• 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.

Low PPD Despite Many Clients

0

014

Limp Gawd
Joined
Nov 6, 2010
Messages
193
These are the names of my systems that have clients running along with their number of processor cores. I don't understand why I only get 30K PPD if a ~$1,300 4p system can blow me out of the water at 250K PPD. I mean, that's a lot more points. I feel like I have a lot of horsepower running here.

STPGLE 4x (quad)
STPXPDEV2 2x (dual)
STPUTIL 4x (quad)
STPUTIL2 4x (quad)
STPBACKUP 4x (quad)
STPSAND1 8x (dual quad)
TACUTIL 8x (dual quad)
TACXPDEV 2x (dual)
TACCTX01 8x (dual quad)
 
It is not just the number of cores you have running.. clock speed, architecture , and OS also come into play as well as what type of units you are crunching. For someone to give pointers, we need something more that what you name your systems. Nobody can tell if you have 4 hamsters running or a 55/56xx with issues. Also, a 4p system with high end Xeons will get a lot of points with bigadv units and if running plain SMP. See this http://hardforum.com/showthread.php?t=1662920 for a quick ref.
An i7-2600k at 4.7GHz on 8 threads will do 30-35k on SMP units in Linux

Someone with more knowledge will be able to help once you give more info.
 
I'm only asking this because you said nothing about it in your original post but - do you have a passkey? If not that could explain why your ppd is so low.

Probably need a little more info on your hardware as well. A dual quad socket 771 system is considerably slower than a dual quad socket 1366 system. Doing the math based on your stats on hardfolding it looks like you're averaging just under 2k per WU...that does seem really low to me to, like it could be a passkey issue. Let us know.
 
Last edited:
I'll break out all the specs after I get off the bus. I am using my passkey on all the systems. They are all SMP. Three do bigadv. I provided the screen shot for the guide to make bigadv run in service mode.

I'll reply with more info tonight.
 
Once the specs are in we can help you out! Hurry up bus!
 
How many run Linux, and how many run windows?
 
Two are XP, two are 2003, one is 7, and the rest 2008 R2. They are bored systems.
 
speeecccs plz

but so far sounds like a passkey issue ...
 
as for your main question about how 4p systems are making 250k points is because of the way the point system works. the faster you can run and turn in a WU the more points you make. and its exponentially more on top of that when you get into bigadv WU's. with SMP its not a huge difference. for example my phenom II 940 @ 3.4Ghz did around 7 or 8k PPD if i remember correctly and even the 2600k's only do something like 20k PPD on the same WU's. but then when you get into bigadv that number changes to 50k + PPD(or at least it use to be, i have no clue what it is now)
 
Half of these are virtual machines on VMWare: TACXPDEV, TACUTIL, STPXPDEV2, STPUTIL2, and STPUTIL.

STPGLE
stpgle1.png

TACXPDEV
tacxpdev1.png

TACUTIL
tacutil1.png

TACCTX01
tacctx011.png

STPXPDEV2
stpxpdev21.png

STPUTIL2
stputil21.png

STPUTIL
stputil1.png

STPSAND1
stpsand11.png

STPBACKUP
stpbackup1.png
 
Last edited:
Check to make sure that the SMP flag is set, and that all the cores are being loaded.
 
given what historical rigs your running... I would say your doing good ppd wise...

I also would recommend you make sure that what your borging are not ....production machines...
 
also on some of the 1c 1t i5 boxen your probably not meeting deadlines for smp...
 
How do you have your VMs setup? I notice a lot of them say 1 or 2 processors/cores in the cpuz. Are you restricting these VMs on speed or cpu count?

On VMs, I've found that if you are restricting based on speed, set your available CPUs = to the amount of cores needed to hit that number. ie, if you have 10Ghz of available cpu over 4 cores (2.5 per core), and wanted to restrict it to 7.5Ghz of available speed, make it a 3 core instead of a 4, it will run better when it utilizes the full power of each core, instead of a partial use of all 4 cores.
 
