• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

F@H SMP clients

P

Pocatello

DC Moderator and [H]ard DCOTM x7
Staff member
2FA
Joined
Jun 15, 2005
Messages
6,869
I have been running GPU clients only for the past year.

I don't have any i3/i5/i7 CPUs. But I want to get my Quad Cores (Q6600s) running SMP again.

What client should I use? What is this A3 I hear about?

Should I upgrade my GPU clients, or are they up to date?

I went to Stanford's site... and I was not impressed.

Thanks in advance.
 
Install the recent SMP client from Stanford's site here: http://www.stanford.edu/group/pandegroup/folding/release/FAH6.29beta-win32-SMP-mpich.exe

When you configure the client make sure you select yes for advanced WUs, or add the -advmethods flag to receive A3 WUs. Also make sure you have the -smp x flag, where x equals the number of cores you want processing the WUs. On my four core systems, I employ the -smp 3 flag because it works better than using all cores for some odd reason. Core 2 architecture is very different from Nehalem, and performance seem to vary a lot, with the config not being straightforward from my experience.

The GPU client has also been updated and it's a drop in executable to update it.
 
And on a side note, I knew what I was looking for and it still took a bit of digging to find these, plus I didn't really know the difference between them (reference to the other thread about how to improve output for the Horde.)
 
Is this correct?
Code: 
[<snip>
[06:35:14] + 2334720 bytes downloaded
[06:35:14] + 2340489 bytes downloaded
[06:35:14] Verifying core Core_a3.fah...
[06:35:14] Signature is VALID
[06:35:14]
[06:35:14] Trying to unzip core FahCore_a3.exe
[06:35:16] Decompressed FahCore_a3.exe (8087552 bytes) successfully
[06:35:21] + Core successfully engaged
[06:35:26]
[06:35:26] + Processing work unit
[06:35:26] Core required: FahCore_a3.exe
[06:35:26] Core found.
[06:35:26] Working on queue slot 01 [April 15 06:35:26 UTC]
[06:35:26] + Working ...
[06:35:26]
[06:35:26] *------------------------------*
[06:35:26] Folding@Home Gromacs SMP Core
[06:35:26] Version 2.17 (Mar 12, 2010)
[06:35:26]
[06:35:26] Preparing to commence simulation
[06:35:26] - Looking at optimizations...
[06:35:26] - Created dyn
[06:35:26] - Files status OK
[06:35:27] - Expanded 1768960 -> 1971489 (decompressed 111.4 percent)
[06:35:27] Called DecompressByteArray: compressed_data_size=1768960 data_size=19
71489, decompressed_data_size=1971489 diff=0
[06:35:27] - Digital signature verified
[06:35:27]
[06:35:27] Project: 6020 (Run 0, Clone 191, Gen 111)
[06:35:27]
[06:35:27] Assembly optimizations on if available.
[06:35:27] Entering M.D.
[06:35:34] Completed 0 out of 500000 steps  (0%)
[06:56:07] Completed 5000 out of 500000 steps  (1%)
[07:15:33] Completed 10000 out of 500000 steps  (2%)
[07:35:01] Completed 15000 out of 500000 steps  (3%)
[07:56:07] Completed 20000 out of 500000 steps  (4%)
[08:16:13] Completed 25000 out of 500000 steps  (5%)
<snip>
 
Yep...A3 core...that is an SMP unit. 6020's are the bigger ones.
 
Should I be concerned?

This boxen is a Q6600 / Vista 64 / 6GB memory.
 
Considering my overclocked E2140 does about 15 minutes per frame on that same work unit, 20 minute frames on a Q6600 are a bit worrying. Are you running any other CPU-intensive programs on your PC? Check in Task Manager to see if Fahcore_a3.exe is using up around 100% of your CPU.
 
Yeah... something's definitely wrong. For comparison my stock Q6600 (Durn Dell workstation) is pulling ~9min frames with a p6015.
 
Zero82z said:
Check in Task Manager to see if Fahcore_a3.exe is using up around 100% of your CPU.
Click to expand...

This.
If you're not at 100% utilization on all 4 cores, something is wrong. Make sure that you've set the client up correctly and it says "if you see this twice, mpich is working".

CraftyOtter said:
Maybe need the -smp 4 flag set??
Click to expand...

Another strong possibility. Should only need -smp though, don't need to specify -smp 4.
 
Are you also running a GPU client? It is important if you are because you might need to change your config.
 
I am not running a GPU client.

{The GPU on this machine always EUE and cannot finish any work.}

The Q6600 is stock.
 
Pocatello said:
I am not running a GPU client.

