folding on HP DL580 G5 quad 2.93 core2 quads Linux

N

nwrtarget

Gawd
Joined
Aug 10, 2010
Messages
903
So I happen to have a Linux test box that I can fold on most of the time. As the subject says it is a four socket box with four quad cores in it. It also has 40 gigs of ram. It is running Cent 5.5. It is currently about done with its ten a3 units and is running forceasm and smp. Can it run the high bonus wu? I have been digging for a few days now and keep finding conflicting data on bigadv so if that is an option on Linux please point me in the right direction. It has a passkey assigned.

I have a couple of other machines folding too but this guy should be able to generate some serious ppd.

If the only way to get the big points is by running windows I will eventually be able to run a windows vm on the esx that will eventually live on this guy but if there is some simple tweak that we can do now it will help out the [H]orde.
 
Quad Socket-604 7300-series Tigerton Xeon MP?? Wow, this is a very rare configuration here at the [H]orde!! Run a server version of Windows and then the -bigadv flag for current maximum PPD. Let us know if you have any questions and best of luck! :cool:
 
-bigadv is windows only currently :(

A VM would work, just not sure if it could handle all of the threads (you would know better than me).

In any case, that thing has got to be cranking some good ppd. What are the time per frames looking like?

Welcome to the [H]orde BTW! Fold on!

 
Wine and WinSMP has a chance, but I couldn't get it to work. That is your only shot short of installing Windows. Hopefully, you have better luck than I did.
 
Kendrak said:
In any case, that thing has got to be cranking some good ppd
Click to expand...
It has 16 logical cores at nearly 3.0GHz, so it should be quite productive. I figure similar to a SR-2 with 16 threads depending on the WU. I could be off in my estimations by quite a bit though...

My dual Harpertown @2.9Ghz with 8 threads dedicated -bigadv machine does about 20k PPD, quite a bit less with the P2684. However, it's using the less efficient 5000V chipset. So, maybe this is capable of 40k or a bit better if Windows Server uses memory more efficiently. I think it's possible the OS could make a difference here since the Core2 architecture seems to be very sensitive with memory sub-system performance and -bigadv.

Welcome to the [H]orde BTW! Fold on!
Click to expand...
Yeah, forgot that in my excitement, welcome to the [H]orde! :cool:
 
Where do I find time per frames?

I do know that on the wu it has done, big and smp, it can churn in 3 hours and 15 minutes. Those seem to only be worth around 921 points. At the end of a run it usually says around 40 Gflops.

I am just trying to make sure I am getting the most pps out of this monster and from what you are saying the answer is no unless I go windoze. To do that right I would have to run two 8 cpu vms with 16 or so gigs of memory each. Do we think those vms could make the high bonus cutoff under bigadv? If so this might work really well except for the windoze part.

Good call on the cpu I hadn't bothered to look it up but the are xeon 73xx series at 2.93.
 
Look for a file called FAHlog.txt

That has all the goings on. But on the screen that you referenced, just count how long it takes to go from % to %.

1% = 1 frame
 
nwrtarget said:
Where do I find time per frames?

I do know that on the wu it has done, big and smp, it can churn in 3 hours and 15 minutes. Those seem to only be worth around 921 points. At the end of a run it usually says around 40 Gflops.

I am just trying to make sure I am getting the most pps out of this monster and from what you are saying the answer is no unless I go windoze. To do that right I would have to run two 8 cpu vms with 16 or so gigs of memory each. Do we think those vms could make the high bonus cutoff under bigadv? If so this might work really well except for the windoze part.

Good call on the cpu I hadn't bothered to look it up but the are xeon 73xx series at 2.93.
Click to expand...
unfortunately, (and APOLLO would be the team's expert here) I don't think 8 cores from the Core2 architecture will get the job done on a consistent basis on the bigadv units. Particularly since I assume there'd be some performance lost due to the dual VMs.

You'd be better off doing what you're doing.
 
nwrtarget said:
I do know that on the wu it has done, big and smp, it can churn in 3 hours and 15 minutes. Those seem to only be worth around 921 points. At the end of a run it usually says around 40 Gflops.
Click to expand...
Complete a WU in just over 3hrs? That is very fast for pre-Nehalem hardware even if it is standard A3 SMP. Very fast...

Good call on the cpu I hadn't bothered to look it up but the are xeon 73xx series at 2.93.
Click to expand...
These are your processors: http://ark.intel.com/Product.aspx?id=30796

You know what, I wonder if these can be BSEL modded to 1333MHz FSB... :eek:
 
Ah so it runs a whole wu in 200 or so minutes and with the logic it cranks a frame every 2 minutes.

The ones I have looked at are 3 hours and some number of minutes around 15. I know not all of the wu are that fast but I have only been folding on this thing for a couple of days.
 
nwrtarget said:
Ah so it runs a whole wu in 200 or so minutes and with the logic it cranks a frame every 2 minutes.

The ones I have looked at are 3 hours and some number of minutes around 15. I know not all of the wu are that fast but I have only been folding on this thing for a couple of days.
Click to expand...

Can you post a snip-it of the log file (the beginning part) looks like this.....

Code: 
[13:44:12] Project: 2682 (Run 1, Clone 1, Gen 19)
[13:44:12] 
[13:44:14] Assembly optimizations on if available.
[13:44:14] Entering M.D.
[13:44:32] Completed 0 out of 250000 steps  (0%)
[13:58:26] + Results successfully sent
[13:58:26] Thank you for your contribution to Folding@Home.
[13:58:26] + Number of Units Completed: 69

[14:06:24] Completed 2500 out of 250000 steps  (1%)
[14:28:06] Completed 5000 out of 250000 steps  (2%)
[14:49:48] Completed 7500 out of 250000 steps  (3%)
[15:11:31] Completed 10000 out of 250000 steps  (4%)
[15:33:27] Completed 12500 out of 250000 steps  (5%)
[15:55:24] Completed 15000 out of 250000 steps  (6%)

That will give us the project # and the time per frame and we can use this calculator to see what your ppd is.
 
nwrtarget said:
Oh and to answer the question the 3 hour 15 minute wu were a3.
Click to expand...
And it was in Linux as well. You will be seeing an improvement in Windows even if you stick to only running standard A3 SMP, so time to completion will be even less...
 
Yeah I can't mod this guy he occasionally has to do actual work but most of the time it is just folding.

Oh and I think the whole server as it sits it more than 30 grand. At one point not to long ago 50...

This isn't a bargain machine!

It sounds like I want to esx this thing as soon as I ca so I can windoze it up!
 
Phone double posted on me.

I wonder what a DL380 G6 can do? Dual I7 based xeons running 2.53 gigahertz.. I could probably throw a win vm on and find out? Will that complete the bigadv in time?
 
Last edited:
jebo_4jc said:
DO IT DO IT DO IT
Click to expand...

APOLLO said:
Yup, this is the [H] after all, and mentioning OCing must be performed within a couple of posts following the OP. It is a tradition here. :D
Click to expand...

I'm not sure moding a $50k boxen (that isn't his, or is it?) is such a good idea.

ivoid.jpg
 
Kendrak said:
I'm not sure moding a $50k boxen (that isn't his, or is it?) is such a good idea.
Click to expand...
No, it's definitely not. LOL, I posted before I saw the actual price figure for the whole server and was half joking. :p
 
Hell yes! If it was mine to mod I would be modding it.

I just joined the forums but have been reading [H] since 1999.

Normally I can find all the data I need by just looking but in this case I wanted to post to be sure.

I will post the tops of the runs tomorrow.
 
Great to have you here. Anything you want to know that is folding-related or otherwise, just drop us a line. Someone here is bound to know the answer or can find it for you. :)
 
Tomorrow I will post the snippets.
Under vmware esxi 4 you don't lose much performance and since each vm will have its own disk arrayof 8 148 10k sas drives... who knows. I will say this though I doubt it will matter if I run one or two vms as they probably won't be able to to the other one is there.

I did have two fedora vms on it before it went to a bare metal install and the big wu were coming at 7 hours if I recall.
 
nwrtarget said:
I wonder what a DL380 G6 can do? Dual I7 based xeons running 2.53 gigahertz.. I could probably throw a win vm on and find out? Will that complete the bigadv in time?
Click to expand...

That should definitely be capable of bigadv. It's basically the same as a non-overclocked SR-2, and my rough guess is you should get in the neighborhood of 40K PPD.
 
Doesn't musky have a system around that config? If you dig thru my old skynet thread I may have hit that OC in my evaluation but I'm not sure if I tried folding atthat low of a speed.

Kendrak's dual xeon was near there I think, see what he got, it was just over 30k I think
Posted via [H] Mobile Device
 
Dual E5530s without turbo (2.4 GHz) ran bigadv units a few K ppd faster than an I7 920 @ 4GHz (around 35K ppd.) My current dual X5550s at 2.93GHz will just clear 50K ppd. Kendrak actualyl ran dual Xeons at 2.53 before he got his SR-2. The hardware itself is fine. The problem I think you are going to have as getting over 8 threads to a VM. Can esx handle 16 threads or more? If you only have 8 available, I don't think you will be able to run bigadv.
 
It will be fun to try it out and find out what it does. I can't speak intelligently on how VMWare ESXi handles virtual CPU's. I know I can only pass 8 cores through to a VM but I wonder if that allows it to use 8 cores / 16 virtual cores or not. I have read recently that there seems to be less benefit to having virtual cores with the newer work units and if that is actually the case then it might work out fine. Testing is good!

For once in my life I wish I could Windows!!!! But it has other purposes in life that require Linux on VMWare.

Anyway I might have that running later today depending on how things go.
 
The DL580 test box was reinstalled yesterday so the logs got roasted. Oh well we will have more later.

Unfortunately until F@H releases -bigadv for native Linux installs again Linux systems like the DL580 G5 aren't really that great in the PPD department. You can't put enough through to a VM to make it work so you have to run Windows direct.

Did M$ make a secret large contribution with strings attached? rhetorical question
 
From my calculations, finishing smaller A3 units in 3 hours and 15 minutes still results in about 35k PPD. Although that is not as good as what you would see from -bigadv, it's still a not inconsiderable amount of points output and it's definitely worth folding that way.
 
That's not bad at all. Means bigadv would result in some serious points
Posted via [H] Mobile Device
 
Yeh 35k ppd is a good bit from a single box.

Hell the avg person here does 6k ppd.
 
From the DL580 G5

top - 15:32:51 up 1 day, 3:42, 4 users, load average: 17.02, 11.45, 5.55
Tasks: 238 total, 2 running, 236 sleeping, 0 stopped, 0 zombie
Cpu(s): 0.0%us, 22.5%sy, 77.5%ni, 0.0%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 41184700k total, 823540k used, 40361160k free, 62712k buffers
Swap: 8388600k total, 0k used, 8388600k free, 422980k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
5399 root 34 19 1164m 125m 2664 S 1497.0 0.3 78:26.97 FahCore_a3.exe
5247 root 39 19 31800 21m 1268 R 100.1 0.1 58:37.64 FahCore_78.exe

It wants to run that single threaded one and I don't know why so I started an normal one too.


[20:27:25] Verifying core Core_a3.fah...
[20:27:25] Signature is VALID
[20:27:25]
[20:27:25] Trying to unzip core FahCore_a3.exe
[20:27:26] Decompressed FahCore_a3.exe (5277040 bytes) successfully
[20:27:26] + Core successfully engaged
[20:27:31]
[20:27:31] + Processing work unit
[20:27:31] Core required: FahCore_a3.exe
[20:27:31] Core found.
[20:27:31] Working on queue slot 02 [August 20 20:27:31 UTC]
[20:27:31] + Working ...
[20:27:31]
[20:27:31] *------------------------------*
[20:27:31] Folding@Home Gromacs SMP Core
[20:27:31] Version 2.22 (June 10, 2010)
[20:27:31]
[20:27:31] Preparing to commence simulation
[20:27:31] - Assembly optimizations manually forced on.
[20:27:31] - Not checking prior termination.
[20:27:31] - Expanded 1763844 -> 2250969 (decompressed 127.6 percent)
[20:27:31] Called DecompressByteArray: compressed_data_size=1763844 data_size=2250969, decompressed_data_size=2250969 diff=0
[20:27:31] - Digital signature verified
[20:27:31]
[20:27:31] Project: 6070 (Run 0, Clone 15, Gen 134)
[20:27:31]
[20:27:31] Assembly optimizations on if available.
[20:27:31] Entering M.D.
Starting 16 threads
NNODES=16, MYRANK=2, HOSTNAME=thread #2
NNODES=16, MYRANK=3, HOSTNAME=thread #3
NNODES=16, MYRANK=4, HOSTNAME=thread #4
NNODES=16, MYRANK=5, HOSTNAME=thread #5
NNODES=16, MYRANK=6, HOSTNAME=thread #6
NNODES=16, MYRANK=7, HOSTNAME=thread #7
NNODES=16, MYRANK=8, HOSTNAME=thread #8
NNODES=16, MYRANK=9, HOSTNAME=thread #9
NNODES=16, MYRANK=0, HOSTNAME=thread #0
Reading file work/wudata_02.tpr, VERSION 4.0.99_development_20090605 (single precision)
NNODES=16, MYRANK=10, HOSTNAME=thread #10
NNODES=16, MYRANK=11, HOSTNAME=thread #11
NNODES=16, MYRANK=12, HOSTNAME=thread #12
NNODES=16, MYRANK=13, HOSTNAME=thread #13
NNODES=16, MYRANK=1, HOSTNAME=thread #1
NNODES=16, MYRANK=14, HOSTNAME=thread #14
NNODES=16, MYRANK=15, HOSTNAME=thread #15
Making 2D domain decomposition 4 x 4 x 1
starting mdrun 'Mutant_scan'
67500008 steps, 135000.0 ps (continuing from step 67000008, 134000.0 ps).

NOTE: Turning on dynamic load balancing

[20:27:38] Completed 0 out of 500000 steps (0%)
[20:30:15] Completed 5000 out of 500000 steps (1%)
[20:32:51] Completed 10000 out of 500000 steps (2%)




And the single threaded one

[19:33:50] Work type 78 not eligible for variable processors
[19:33:50] Core required: FahCore_78.exe
[19:33:50] Core found.
[19:33:50] Working on queue slot 01 [August 20 19:33:50 UTC]
[19:33:50] + Working ...
[19:33:51]
[19:33:51] *------------------------------*
[19:33:51] Folding@Home Gromacs Core
[19:33:51] Version 1.90 (March 8, 2006)
[19:33:51]
[19:33:51] Preparing to commence simulation
[19:33:51] - Assembly optimizations manually forced on.
[19:33:51] - Not checking prior termination.
[19:33:51] - Expanded 504099 -> 2514349 (decompressed 498.7 percent)
[19:33:51]
[19:33:51] Project: 6505 (Run 11, Clone 177, Gen 46)
[19:33:51]
[19:33:51] Assembly optimizations on if available.
[19:33:51] Entering M.D.