Untitledone said:
Check to make sure that the SMP flag is set, and that all the cores are being loaded.
Click to expand...
I've done this before and did it again this morning. The a3 core does run. (That's the SMP one, right?) Both the smp and bigadv flags are working even though I have the clients running as a service. Only TACCTX01, TACUTIL, and STPSAND1 run bigadv. All do SMP.

Patriot said:
given what historical rigs your running... I would say your doing good ppd wise...
Click to expand...
This is probably all I need to hear. The saddening fact may just be that my systems aren't that fast for folding. They are more than adequate for everything else.

W.Feather said:
also on some of the 1c 1t i5 boxen your probably not meeting deadlines for smp...
Click to expand...
I am meeting all deadlines with lots of time to spare. The only time I didn't meet a deadline was when I tried running the GPU client on my workstation at work.

Syran said:
How do you have your VMs setup? I notice a lot of them say 1 or 2 processors/cores in the cpuz. Are you restricting these VMs on speed or cpu count?
Click to expand...
I don't limit CPU speed, I only limit cores. But take note of the CPUID screen shots I posted. Some of them have drop-downs to show info about additional processors. I didn't notice that myself until after I posted them here. Refer to my first post to see the core count of each machine.
 
I just thought of why my points might be lower. The systems that are doing bigadv haven't done ten work units yet. They have done 8-9, so they are close. I know you have to do ten work units before you can get extra points for doing bigadv. That must be affecting my points. I wonder how much.

Maybe I will switch them to regular SMP today so they can crunch a unit or two really quickly to get to ten. On the other hand, I may just let it run the big ones and wait it out. Either way, I'll update this thread in a while (maybe a day or two) with any changes I see in my PPD. I'm hopeful this is why my points are lower. :)
 
The 10 WU thing is per passkey, not per machine. Once you have run 10 SMP units on your passkey, any new machine will get the bonus assuming it uses the same passkey.
 
None of your machines are bigadv qualified machines.

Your best there is a quad core.

Why is it if you can't fold bigadv, people continously proclaim their computers not fast enough to fold?

You're doing pretty good for your level of hardware
 
Patonb said:
None of your machines are bigadv qualified machines.
Click to expand...
I read the requirement was 8 threads and 6GB of RAM. I have three machines that fit that. They all have 6GB of RAM. Two have 2x dual cores with HT (TACCTX01 and STPSAND1), and the other has 2x quad cores (TACUTIL).
 
he is saying that there is no one machine listed in your CPUz screen shots that meet the requirements for big adv .... the highest we see in those screen shots is 4c 4t machines......
 
014 said:
I read the requirement was 8 threads and 6GB of RAM. I have three machines that fit that. They all have 6GB of RAM. Two have 2x dual cores with HT (TACCTX01 and STPSAND1), and the other has 2x quad cores (TACUTIL).
Click to expand...

Those are the vague guidelines... Core i7 era for arch...
 
W.Feather said:
he is saying that there is no one machine listed in your CPUz screen shots that meet the requirements for big adv .... the highest we see in those screen shots is 4c 4t machines......
Click to expand...
Some of them are dual processor, you have to compare the CPU Z screenshot with how many CPUs he stated in the particular machine in the OP.

I wouldn't think the Pentium D based machines could make the deadlines on a bigadv since they are roughly multi-core Pentium 4s from my understanding.

You could plug in what units are getting assigned, and the time per frame to get a breakdown of where your PPD is coming from using the points calculator. That would help narrow down which machines are working as expected, and which are not.
 
W.Feather said:
he is saying that there is no one machine listed in your CPUz screen shots that meet the requirements for big adv .... the highest we see in those screen shots is 4c 4t machines......
Click to expand...

Yha, but relooking, TACUTIL seems it has a 2nd processer, well the cpu drop down box looks as if theres a second cpu. So that one would then.

