FERMI Confusion

Message boards : Number crunching : FERMI Confusion
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Dave Barstow

Send message
Joined: 14 May 99
Posts: 76
Credit: 15,064,044
RAC: 0
Philippines
Message 1113001 - Posted: 4 Jun 2011, 14:01:05 UTC

I recently 'upgraded' from the GT 220 that came with my computer to a GT 430. Since then my RAC has descended from ~7200 to ~6200 per day.

Here is my BOINC logon message:
DaBar-PC-HP6190

1 6/2/2011 10:25:40 PM Starting BOINC client version 6.12.28 for windows_x86_64
2 6/2/2011 10:25:40 PM log flags: file_xfer, sched_ops, task
3 6/2/2011 10:25:40 PM Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.5
4 6/2/2011 10:25:40 PM Data directory: C:\ProgramData\BOINC
5 6/2/2011 10:25:40 PM Running under account DaBar
6 6/2/2011 10:25:40 PM Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q8400 @ 2.66GHz [Family 6 Model 23 Stepping 10]
7 6/2/2011 10:25:40 PM Processor: 2.00 MB cache
8 6/2/2011 10:25:40 PM 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 sse4_1 syscall nx lm vmx tm2 pbe
9 6/2/2011 10:25:40 PM OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
10 6/2/2011 10:25:40 PM Memory: 8.00 GB physical, 20.69 GB virtual
11 6/2/2011 10:25:40 PM Disk: 687.80 GB total, 417.21 GB free
12 6/2/2011 10:25:40 PM Local time is UTC +8 hours
13 6/2/2011 10:25:40 PM NVIDIA GPU 0: GeForce GT 430 (driver version 27533, CUDA version 4000, compute capability 2.1, 962MB, 179 GFLOPS peak)
14 SETI@home 6/2/2011 10:25:40 PM Found app_info.xml; using anonymous platform
15 SETI@home 6/2/2011 10:25:40 PM URL http://setiathome.berkeley.edu/; Computer ID 5617853; resource share 400
16 SETI@home 6/2/2011 10:25:40 PM General prefs: from SETI@home (last modified 14-Jan-2011 03:06:09)
17 SETI@home 6/2/2011 10:25:40 PM Computer location: home
18 SETI@home 6/2/2011 10:25:40 PM General prefs: no separate prefs for home; using your defaults
19 6/2/2011 10:25:40 PM Reading preferences override file
20 6/2/2011 10:25:40 PM Preferences:
21 6/2/2011 10:25:40 PM max memory usage when active: 3686.03MB
22 6/2/2011 10:25:40 PM max memory usage when idle: 7781.62MB
23 6/2/2011 10:25:40 PM max disk usage: 2.00GB
24 6/2/2011 10:25:40 PM suspend work if non-BOINC CPU load exceeds 50 %
25 6/2/2011 10:25:40 PM (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
26 6/2/2011 10:25:40 PM Not using a proxy

Any thoughts?
ID: 1113001 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1113013 - Posted: 4 Jun 2011, 14:22:57 UTC - in response to Message 1113001.  
Last modified: 4 Jun 2011, 14:24:21 UTC

You're changed from a GeForce GT 220 with 6 multiProcessors to a GT 430 with 2,

The GT 220 does a Mid AR Wu in ~2880 secs: resultid=1917887252

The GT 430 does a Mid AR Wu in ~9088 secs: resultid=1927274503

Claggy
ID: 1113013 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14653
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1113022 - Posted: 4 Jun 2011, 14:50:24 UTC - in response to Message 1113013.  

You're changed from a GeForce GT 220 with 6 multiProcessors to a GT 430 with 2,

The GT 220 does a Mid AR Wu in ~2880 secs: resultid=1917887252

The GT 430 does a Mid AR Wu in ~9088 secs: resultid=1927274503

Claggy

Having said that, the Compute Capability 2.1 Fermis have 48 CUDA cores per multiprocessor, where the old CC 1.2 range had 8 cores per MP. On core count, the new card wins by 96:48

According to the listings at Wikipedia, the new card wins on speed as well, by 268.8:192 GFLOPs Peak. (I don't think a WikiGFlop is necessarily the same unit as a BoincGFlop).

Neither comparison would explain such a big slowdown - unless it's the 275.33 driver downclock bug biting?
ID: 1113022 · Report as offensive
Profile Dave Barstow

Send message
Joined: 14 May 99
Posts: 76
Credit: 15,064,044
RAC: 0
Philippines
Message 1113032 - Posted: 4 Jun 2011, 14:58:59 UTC - in response to Message 1113013.  

Hmmm? According to the specs, as I understood them, the GT 430 FERMI card is about 3 times faster than the GT 220, with LOTS more shaders.

Oh well... I am trying to locate a GT 460 1GB card and a new PS. May have to fly to Manila to get them... :-((

Still confused with the GT 430 being a newer FERMI card... BUT SLOWER. HUH?

I have also noted that some of my graphics apps are noticeably faster and smoother with the GT 430... but not S@H/BOINC.
ID: 1113032 · Report as offensive
Profile Dave Barstow

Send message
Joined: 14 May 99
Posts: 76
Credit: 15,064,044
RAC: 0
Philippines
Message 1113037 - Posted: 4 Jun 2011, 15:03:28 UTC - in response to Message 1113022.  

I went from 125 Gflops to 179 Gflops when I changed cards. So... I suspect there is something else mucking-up the works, as you suggest.

I'll see what additional input I get, and then re-attack the issue.

Thanks for your, as always, thoughtful input!
ID: 1113037 · Report as offensive
Profile Dave Barstow

Send message
Joined: 14 May 99
Posts: 76
Credit: 15,064,044
RAC: 0
Philippines
Message 1113041 - Posted: 4 Jun 2011, 15:06:32 UTC - in response to Message 1113022.  

Oh yeah... Was getting the downclock bug, but have not seen it with the GT 430... yet.
ID: 1113041 · Report as offensive
Cruncher-American Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor

Send message
Joined: 25 Mar 02
Posts: 1513
Credit: 370,893,186
RAC: 340
United States
Message 1113052 - Posted: 4 Jun 2011, 15:18:11 UTC

Check this out for more info.

Apparently, the GT 430 IS slower in some ways:

http://www.guru3d.com/article/geforce-gt-430-review/
ID: 1113052 · Report as offensive
Profile Sutaru Tsureku
Volunteer tester

Send message
Joined: 6 Apr 07
Posts: 7105
Credit: 147,663,825
RAC: 5
Germany
Message 1113155 - Posted: 4 Jun 2011, 20:22:47 UTC - in response to Message 1113001.  
Last modified: 4 Jun 2011, 20:25:12 UTC

I guess on the GT220 you can't let run 2+ WUs/GPU simultaneously.
What's with the GT430?
If you look e.g. with GPU-Z, how high is the average GPU Load with 1 WU/GPU?


- Best regards! - Sutaru Tsureku, team seti.international founder. - Optimize your PC for higher RAC. - SETI@home needs your help. -
ID: 1113155 · Report as offensive
Profile Dave Barstow

Send message
Joined: 14 May 99
Posts: 76
Credit: 15,064,044
RAC: 0
Philippines
Message 1113162 - Posted: 4 Jun 2011, 20:43:22 UTC - in response to Message 1113155.  

On the GT 220 2 WU was ~10% slower and 3 WU ~50% slower that running one only.

On the GT 430 2 WUs gives a slight gain.

The GPU load runs around 88% with one WU, and about 98% with two.

CPU w/GPU runs 4-5% on the GT 430.
ID: 1113162 · Report as offensive

Message boards : Number crunching : FERMI Confusion


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.