Trying to understand differences in machines

Message boards : Number crunching : Trying to understand differences in machines
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Smariga
Avatar

Send message
Joined: 13 Jun 99
Posts: 49
Credit: 30,454,070
RAC: 0
United States
Message 986379 - Posted: 3 Apr 2010, 18:41:44 UTC

I have 2 machines I made up specific for BOINC work (SETI). One is a Q9300 (ID: 4429891) with an NVIDIA 260-216 and its RAC is 15,150. The other is an I7-860 (ID: 5158235) with an NVIDIA 275 and its RAC is 15,155. The first runs XP-Pro 32 and the 2nd runs Win 7-64.
As they both sit all day running only BOINC, I would expect the 2nd machine to significantly outperform the 1st, but that is not happening. These machines have been in these configurations and states for a while, so the RAC should be stable.
Also, when I look at valid tasks performed by the machines, the 1st (slower) takes 60-70 secs for an elapsed time of 600-700 sec to achieve 120-150 credits. The 2nd (faster) machine takes twice as much cpu (140-160 secs) for an elapsed time of 750-800 secs to achieve the same 120-150 credits.
Is Win7 that much worse than XP Pro? The CUDA drivers are different 196.21 vs 197.13, but both are 3.0 drivers, I think.
Can anyone point out elsewhere to change to get better performance, out of either or both.
Thanks.
ID: 986379 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 986380 - Posted: 3 Apr 2010, 18:49:14 UTC - in response to Message 986379.  

Try to disable HyperThreading on i7 CPU for few hours and see will CPU and elapsed times for CUDA tasks decrease or not.
ID: 986380 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 986382 - Posted: 3 Apr 2010, 18:58:36 UTC - in response to Message 986379.  
Last modified: 3 Apr 2010, 18:58:53 UTC

I'd try using the SSSE3x version of the AK_V8 app for a start, the SSE4.1 version has been noted to be only better on fast C2D Dual Cores, where there is less bus contention,
while the SSSE3x version is better on Quads,

Claggy
ID: 986382 · Report as offensive
FiveHamlet
Avatar

Send message
Joined: 5 Oct 99
Posts: 783
Credit: 32,638,578
RAC: 0
United Kingdom
Message 986384 - Posted: 3 Apr 2010, 19:07:25 UTC - in response to Message 986379.  

I would run the 19107 driver on both rigs as well as the other tips listed.

Dave
ID: 986384 · Report as offensive
Profile 52 Aces
Avatar

Send message
Joined: 7 Jan 02
Posts: 497
Credit: 14,261,068
RAC: 67
United States
Message 986402 - Posted: 3 Apr 2010, 20:38:50 UTC
Last modified: 3 Apr 2010, 20:39:13 UTC

Per Win 7, it's more efficient on processor scheduling, and I crunch about 8% more apples to apples vs XP.

But you have to make sure you've tuned it as well as what you're comparing it to. TaskMgr is your friend, it'll tell you what is eating CPU & Memory real time (I turn off pretty much everything, including scanners). Act accordingly to your priorities. Also, turn off S@H screen saver, check your local boinc settings, turn off all forms of machine power save, throttle back, hibernate, etc. For the final tweaks, check OC of CPU & GPU, and disable Aero.

Enjoy
ID: 986402 · Report as offensive
Profile Smariga
Avatar

Send message
Joined: 13 Jun 99
Posts: 49
Credit: 30,454,070
RAC: 0
United States
Message 986552 - Posted: 4 Apr 2010, 13:58:24 UTC

Well, I did all the suggestions to the Win 7 machine. Took me a while to find the HT bios option but changed it. Reinstalled the unified Lunitics drivers. How do I readily tell that they are running? The names look the same. CUDA looks good, as BOINC tells me as its starting up.
I usually do monitor Task Manager and get rid of a few things that start unwanted.
And it looks like RAC is in flight again. We'll see in a day or two.
If it works well on this machine, I'll go back to the other and make the changes to it.
Thanks all.
Alex
ID: 986552 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 986575 - Posted: 4 Apr 2010, 15:48:49 UTC - in response to Message 986552.  
Last modified: 4 Apr 2010, 15:56:29 UTC

Well, I did all the suggestions to the Win 7 machine. Took me a while to find the HT bios option but changed it. Reinstalled the unified Lunitics drivers. How do I readily tell that they are running? The names look the same. CUDA looks good, as BOINC tells me as its starting up.
I usually do monitor Task Manager and get rid of a few things that start unwanted.
And it looks like RAC is in flight again. We'll see in a day or two.
If it works well on this machine, I'll go back to the other and make the changes to it.
Thanks all.
Alex


"The names look the same":
What are the names of the SETI apps shown in Windows Task Manager?

You can see the SETI apps .exe files in
C:\ProgramData\BOINC\projects\setiathome.berkeley.edu\

Did you use
Win64 Lunatics' Unified Installer v0.2
http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=206

Did you install (manually copy)
Cuda 2.3 DLLs
http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=208

.
 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 986575 · Report as offensive
Profile Smariga
Avatar

Send message
Joined: 13 Jun 99
Posts: 49
Credit: 30,454,070
RAC: 0
United States
Message 986596 - Posted: 4 Apr 2010, 17:08:30 UTC - in response to Message 986575.  

Names in Task manager are:
AK_v8b_win_x64_SSSE3x.exe and MB_6.08_CUDA_V12_VLARKill_FPLim2048.exe *32
So I guess I am using the correct files.

I did use the Unified Lunatics x64 installer and changed the selection options to SSSE3 and CUDA. Looks OK from here.

I am using the CUDA 2.3 dll's, dated 7/3/09. Looks like I never migrated the 3.0 dll's at all.

When I look at the complete(valid) work units, how do I know when they were finished on my machine. Which date is correct so I can see if they are processing faster.

Thanks.

Alex
ID: 986596 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 986605 - Posted: 4 Apr 2010, 17:37:32 UTC - in response to Message 986596.  

Yes, you using correct apps.
Look at date of report. It's the best estimate of time when task was processed, especially if you have large WU cache.
ID: 986605 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 986767 - Posted: 5 Apr 2010, 4:22:54 UTC - in response to Message 986596.  

...
When I look at the complete(valid) work units, how do I know when they were finished on my machine. Which date is correct so I can see if they are processing faster.

Thanks.

Alex

"Sent" indicates when the Scheduler told your host about the work and "Received" when your host reported to the Scheduler the work was done. You know from those that sometime within that interval your host downloaded and crunched the WU then uploaded the result.

"Run time" is the best for comparing speed. That's BOINC's measure of the elapsed time during which the work was actively being crunched.

"CPU time" is a useful measure of efficiency. For CPU work it should be close to "Run time", the difference is a measure of time stolen by other activity on the host. For GPU work there would ideally be near zero "CPU time" since it indicates the part of the "Run time" when the GPU was waiting to be told what to do.
                                                               Joe
ID: 986767 · Report as offensive

Message boards : Number crunching : Trying to understand differences in machines


 
©2025 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.