I'd for sure relook at the pts calc and see what your best ppd combo is, as if you finish just past perferred time its only base of 3000k total. where you easily can get 3k per day on smp.
 
I think this is just coming down to accepting that they are going as fast as they can.

TACCTX01
tacctx012.png


TACUTIL
tacutil2.png


STPSAND1
STPSAND12.png
 
I can get ~6.8k to 7k PPD with a Q6600 @ 3.1GHz in Windows, shouldn't TACUTIL be able to beat that? Also is TACUTIL a dual quad or single quad (8 or 16 threads) ? I guess it could depend if it is idle or not.
 
Depending on other loads on the system and thread restrictions..... yeh, that is probably correct.
 
TACUTIL is dual quad, but it only has 8 threads. It's a virtual machine that does nothing but fold. It's on a VMWare host that has no other virtual guests. The host has dual quads. I created the VM to match.

Hmm. These are the processor specs. VMWare says the Dell PowerEdge R710 host has 2 processors sockets, 4 cores per socket, and 16 logical processors. I wonder if it would make a positive difference for me to recreate this VM with two 8-core virtual processors. I might as well try it, even though the host CPU is pegged near 100%.

I checked the calculator again because it looks like this machine moved onto a new project, 7504. The PPD prediction is now 10K. I'm still going to shut the system down and give it MOAR CORES for kicks. It may not be beneficial, but I'm going to try it anyway. Finding out will let me know for production servers too.
 
Last edited:
Thats a dual xeon 1366 which has 8 real cores and 8 more hyper threaded cores for 16 total "core" setup.

Setup right that should do 30+k alone!

You should set the host to 16 cores. Or can you in vmware? Virtualbox does but unsure on VMWare.
 
Running Windows, the R710 should be capable of about 30k ppd on average with SMP and almost 40k with bigadv(when they are available). I have an R410 running, with those same processors.
 
Yes, you can in VMWare. I shut down the guest VM and gave it 2 virtual sockets and 8 cores each, so 16. Task Manager shows 16 threads now at 98-99%. I'm just waiting for the FAHlog to update with some even numbers so I can throw them into the calculator and see if more threads helps folding even though it's the same speed.

Well, it doesn't look that great. Here are a few lines from when it ran on 8 threads (project 7504):
[21:32:55] Completed 260000 out of 500000 steps (52%)
[21:38:33] Completed 265000 out of 500000 steps (53%)
[21:44:04] Completed 270000 out of 500000 steps (54%)
[21:49:37] Completed 275000 out of 500000 steps (55%)
[21:55:26] Completed 280000 out of 500000 steps (56%)

Here are two lines from the same project on 16 threads:
[22:06:32] Completed 285000 out of 500000 steps (57%)
[22:23:43] Completed 290000 out of 500000 steps (58%)

I'm scratching my head. Hrm.

Aha. The guest VM is only using half of the host's CPU now whereas before it was using all of it. Strange. But I guess I will switch it back to 8 threads.
tacutil4.png
 
Last edited:
The log shows it used 16 threads. On the OS level, the threads are at 100%, so it has to be a VMWare thing that it only used half of the host's CPU.

Launch directory: C:\Folding@home SMP
Service: C:\Folding@home SMP\Folding@home-Win32-x86.exe
Arguments: -smp -bigadv -svcstart -d C:\Folding@home SMP

Launched as a service.
Entered C:\Folding@home SMP to do work.

