AOA Forums

AOA Forums (http://www.aoaforums.com/forum/)
-   Windows/Linux SMP Clients (http://www.aoaforums.com/forum/123-windows-linux-smp-clients/)
-   -   Disappointing Quad Core Performance (http://www.aoaforums.com/forum/windows-linux-smp-clients/47014-disappointing-quad-core-performance.html)

dabaerman 4th April, 2009 02:24 AM

Disappointing Quad Core Performance
 
my recent purchase of the 9600 Agena had been putting out some good folding numbers. I liked it enough to get another one. now I have 2 dual core cpu's just killing the 2 quads in folding.
was it;
A: a bad choice of quad core cpu's?
B: a bad install of FAH?
C: a need to reconfigure the client, or
D: delete and reinstall FAH?
I just updated the bios then pulled the duals out and plugged the quads in. starting the client shows 4 cores detected. and task mgr shows 4 cores (FahCore_a1.exe) @ 25% for the quads. 1 FahCore_a1.exe @ 34%, 1 @ 28%, 1 @ 22% and the last @ 16% for one of the duals'. the % for the other dual core are in the same range and the are dynamic. the quads are clocked @ 2.4 and the duals' are @ 3.0 if that makes a difference.
the 9600,s are getting a workout here before moving along and @ $85 a bargain for their new owners. I just have my doubts about them for folding.

Ron

ThunderRd 4th April, 2009 11:05 AM

Are you sure that they don't simply have slower WUs?

For example, the 2653s are the fastest by far and the most productive, and there are a lot of them around right now. If the duals have 2653s and the quads have a slower, less productive WU, that could explain it.

Outside of that, I probably would have reinstalled the client anyhow with a cpu change. The assignment server gets information from the client as to how much memory is installed, the cpu, etc. Maybe this didn't update automatically when you changed the cpu, and the client reports a dual instead. IDK, just a thought.

dabaerman 4th April, 2009 02:58 PM

I was thinking along those lines also. the duals are on 2653's and the quads are on 2665's. I'll reinstall later today. off to school.

Ron

dabaerman 5th April, 2009 02:23 AM

it's definately the 2665's. they are SLOW. the 2653 on a quad is faster than the dual by 200ppd. time to step on the clock!

Ron

dsio 5th April, 2009 05:01 AM

Two quads running atm, Q9450 @ 3.6Ghz = 8.3 hours / 1920 WU.

Q6600 @ 2.4Ghz = 25 hours / 2165 WU.

The Q6600 was running a 5101 WU.

Now the Q6600 is running a 2671 WU which is looking like 10.8 hours / 1920 WU.

The Q9450 @ 3.6 is doing some nice work atm, pushed as much as a 1920 WU in 5 hours when it had no load on it, and its actually running under stock voltage @ 50C loaded.

ThunderRd 5th April, 2009 08:13 AM

Quote:

Originally Posted by dabaerman (Post 491791)
the 2653 on a quad is faster than the dual by 200ppd.
Ron

It's gotta be more than that, mate, or something is wrong...

If you are running one client only on the quad, that is. Should be double the dual, or thereabouts. My q6600 has been running a boatload of 2653's...about 7.30 minutes per step. They are the fastest and the best-yielding WUs by far. That's in addition to the two GPU clients in the same machine.

Try this as a test. Copy your work folder and queue.dat from the quad to the dual, and run both machines on the same WU; this way you have a direct comparison.

My comparable dual-cores all run Linux, so I can't attest to a direct comparison. They get different WUs on the a2 core.

dsio 5th April, 2009 10:32 AM

Sorry to thread-jack btw, but what's with this:

Code:

[08:14:14] Completed 92500 out of 250000 steps  (37%)
[08:19:09] Completed 95000 out of 250000 steps  (38%)
[08:23:57] Completed 97500 out of 250000 steps  (39%)
[08:25:44] CoreStatus = 0 (0)
[08:25:44] Sending work to server
[08:25:44] Project: 2671 (Run 0, Clone 78, Gen 4)
[08:25:44] - Error: Could not get length of results file work/wuresults_08.dat
[08:25:44] - Error: Could not read unit 08 file. Removing from queue.
[08:25:44] - Preparing to get new work unit...
[08:25:44] + Attempting to get work packet
[08:25:44] - Connecting to assignment server
[08:25:45] - Successful: assigned to (171.67.108.24).
[08:25:45] + News From Folding@Home: Welcome to Folding@Home
[08:25:45] Loaded queue successfully.
[08:25:58] + Closed connections
[08:26:03]
[08:26:03] + Processing work unit
[08:26:03] At least 4 processors must be requested.Core required: FahCore_a2.exe
[08:26:03] Core found.
[08:26:03] Working on queue slot 09 [April 5 08:26:03 UTC]
[08:26:03] + Working ...
[08:26:03]
[08:26:03] *------------------------------*
[08:26:03] Folding@Home Gromacs SMP Core
[08:26:03] Version 2.04 (Thu Jan 29 16:43:57 PST 2009)
[08:26:03]
[08:26:03] Preparing to commence simulation
[08:26:03] - Ensuring status. Please wait.
[08:26:13] - Looking at optimizations...
[08:26:13] - Working with standard loops on this execution.
[08:26:13] - Files status OK
[08:26:13] - Expanded 4840369 -> 24012993 (decompressed 496.0 percent)
[08:26:13] Called DecompressByteArray: compressed_data_size=4840369 data_size=24012993, decompressed_data_size=24012993 diff=0
[08:26:14] - Digital signature verified
[08:26:14]
[08:26:14] Project: 2671 (Run 12, Clone 97, Gen 4)
[08:26:14]
[08:26:14] Entering M.D.
[08:31:11] Completed 2500 out of 250000 steps  (1%)
[08:36:03] Completed 5000 out of 250000 steps  (2%)


ThunderRd 5th April, 2009 11:02 AM

This might help you, it's a Linux-only error afaik. I've seen it before on occasion:

Error 0x0 and 0x1 - FaHWiki

dabaerman 6th April, 2009 04:50 AM

I know there may be subtle difference in projects and a direct comparison would be more accurate. I will say that the dual-2665= 37:37/frame or 759.21ppd. a 2665 on one quad= 26:42/frame or 8o4.12ppd. it is currently increasing in speed. the #2 quad=26:05/frame or 874+ ppd. there must be something wrong, somewhere.
let me assure you, on some things that require more than knowing what a cpu looks like, I miss out on the details. so; a single core cpu @ 100% has it's single core running @ 100%. a dual @ 100% has 50% on each core, and I wounder why not 100% on each core. which leads me to the quad core. my guess is that 2 or 4 cores = 1 cpu @ 100% making each core worth 50% or 25%. the blank stare on my face must be priceless.


Ron

ThunderRd 6th April, 2009 07:15 AM

@Ron:
You almost certainly have problems on those quad installations if that is all they can do on 2665.

I am currently running a 2665 on my Q6600[oc to 3.4G], it's at 60%. 2324 ppd, 12 minutes 15 secs per step. Even if you aren't overclocked, it shouldn't be that slow.

Windows reports 25% per core, and that is normal. Do this: check task manager. Do you show 4 fahcore_a1.exe processes? Are they at 25% under load? I run the DEINO version, and have found it to be more stable than the MPI version. There are 4 core processes, 5 deino processes, and 1 for mpiexec.

Are you running GPU clients on those machines as well? Try running the SMP client alone, and see if the result is the same.

Personally, I would rip the whole install out completely, and reinstall again, including a repair or reinstall of .net 2.0 under the admin credentials. Then see what happens.

Follow the install guide here carefully:
Folding@home - Guide

BTW, what OS is that? Everything I have said here is based on XP. Vista has different quirks; the install guides will show you.

dabaerman 7th April, 2009 03:04 AM

TR, os is XP, Deino x 5, mpiexe x 1 and 4 cores (_a1). I'll shut down the gpu for a bit on both quads. L8R.


Ron

dabaerman 7th April, 2009 02:50 PM

no big change, <100 ppd, with gpu off. I will try a complete reinstall over the weekend.

Ron

ThunderRd 7th April, 2009 04:34 PM

Quote:

Originally Posted by dabaerman (Post 491855)
no big change, <100 ppd, with gpu off. I will try a complete reinstall over the weekend.

Ron

I'm afraid that's what you have to do.

Make sure you go to the website and download a new copy of the latest Deino file. Also don't forget to reinstall or repair .net 2.0.

Use add/remove to take out the old client and the Deino service. Delete the old installation files and scan the registry for the following:
Pande
folding@home
mpiexec
deino
Remove the keys manually if there are any of those left over. Then reinstall the client and report back. I'm pretty sure that it will solve the problem, as long as the comp is picking up the quad cpu properly.

Let me know when you are running again. I won't gain much on you, I've been hit or miss the last week or two with general maintenance on my boxes, so you've put a dent in my lead anyhow ;)

dabaerman 8th April, 2009 02:53 PM

Quote:

Originally Posted by ThunderRd (Post 491857)
you've put a dent in my lead anyhow ;)

call that a dent?;) yea, I've been lucky lately. time to go to work.:(

Ron

dabaerman 9th April, 2009 05:38 AM

seem to have another issue on one box. seems like windows security certer is fubar. all I get is the header and it wont open from control pannel. however, I have removed all the old FAH/SMP/Deino stuff, reinstalled and restarted. I'll check on it in the AM. off to get some sleep.


Ron

dabaerman 9th April, 2009 02:55 PM

OK, it's morning and the fresh install was a success. now putting out 1800+ ppd. before; 850's. so it is at least double. will fix the other box tonight.


Ron

dabaerman 12th October, 2010 04:33 AM

time for an update! one the X4 9600 Agenas' is off to a new home. I have replaced the dualies with quads. 2 X4 955-3.2MHz, 2 X4 945-3.0MHz, one X4-940-3.0MHz and a lone X4 9750@2.4MHz. 17.8MHz folding v6.0 SMP for the TEAM!

ok, I like too!

Ron

ThunderRd 12th October, 2010 11:59 AM

Nice one. Looking forward to having one of the top 100 folders on our team... ;)

dabaerman 5th November, 2010 04:27 AM

forgot to mention the cpu's are working with 3 tri-sli and 3 sli mb's. 3XDDR2 and 3XDDR3. all those gpu's had to go somewhere!:drink:

Ron

*no longer disappointed with SMP and these quad's are running very well*


All times are GMT +1. The time now is 04:56 AM.


Copyright ©2001 - 2010, AOA Forums


Search Engine Friendly URLs by vBSEO 3.3.0