Finally Figured out SMP

[DAK1640]

:) :) :) Finally, My IQ must be less than 12...I finally figured out how to make the SMP client work...Evidently I needed to assign a password to my PC (I've never done this EVER). Well, after I assigned a password under user accounts, all went well and I actually launched my 1st SMP client...Yippee...Wow, these puppies are worth 1523 points ???Holy Geez, and I was using my GPU's? I also loaded FahMon and figured it out also, way easier than BoinView I might add. It only took me 2.3473 years to figure this out.

If anyone is having trouble and doesn't want to get embarrased on how/what, simply PM myself. The light went on finally, and if a forgetful Old man can do this, anyone can.
\

:banana2: :banana2: :banana2: :party: :party2: :party3:

[Deux]

I've got an e6600 an 3.2 and 1950 pro, what do you recommend I run on it? ATM I run the standard GUI client on one core and the GPU which also uses a core.

[sc00p]

I am very glad that you DAK got this handled...
...but on the other hand sorry I couldn't help you with this.
And again I believe that no-one couldn't have helped you with so few information you gave to us.
It would sometimes very helpful if FAHlog.txt is posted along with other description about the problem. And PMs don't help others struggling with same problem. I suggest keep these public.

I believe you got FFFFFFFF (-1) -problem. I did have that same (maybe with log file posted I could've helped you). Maybe Stanford didn't think anyone would use Windows without logging in.
My problem was:
-I knew I was logging in (automatically) with an account with administrator priviledges (<-not enough)
-I had changed the name from Administrator to a another one
->Install.bat (or actually mpiexec) don't work like it should -> it turned me and my account name&password down saying something like "Credentials for X rejected connecting to X"
Aborting...

The solve:
-Control panel->User accounts->Change the way users log on or off->(uncheck)Use the welcome screen.
This way you're obliged to use password when starting windows. Important to use Administrator as a account name. Hope you have still the Administrator account password in your memory. Without that there's a little problem ;)
-Installed SMP at there using MPIEXEC with -register -flag... again with same Administator account name and password.
-After that launched F@H.exe to see everything was OK. Then (for the convenience) changed back the way Windows starts/logs in (<-automagically to my custom account).
-SMP should be launching/continue from there where it leaved when exiting (CTRL-C) at test phase under Administrator account.

Actually DAK you shouldn't feel embarrassed... anything else. Even Stanford isn't fully aware of this FFFFFFFF(-1) -prblem... and what could cause this.


Everyone with C2Ds should run SMPs IMO. But should be aware deadlines are for dualcores @ stock a little bit tight. There cannot be any weekend LAN partys or other delays interfering.

(hope this makes any sense to anyone... I've been having quite a few drinky-winkies :p:)

[DAK1640]

Thank you sc00p...Excellent explanation...I wished I had this before.

Deux, I too have an E6600 and an X1950 Pro...I found out that they can both run simultaneously; however you must launch SMP 1st (know idea why though).

I now have 3 SMP clients and 3 GPU's running concurrently...

I have aborted WCG for the time being to see what kinda output I can get with F@H

Observation: WCG dual core runs hotter than F@H SMP dual core, by about 5-6C

Interesting.

Thanks to all for the assist.:up: :up: :up:

[angra]

on a related note, do we know the max number of threads the SMP client (linux-64) will support? I am about to get an 8-core machine into my cluster, so doing a little planning ahead :)

[jimwah]

.. I've been having quite a few drinky-winkies :p:)

:toast: it's friday mate I'm right on it too :D

@DAK good to hear you are sorted mate, the heat difference compared to WCG (or even 2x normal FAH clients) may be because the SMP client doesn't tend to use a pure 100% load, a little bit of cpu time is wasted keeping the threads sync'ed from what I have read & observed. Good for summer folding at least :up:

@angra - wow 8 cores nice :D I would have to look into the possibility of running 4 vm's on that box with 2 cpu's each, although I bet it's one of those "higher ppd / but Stanford prefer single units completed quicker", type of situations, and would take a bit of managing.