[22:01:43] - Ask before connecting: No
[22:01:43] - User name: 014 (Team 33)
[22:01:43] - User ID: ***
[22:01:43] - Machine ID: 1
[22:01:43]
[22:01:43] Loaded queue successfully.
[22:01:43]
[22:01:43] + Processing work unit
[22:01:43] Core required: FahCore_a3.exe
[22:01:43] Core found.
[22:01:43] Working on queue slot 05 [January 4 22:01:43 UTC]
[22:01:43] + Working ...
[22:01:44]
[22:01:44] *------------------------------*
[22:01:44] Folding@Home Gromacs SMP Core
[22:01:44] Version 2.27 (Dec. 15, 2010)
[22:01:44]
[22:01:44] Preparing to commence simulation
[22:01:44] - Looking at optimizations...
[22:01:44] - Files status OK
[22:01:45] - Expanded 1761657 -> 2700832 (decompressed 153.3 percent)
[22:01:45] Called DecompressByteArray: compressed_data_size=1761657 data_size=2700832, decompressed_data_size=2700832 diff=0
[22:01:45] - Digital signature verified
[22:01:45]
[22:01:45] Project: 7504 (Run 0, Clone 14, Gen 166)
[22:01:45]
[22:01:45] Assembly optimizations on if available.
[22:01:45] Entering M.D.
[22:02:30] Using Gromacs checkpoints
[22:02:30] Mapping NT from 16 to 16
[22:02:33] Resuming from checkpoint
[22:02:33] Verified work/wudata_05.log
[22:02:34] Verified work/wudata_05.trr
[22:02:34] Verified work/wudata_05.xtc
[22:02:34] Verified work/wudata_05.edr
[22:02:34] Completed 283910 out of 500000 steps (56%)
[22:06:32] Completed 285000 out of 500000 steps (57%)
[22:23:43] Completed 290000 out of 500000 steps (58%)

After switching the VM back to dual quad cores, it is running faster again.
[23:53:02] Completed 360000 out of 500000 steps (72%)
[23:58:33] Completed 365000 out of 500000 steps (73%)
[00:04:05] Completed 370000 out of 500000 steps (74%)

EDIT: Well, I messed around with many different core combinations and HT settings, and the one that uses 100% of the host's CPU is dual quad cores (the way I had it in the first place). I think I have to settle that this server, or at least this job on this server, is going to give me points at the rate of 10K per day.
 
Last edited:
Can you try running natively in Windows to rule out VMWare as the issue?
 
FWIW I've been using a dedicated Dell 720 (while I'm able) with dual 8-core + HT processors. Was getting impatient with the bigadv units, so just let the FAH servers send me whatever they want. (Plus the system does get shut off once in a while as it heats up our lab quite a bit! :p) Here's what it's working on currently, roughly 70 seconds per/% Think I'm seeing around 80k PPD, despite FAHMON thinking it should be over 100k.

[01:04:53] Project: 6070 (Run 1, Clone 186, Gen 335)
[01:04:53]
[01:04:53] Assembly optimizations on if available.
[01:04:53] Entering M.D.
[01:04:59] Mapping NT from 32 to 32
[01:04:59] Completed 0 out of 500000 steps (0%)
[01:06:13] Completed 5000 out of 500000 steps (1%)
[01:07:22] Completed 10000 out of 500000 steps (2%)
[01:08:32] Completed 15000 out of 500000 steps (3%)
[01:09:42] Completed 20000 out of 500000 steps (4%)
[01:10:51] Completed 25000 out of 500000 steps (5%)

Screen shot from when I first had it up and running with bigadv, but shows the CPUID info. Seems it was taking 14m/% with bigadv on project 6900... not sure how that compares with what you're seeing.

Just thought it might be of some slight use of comparing the VM setup vs. native. :)
 
AgrFan said:
Can you try running natively in Windows to rule out VMWare as the issue?
Click to expand...
VMWare is running on ESX servers, so I can't use those to install Windows natively.

TACCTX01, STPBACKUP, STPSAND1, and STPGLE are physical machines.
 
Thought I'd throw out an update to say that I deleted the TACUTIL VM and replaced it with (same virtual hardware) Ubuntu server. I don't know if it's a chance thing or not, but I noticed today it started processing bigadv units. It's my only system that has pulled them down. It's doing 28K PPD. :D This will be a huge boost to my total PPD!
 
You must log in or register to reply here.
Back
Top