[19:34:11] (Starting from checkpoint)
[19:34:11] Protein: 1CFC_A_12 in water
[19:34:11]
[19:34:11] Writing local files
[19:34:11] Completed 5000 out of 250000 steps (2%)
[19:34:12] Extra SSE boost OK.
[19:38:46] Writing local files
[19:38:46] Completed 7500 out of 250000 steps (3%)
[19:43:20] Writing local files
[19:43:20] Completed 10000 out of 250000 steps (4%)
[19:47:54] Writing local files
[19:47:54] Completed 12500 out of 250000 steps (5%)
[19:52:28] Writing local files
[19:52:28] Completed 15000 out of 250000 steps (6%)
[19:57:01] Writing local files
[19:57:01] Completed 17500 out of 250000 steps (7%)
[20:01:35] Writing local files
[20:01:35] Completed 20000 out of 250000 steps (8%)
[20:06:09] Writing local files
[20:06:09] Completed 22500 out of 250000 steps (9%)
[20:10:42] Writing local files
[20:10:43] Completed 25000 out of 250000 steps (10%)
[20:15:16] Writing local files
[20:15:16] Completed 27500 out of 250000 steps (11%)
[20:19:49] Writing local files
[20:19:49] Completed 30000 out of 250000 steps (12%)
[20:24:23] Writing local files
[20:24:23] Completed 32500 out of 250000 steps (13%)
[20:29:01] Writing local files
[20:29:01] Completed 35000 out of 250000 steps (14%)
[20:33:50] Writing local files
[20:33:50] Completed 37500 out of 250000 steps (15%)

The ram for the DL380 isn't here yet otherwise I would be lighting it up right now.
 
On the 6070 WU the point calc says

Folding@home SMP2 Bonus Point Calculator
2 min 40 secs
PPD 21425.01

During that time it was running a little slower because of WU 6505 is still running on a single thread.

After stopping the single threaded one the time per frame dropped to 2:05 yielding a 31026 PPD score.

So I have been running this thing for a day or so how is my contribution to the [H]orde only 9000?
 
nwrtarget said:
So I have been running this thing for a day or so how is my contribution to the [H]orde only 9000?
Click to expand...
You need to complete 10 WUs with your passkey to be eligible for the bonus points. Until then you'll receive only the base credit for each WU. At this rate you will accomplish it in little time.
 
I'm not an expert, but based on that log info you posted, I'm confused why you would be running a 16 thread smp WU while running a 1 thread(core) WU at the same time..You are making them fight each other that way, so drop the single thread and just run the SMP client..

As you seemed to figure out, your TPF will be much better and net you a much higher PPD...Also, welcome to the [H]orde!
 
Yeah I did it to get through the single threaded one without wasting 15 cores for ever while it ran but stopped it over the weekend and maybe permanently as it was just killing productivity of the main one.

I posted 1731 in the last update! I am pretty sure that was more than one machine but it was still pretty nice.

Does anyone know if the ten WU you have to run against your key all have to be on the same "machine"? The DL580 got reinstalled so I was curious if it would have to start over on another set of 10 WU? If so do I just need to copy the directory over so that doesn't happen again?

Right now there are three DL380's in boxes, waiting, we will see how much burn in time I can get in on each of them. Two of them have 2.8's instead of the slower 2.53. Although that circuit might run out of power :(

DL 380 G6 x3
200-300 watts or so per so 600 to 900
DL580
500 watts
Storage array with crap load o drives
300

1700 Watts well that will probably not blow the breaker so long as I don't start them all at once.

I also have a DL585 G2 that is sitting around powered off. I could slap some stuff on it and get it churning away. It has four dual core AMD Opteron's and around 16 gigs of ram. Doesn't sound like it would do very well at -bigadv being AMD but just wanted to float it out there. Any thoughts is it worth the power and effort of firing it up? It wouldn't be in the above power equation as it would be in another room.

Everyone thanks for the warm welcome.
 
To get Bonus:
A3 WU with your pass key
80% finish rate

Does not matter if it is 1 box or 100, you just need 10 finished under your passkey.
 
nwrtarget said:
I also have a DL585 G2 that is sitting around powered off. I could slap some stuff on it and get it churning away. It has four dual core AMD Opteron's and around 16 gigs of ram. Doesn't sound like it would do very well at -bigadv being AMD but just wanted to float it out there. Any thoughts is it worth the power and effort of firing it up? It wouldn't be in the above power equation as it would be in another room.

Everyone thanks for the warm welcome.
Click to expand...

Depends on which opteron model but even if it can't run -bigadv it should still run quite nicely on normal smp WU, i don't think ive ever seen PPD for 4 x dual core opteron so it would be interesting to see.
 
You must log in or register to reply here.
Back
Top