[xVeinx]

As the forum appears to be down for F@H, I thought I'd ask around here. I've been having trouble more recently with the SMP client. It works for awhile, and then it just stops. I run it via command prompt, no fudged service install or anything fancy. It just stops though without ANY error. Any ideas? My system is rock stable and is (blushes) at stock :p:. No problems running Orthos for several hours, so I don't think that it is a hardware issue.

[sc00p]

As the forum appears to be down for F@H, I thought I'd ask around here. I've been having trouble more recently with the SMP client. It works for awhile, and then it just stops. I run it via command prompt, no fudged service install or anything fancy. It just stops though without ANY error. Any ideas? My system is rock stable and is (blushes) at stock :p:. No problems running Orthos for several hours, so I don't think that it is a hardware issue.*Guessing... as always*
...could you please check (actually uncheck) that your NIC's don't have at Power Management Allow the computer to turn off this device to save power -feature on... and see if it helps at all :shrug: If not I would still keep looking solution from somethings like nic-drivers etc IDK...

SMP client uses nic as a loopback adapter for communications (sync) between cores (you probably knew that already).

Some has told that continuous (physical) network connection is needed with SMP-client or that folding with SMP-client can be interrupted just by unplugging pc from network :confused:
On my case client didn't act this way... even disabled nic. Until I remembered that I've Hamachi installed which keeps SMP-client alive.

[sc00p]

:D :toast:
:toast: it's friday mate I'm right on it too :D

the heat difference compared to WCG (or even 2x normal FAH clients) may be because the SMP client doesn't tend to use a pure 100% load, a little bit of cpu time is wasted keeping the threads sync'ed from what I have read & observed. Good for summer folding at least :up:I agree and maybe it's "worse" (or better, depending how do you look at it) when using only a dualcore cause SMP is after all quadcore client. I guess dualcore loses efficiency for playing with (2)HyperThreads and the reason why HyperThreading should be disabled folding with true 4-cores.
BTW: p2651 (=1760 points) seems to run 3-4c hotter than p2610... which on the other hand has ~100 higher PPD.

@angra: maybe you could directly send private message to Mr Kasson at official forums about that problem :rofl: :rocker: :D
I haven't read too much about Linux client :(

[SparkyJJO]

on a related note, do we know the max number of threads the SMP client (linux-64) will support? I am about to get an 8-core machine into my cluster, so doing a little planning ahead :)

I believe they designed it with 4 threads. So running 2 SMP clients on an 8 core box would work well I'd think :)

[apexracing]

I have read that the SMP client is designed for quadcore rigs, it just happens to run on dualcore ones. Each core gets its own process.

While it would be possible to run two SMP instances on a quadcore, I'm not sure how much more one gets from it.

[Sir_Loin]

on a related note, do we know the max number of threads the SMP client (linux-64) will support? I am about to get an 8-core machine into my cluster, so doing a little planning ahead :)

I believe they designed it with 4 threads. So running 2 SMP clients on an 8 core box would work well I'd think :)

I have read that the SMP client is designed for quadcore rigs, it just happens to run on dualcore ones. Each core gets its own process.

While it would be possible to run two SMP instances on a quadcore, I'm not sure how much more one gets from it.

@ angra - These guys are correct...in one instance of F@H the client can utilize up to four cores. :up:

@ SparkyJJO - Yessir, 2 smp clients would be ideal in that situation :)

@ apexracing - You can run two intances on a four core machine, but each instance of F@H will suffer performance wise. There isn't enough or any gain to warrant that approach since you're slowing the process on a 4 day deadline :eek:

[sc00p]

:welcome: Sir :D Thank you for the visit :cool:

How is it going at top 10 (=2CPU.com) :toast: :up:

[Sir_Loin]

:welcome: Sir :D Thank you for the visit :cool:

How is it going at top 10 (=2CPU.com) :toast: :up:

Np on the visit :cool: I decided it was time to stretch my legs a bit. ;)

Unfortunately, they will not retain top 10 status. I haven't folded for them for since last October due to personal reasons. However, there are still some very good people there I'm glad to call my friends. I look forward to making new ones here....when I can get some time off work!! :shrug: