8 core xeon folding help

R

RAD_MAN

n00b
Joined
Dec 5, 2005
Messages
31
I have been folding on and off for years with team 33, nothing major though just a couple home boxes. I recently came into a couple nice dual socket quad-core Xeon servers lately and at first I was just going to use them for a test environment. I quickly realized that I am going to barely be hitting these computers CPU wise and since they will be turned on 24/7 anyway I might as well put them to good folding use.
To get to my point here, I have 2 dual socket quad-core xeon boxes (2 x X5550, 6GB DDR3 @ 1333 & 2 x E5310, 8GB DDR2 @ 667), they both run Server 2008 R2 Enterprise x64 with Hyper-V enabled. I went with MS virtualization because I had the best management tools for that platform. I installed the 6.29 beta for MPICH client with the -smp flag as that looked to be the best fit to get started. After some looking around in the forum at what other people are doing and the results they are getting I figured something was wrong cause it didn't look like I was getting the results everybody else was. I think I am pulling like ~2500 PPD and form the looks of things I should be getting more. I do know I am not runing -bigadv but I don't think I can run that as the Hyper-V only lets me put 4 virtual CPU's to a VM and I am already running a Hypervisor. Am I just screwed on cranking out some serious PPD on these guys or is there something I haven't thought of yet. Thanks for any tips.
 
if you cant do a 8 core VM then A3 SMP is your friend. You should be getting a nice amount from those rigs, and I could only imagine those 5550's getting set free on a SR2 lol. Even at that, 8 real cores should put up some decent numbers at stock, im sure we/ll get you squared away, but i am kinda busy now, just wanted to say hi and welcome to the forums
 
If you post your log file that will go a long way to help us figure out what is going on.

However my guess is that the MS virtualization is killing/starving the client. (or we don't have the correct one running)

Just run the beta SMP2 client in windows. It should be set to idle priority so nothing that you are running will have to fight for CPU cycles. It will also cut down the the RAM usage.

On Nexus (rig specs in sig) running the SMP2 client on A3 WU I can get ~17k ppd just of that rig.

If you follow this and run with the -advmethods flag you should be all set.

http://hardforum.com/showpost.php?p=1032943603&postcount=2
 
--- Opening Log file [April 1 13:15:15 UTC]


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

Folding@Home Client Version 6.29

http://folding.stanford.edu

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

Launch directory: C:\FAH
Executable: C:\FAH\[email protected]
Arguments: -smp

[13:15:15] - Ask before connecting: No
[13:15:15] - User name: RAD_MAN (Team 33)
[13:15:15] - User ID: 7021B1EC3C933E74
[13:15:15] - Machine ID: 10
[13:15:15]
[13:15:15] Loaded queue successfully.
[13:15:15]
[13:15:15] + Processing work unit
[13:15:15] Core required: FahCore_a3.exe
[13:15:15] Core found.
[13:15:15] Working on queue slot 09 [April 1 13:15:15 UTC]
[13:15:15] + Working ...
[13:15:15]
[13:15:15] *------------------------------*
[13:15:15] Folding@Home Gromacs SMP Core
[13:15:15] Version 2.17 (Mar 12, 2010)
[13:15:15]
[13:15:15] Preparing to commence simulation
[13:15:15] - Looking at optimizations...
[13:15:15] - Files status OK
[13:15:16] - Expanded 1799574 -> 2396877 (decompressed 133.1 percent)
[13:15:16] Called DecompressByteArray: compressed_data_size=1799574 data_size=2396877, decompressed_data_size=2396877 diff=0
[13:15:16] - Digital signature verified
[13:15:16]
[13:15:16] Project: 6014 (Run 0, Clone 168, Gen 108)
[13:15:16]
[13:15:16] Assembly optimizations on if available.
[13:15:16] Entering M.D.
[13:15:22] Using Gromacs checkpoints
[13:15:23] Resuming from checkpoint
[13:15:23] Verified work/wudata_09.log
[13:15:24] Verified work/wudata_09.trr
[13:15:24] Verified work/wudata_09.edr
[13:15:24] Completed 192446 out of 500000 steps (38%)
[13:17:08] Completed 195000 out of 500000 steps (39%)
[13:20:34] Completed 200000 out of 500000 steps (40%)
 
That is my most recent log file on my dual X5550. That one has 8 cores and each core has HT, they are all pegged 100%.

Also, when i ran through the setup / -configonly I told it I wanted to do -advmethods work units.
 
Are you using FAHMon to look at your ppd?

If so that is the issue. FAHMon does not know how to calculate the bonus that you get when you run A3.

With you TPF your should be getting around 13k ppd. Everything looks fine. Let the TPF settle down and you will get a better reading, but HFM.net is what you want to run to look at your ppd when running A3 WU.

 
The first thing I noticed from the log file was he was just using the "-smp" flag. If you change that to "-smp 8" then you should utilize all eight cores, producing a lot more bang for your buck. I might be missing something here though... :(
 
Axdrenalin said:
The first thing I noticed from the log file was he was just using the "-smp" flag. If you change that to "-smp 8" then you should utilize all eight cores, producing a lot more bang for your buck. I might be missing something here though... :(
Click to expand...

using just -smp will saturate all the cores.
You only need to run -smp # when you want to specify less than all of them. Such as -smp 7.

Also your log looks good, and you should be getting some decent PPD with that setup.
Do you have Turbo enabled?
 
nitrobass24 said:
using just -smp will saturate all the cores.
You only need to run -smp # when you want to specify less than all of them. Such as -smp 7.
Click to expand...

Thanks for the clarification on that Nitro...I've learned something more today ;)
 
I was wondering about the -smp flag when I decided to do some play time with the Windows SMP client again.
 
I am almost wondering if I am being hammered by the Hyper-V on the CPU. Between the 2 servers I have 16 cores and 24 threads and I am just not seeing the production I expect. Has anybody run F@H on a box using Hyper-V (or Hyper-V R2) before?
 
You must log in or register to reply here.
Back
Top