{The GPU on this machine always EUE and cannot finish any work.}

The Q6600 is stock.
Click to expand...
I was going to suggest to try the -smp 3 flag since this has helped my frame times substantially, but I run dual GPU clients. It may or may not help your particular situation. Don't know but it's worth a shot. :confused:
 
APOLLO said:
I was going to suggest to try the -smp 3 flag since this has helped my frame times substantially, but I run dual GPU clients.
Click to expand...
Can you post more details about this? Before and after PPD values and TPFs, etc.? Also what CPU you are doing this with, what GPUs, and what OS. I am also running dual GPU clients and I'm willing to give this a shot if it'll improve performance for me as well.
 
Zero82z said:
Can you post more details about this? Before and after PPD values and TPFs, etc.? Also what CPU you are doing this with, what GPUs, and what OS. I am also running dual GPU clients and I'm willing to give this a shot if it'll improve performance for me as well.
Click to expand...
I'll post some TPFs a little later in the evening. The cards I run are all 8800GT. I have the -smp 3 flag with my dual dual-core systems (dual Opteron 280s and dual Woodcrests), and the -smp 7 flag on my dual Cloverfields with quad 8800GT cards. My performance drops when I try to have the client access all cores while running the GPUs. If I drop the GPU clients, the times are better using all cores. This is with the priority setting on idle for the SMP2 client and higher for the GPUs. The difference in time isn't huge, but when you are running TPFs around the 10 minute mark or higher, every little bit helps, LOL.
 
Zero82z said:
Can you post more details about this? Before and after PPD values and TPFs, etc.? Also what CPU you are doing this with, what GPUs, and what OS.
Click to expand...
OK, I tried it on my dual Woodcrest system running a bit under 2.3GHz (4 cores total). First log excerpt is with the -smp 3 flag:

[07:22:50] Project: 6025 (Run 1, Clone 108, Gen 81)
[07:22:50]
[07:22:50] Entering M.D.
[07:22:56] Using Gromacs checkpoints
[07:22:57] Resuming from checkpoint
[07:22:57] Verified work/wudata_02.log
[07:22:57] Verified work/wudata_02.trr
[07:22:57] Verified work/wudata_02.edr
[07:22:57] Completed 30516 out of 500000 steps (6%)
[07:31:55] Completed 35000 out of 500000 steps (7%)
[07:41:54] Completed 40000 out of 500000 steps (8%)
[07:51:53] Completed 45000 out of 500000 steps (9%)
[08:01:54] Completed 50000 out of 500000 steps (10%)

Following log is the same WU but with the -smp 4 flag:

[16:42:31] Project: 6025 (Run 1, Clone 108, Gen 81)
[16:42:31]
[16:42:31] Entering M.D.
[16:42:37] Using Gromacs checkpoints
[16:42:37] Resuming from checkpoint
[16:42:37] Verified work/wudata_02.log
[16:42:37] Verified work/wudata_02.trr
[16:42:37] Verified work/wudata_02.edr
[16:42:38] Completed 305696 out of 500000 steps (61%)
[16:52:37] Completed 310000 out of 500000 steps (62%)
[17:04:15] Completed 315000 out of 500000 steps (63%)
[17:15:53] Completed 320000 out of 500000 steps (64%)
[17:27:31] Completed 325000 out of 500000 steps (65%)

The logs show well over a minute difference per frame in favor of the -smp 3 flag (~10:00 vs ~11:40 TPF). It's not huge in this particular system, but I'll take what I can get. It's running Win XP-32 SP2 with just a single 8800GT GPU client and v178.24 nVidia drivers. I would give you PPD figures but I am currently having problems with my monitoring programs not calculating the bonus points for some weird reason. Later, I will post logs of my Opteron system that also has a total of 4 cores using both flags.

I am also running dual GPU clients and I'm willing to give this a shot if it'll improve performance for me as well.
Click to expand...
Try it for several frames on your quad. Worse thing that can happen is you lose a bit of time if it ends up being slower.
 
I did actually try running -smp 3 earlier and I saw an increase in my TPF of about a minute. I'm not sure why it's doing better with your rigs, but it certainly didn't have the same result on mine.
 
Zero82z said:
I did actually try running -smp 3 earlier and I saw an increase in my TPF of about a minute. I'm not sure why it's doing better with your rigs, but it certainly didn't have the same result on mine.
Click to expand...
I think it may have to do with the fact I'm using multiple processors. It's the thing that comes foremost in my mind.
 
I got some problems.

First of all... I got the SMP running and I turned in two 100% WUs with no problems. I did some Windows Updates this afternoon and now I cannot get the SMP working at all.

When I try to run the SMP, I get a message telling me to try to run install.bat again. So I tried that. No good. So I turned off UAC and rebooted. Now when I run install.bat I get some errors. This is weird since I already had it up and running!

Here is my error message:

Code: 
'smpd' is not recognized as an internal or external command,
operable program or batch file.
account (domain\user) [CM-QUAD\dlmystique]:
password:
confirm password:
Password encrypted into the Registry.
Unable to connect to 'CM-QUAD:8676',
sock error: generic socket failure, error stack:
MPIDU_Sock_post_connect(1228): unable to connect to CM-QUAD on port 8676, exhaus
ted all endpoints (errno -1)
MPIDU_Sock_post_connect(1275): unable to connect to CM-QUAD on port 8676, No con
nection could be made because the target machine actively refused it. (errno 100
61)
Press any key to continue . . .
 
And when I try to run the SMP program itself, I get a similar error:

Code: 
'smpd' is not recognized as an internal or external command,
operable program or batch file.
account (domain\user) [CM-QUAD\dlmystique]:
password:
confirm password:
Password encrypted into the Registry.
Unable to connect to 'CM-QUAD:8676',
sock error: generic socket failure, error stack:
MPIDU_Sock_post_connect(1228): unable to connect to CM-QUAD on port 8676, exhaus
ted all endpoints (errno -1)
MPIDU_Sock_post_connect(1275): unable to connect to CM-QUAD on port 8676, No con
nection could be made because the target machine actively refused it. (errno 100
61)
Press any key to continue . . .

Current work unit:

Code: 
Current Work Unit
-----------------
Name: Gromacs
Tag: P6012R0C226G140
Download time: April 16 17:49:55
Due time: April 22 17:49:55
Progress: 87%  [||||||||__]
 
Try running install.bat from your install folder again. This one worries me. It sounds like MS released a patch that breaks the SMP client.

Edit: Never mind. I think that is what you just tried.
 
I got it working again.

I renamed the "SMP" folder to something else, and ran the install program again and then the install.bat program again.

Then I copied over all of the unique files, but not the ones that were already in my new "SMP" folder. I guess one of the files got corrupted.

Weird.

Code: 
Note: Please read the license agreement (Folding@home-Win32-x86.exe -license). F
urther
use of this software requires that you have read and accepted this agreement.

4 cores detected
If you see this twice, MPI is working
If you see this twice, MPI is working


--- Opening Log file [April 19 02:08:17 UTC]


# Windows SMP Console Edition #################################################
###############################################################################

                       Folding@Home Client Version 6.29

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: C:\Folding\smp
Executable: C:\Folding\smp\Folding@home-Win32-x86.exe
Arguments: -smp

[02:08:17] - Ask before connecting: No
[02:08:17] - User name: pocatello (Team 33)
[02:08:17] - User ID: 55xxxxxxxxxxxxxxxxxxxxx
[02:08:17] - Machine ID: 1
[02:08:17]
[02:08:17] Loaded queue successfully.
[02:08:17]
[02:08:17] + Processing work unit
[02:08:17] Core required: FahCore_a3.exe
[02:08:17] Core found.
[02:08:17] Working on queue slot 02 [April 19 02:08:17 UTC]
[02:08:17] + Working ...
[02:08:17]
[02:08:17] *------------------------------*
[02:08:17] Folding@Home Gromacs SMP Core
[02:08:17] Version 2.17 (Mar 12, 2010)
[02:08:17]
[02:08:17] Preparing to commence simulation
[02:08:17] - Ensuring status. Please wait.
[02:08:26] - Looking at optimizations...
[02:08:26] - Working with standard loops on this execution.
[02:08:27] - Previous termination of core was improper.
[02:08:27] - Files status OK
[02:08:27] - Expanded 1796989 -> 2078149 (decompressed 115.6 percent)
[02:08:27] Called DecompressByteArray: compressed_data_size=1796989 data_size=20
78149, decompressed_data_size=2078149 diff=0
[02:08:27] - Digital signature verified
[02:08:27]
[02:08:27] Project: 6012 (Run 0, Clone 226, Gen 140)
[02:08:27]
[02:08:27] Entering M.D.
[02:08:33] Using Gromacs checkpoints
[02:08:34] Resuming from checkpoint
[02:08:34] Verified work/wudata_02.log
[02:08:34] Verified work/wudata_02.trr
[02:08:34] Verified work/wudata_02.edr
[02:08:35] Completed 435912 out of 500000 steps  (87%)
 
You must log in or register to reply here.
Back
Top