC

I have a Pentium Dual Core E5800 and NVIDIA GeForce 8400GS, 4Gb RAM,
Windows XP Pro w/ SP3 32-bit.

I noticed that processing using a CPU is estimated to take around 4 hours;
while processing using GPU is estimated to take around 13.5 hours.

I assumed that GPU is faster the CPU. Or maybe I have the wrong assumption.

Or does SETI@home v7 7.00 for CPU handles different kind of data than
SETI@home v7 7.00 cuda50 for GPU?


8/27/2013 8:27:03 AM | | No config file found - using defaults
8/27/2013 8:27:07 AM | | Starting BOINC client version 7.0.64 for windows_intelx86
8/27/2013 8:27:07 AM | | log flags: file_xfer, sched_ops, task
8/27/2013 8:27:07 AM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
8/27/2013 8:27:07 AM | | Data directory: e:\BOINC
8/27/2013 8:27:07 AM | | Running under account boinc
8/27/2013 8:27:07 AM | | Processor: 2 GenuineIntel Pentium(R) Dual-Core CPU E5800 @ 3.20GHz [Family 6 Model 23 Stepping 10]
8/27/2013 8:27:07 AM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 nx lm vmx tm2 pbe
8/27/2013 8:27:07 AM | | OS: Microsoft Windows XP: Professional x86 Edition, Service Pack 3, (05.01.2600.00)
8/27/2013 8:27:07 AM | | Memory: 3.50 GB physical, 6.84 GB virtual
8/27/2013 8:27:07 AM | | Disk: 18.38 GB total, 9.73 GB free
8/27/2013 8:27:07 AM | | Local time is UTC +8 hours
8/27/2013 8:27:07 AM | | VirtualBox version: 4.2.8
8/27/2013 8:27:07 AM | | CUDA: NVIDIA GPU 0: GeForce 8400 GS (driver version 320.49, CUDA version 5.50, compute capability 1.1, 512MB, 490MB available, 34 GFLOPS peak)
8/27/2013 8:27:07 AM | | OpenCL: NVIDIA GPU 0: GeForce 8400 GS (driver version 320.49, device version OpenCL 1.0 CUDA, 512MB, 490MB available, 34 GFLOPS peak)

[/img]

Profile Rae of Quakesville,CHCH NZ
Avatar
Send message
Joined: 30 Nov 01
Posts: 56
Credit: 195,657
RAC: 0
New Zealand
Message 1408280 - Posted: 27 Aug 2013, 8:39:27 UTC - in response to Message 1408278.

hi. seti work units vary in estimated computer time. while some units maybe 3-4 hrs, others might be 9 - 12 hrs or more like the astropulse units, which are closer to 55 - 60 hrs on average. Yes your GPU computes faster than your cpu & work unit estimated times also vary a lot in those. If you run only 1 work unit on your cpu, your gpu will complete its tasks quicker still. Since your cpu supports SSE2, I suggest if you want toi complete your work units even faster, try using the lunatics installer v 0.41 whichj has optimized data processing programs for both your cpu & gpu.
Happy crucnching

Rae
____________
A city destroyed by an earthquake is an opportunity to Rebuild, redeign & make it a better place to be. Better, stronger, faster like the 6 Million Dollar Man

Profile William
Volunteer tester
Avatar
Send message
Joined: 14 Feb 13
Posts: 1517
Credit: 6,856,058
RAC: 26,975
Message 1408287 - Posted: 27 Aug 2013, 9:19:13 UTC

First of all there is no difference between tasks for GPU and for CPU. You could process any task you get on any device.

Now, the estimates.

What you are seeing are the initial estimates.
BOINC has no idea how fast your system might be, so it takes a guess, drawn together from information on your GPU and CPU and averages in the database.
Initial estimates are often overcautious and notoriously bad. The GPU shouldn't take more than a few hours, even on a slow GPU.
Take a look at the app details for the host. You see an entry per app you've received. Take note of the 'avarage processing rate' or APR. That figure is compiled from the tasks you crunched. Once you have 11 valid tasks, it will be used to estimate the runtime of your tasks - from then on estimates hould be quite ok.

Furthermore, the cuda 5.0 app isn't very efficient on that old hardware. You will see best performance from either the 2.3 or the 3.2 app. The server should be sending all the possible versions until it has established a speed for all of them and then homes in on the 'best' app.

All in all, it's very early days yet - wait a couple of days to see where it's heading.
____________
A person who won't read has no advantage over one who can't read. (Mark Twain)

Profile Raistmer
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 16 Jun 01
Posts: 3262
Credit: 40,685,433
RAC: 60,181
Russia
Message 1408350 - Posted: 27 Aug 2013, 14:01:49 UTC
Last modified: 27 Aug 2013, 14:02:42 UTC

And to add to William

Last: NV 8400GS is entry level CUDA device. Looks like it has DDR2 memory (http://www.nix.ru/support/compare_tables_builder.html?item%5B1%5D=833)
So, its performance quite low per se, even after proposed software optimization (CUDA50 -> CUDA 23).
____________
News about SETI opt app releases: https://twitter.com/Raistmer

Message boards : Number crunching : CPU & GPU processing of SETI@home v7 7.00 and cuda50

Copyright © 2014 University of California