• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

SMP is a POS

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

mxthunder

Member
Joined
Dec 27, 2004
Location
Northeast Ohio
Getting this error when trying to install the SMP client on my main rig, just to learn how to do it for my folding rig:
2010%5C01%5C23%5Cbikepics-1883970-full.jpg



Is there any advantage to running the SMP client over running 4 of the standard CPU clients installed as a service, as far as PPD goes?
Thanks


btw, did anyone else see the depressing event that happend for team 32 this morning?? :-/:-/:-/
 
The SMP client is substantially better than 4 single clients.

Did you run the install.bat file in the directory before launching the program? Basically, that error means that it can't connect to the loopback interface on the computer due to lack of credentials.
 
Hmm interesting thats good to know I guess. I ran the install.bat, and apparently it doesnt like blank credentials so I made a windows password and it 'worked', but I dont want a windows password so I guess I wont be able to use this client.
The client itself didnt work anyways, now I am getting this error:
2010%5C01%5C23%5Cbikepics-1884285-full.jpg


and for the brief period of time it was working, my CPU usage was only at 25% anyways. I have the -smp flag set in the windows shortcut.
 
MX, you do need a password for this to work.

However, you can set Windows to auto-login for that account, so you don't have to type it in when you boot up.

However, if you want to avoid it entirely, VMWare Player and Ubuntu should work for you.
 
However, you can set Windows to auto-login for that account, so you don't have to type it in when you boot up.
I would highly suggest NOT doing this. I've put a computer in an endless cycle where I ended up formatting it. For some reason, it needed an admin login to fix something. Problem was, it was trying to login automatically, which isn't allowed to do that operation. So we had an endless loop of it logging in, saying "you can't do this from an account that automatically logs in", logging the user out and repeating. There was nothing you could do to get around it.

Just use a password, it isn't that hard.
 
MX, you do need a password for this to work.

However, you can set Windows to auto-login for that account, so you don't have to type it in when you boot up.

However, if you want to avoid it entirely, VMWare Player and Ubuntu should work for you.


+1. If you have a compelling reason to continue using Windows, then running an SMP client in Linux with VMplayer will fill-the-bill.

Or, as thideras suggested- using a logon password is not that much of an inconvenience, but can be an asset for long term system administration. ;)
 
Awesome. Thanks for the input. It ended up working when I restarted the client, and its using 100% CPU, getting about 3300ppd. I will have to see how this affects my ppd in my folding rig, its a slower CPU speed, and I dont want to sacrifice any PPD on the GPU clients I have running.
 
I am having more issues configuring this client properly.
Downloaded newest version from stanford, followed THEIR install guide to a TEE. I get 25% CPU usage repeatedly. I have tried this process several times in several different directories, same result every time. I really hate CPU folding, so complex! :p


edit*** nvm, got it, they didnt say you needed the -SMP ANNNNDD the number of cores following it when you use the -smp flag.
 
Last edited:
It shouldn't... only if you're going to limit it to less cores than you have available... and I don't think you can force it lower than 4. What version 6.29 or 6.30?
 
6.30 it the -smp # works just as one would think. I saw that my GPU was only at 80% utilization after the CPU was folding at 100% so I reconfigured the client with 3 cores, now the GPU is folding at 99% again, and the CPU is holding at 75-80%
 
You may want to see if there are updated drivers for your GPU(s). There have been reports of drastically reduced CPU use by the GPU with updated drivers on certain cards. It could have a pretty big PPD increase for you. :)
 
-smp without a number works 100% of the time. What doesn't work is to run the client for the first time with the -config flag or by clicking on the executable, if you put -smp in the extra parameters of the configuration. Flags put there take effect after a restart and the client does not restart if run with -config or by clicking the executable. You should always configure the SMP client for the first time with the -configonly flag. If you don't, you will continue to run uniprocessor WUs until you restart the client.

Max ppd is achieved running the smp client on 4 cores of a quad along with the gpu. Assuming the sig rig, you should run the gpu client at low priority (slightly higher in systray) so that the gpu is unhindered by the cpu client run at idle priority. Also remove the affinity lock on the gpu and enter the environment variable NV_FAH_CPU_AFFINITY and set the value to 15. This will allow windows to balance the gpu client load across all processors and keep it from hindering one thread of the SMP client and having the other three wait to allow the fourth to catch up.
 
Max ppd is achieved running the smp client on 4 cores of a quad along with the gpu.
:thup: Dropping a core really gives a hit to the SMP output. I have a couple of PhenomX3 processors, and they make less than 1/2 vs. my AthlonII. (Granted, they are lower clocked, but the differences aren't enough to make up for the PPD loss)

Some examples: (Phenomx3 @ 2.4GHz, AthlonIIx4 @ 2.9GHz)

p6060
  • AthlonIIx4 - 4,730 PPD
  • Phenomx3 - 1,891 PPD

p6063
  • AthlonIIx4 - 4,482 PPD
  • Phenomx3 - 1,998 PPD

p6701
  • AthlonIIx4 - 3,525 PPD
  • Phenomx3 - 1,406 PPD
 
Thanks for the info guys.
ChasR, that was going to be my next question, 4 cores with a lower priority, or 3 with slightly higher. I will work on setting this up when I get home. Ill do another -config and go through it again.
 
Since the bonus system is non-linear, I'd expect those observations to be the case.

Here are some readings (rounded for clarity) from my systems, all p6701:

Rig #1: 12:00tpf / 6000ppd

Rig #2: 08:30tpf / 10000ppd

Rig #2 is 29% faster than Rig #1 ((12 - 8.5) / 12). Yet increasing the PPD produced by Rig #1 by 29% (6000 * 1.29) yields just 7740ppd. As you can see the actual increase in PPD from Rig#1 to Rig #2 is 66% (6000 * 1.66).

The faster the system the bigger the bonus... it's quite ingenious. :)
 
@mx
On a dedicated machine you will find there is no difference in production between -smp 3 idle priority and -smp 3 low priority. On a daily use machine there would be so little difference you couldn't differentiate between WU variability and production increase. SMP with 4 cores at idle priority will yield about 22% faster frame times while running the gpu client, as described in my earlier post, than - smp 3.
 
I would highly suggest NOT doing this. I've put a computer in an endless cycle where I ended up formatting it. For some reason, it needed an admin login to fix something. Problem was, it was trying to login automatically, which isn't allowed to do that operation. So we had an endless loop of it logging in, saying "you can't do this from an account that automatically logs in", logging the user out and repeating. There was nothing you could do to get around it.

Just use a password, it isn't that hard.

Should be able to hold the shift key down to stop windows from autologging, fwiw
 
Should be able to hold the shift key down to stop windows from autologging, fwiw
Nothing worked when that happened. Luckily, we had a backup and ended up just wiping the drive with a restore.

That post is really old BTW. :shock:
 
Back