Old question; 1 CPU, 2x clients?
Thrax
🐌Austin, TX Icrontian
I have a 1.53GHz Athlon XP on my brother's machine folding. It gets about 13 minutes to the frame on p544_BBA5_ext WUs.
I just double-stacked clients on my 2.1GHz Athlon XP 1700, and it gets the same time (13 minutes) pulling 2 of the same (p544) WUs at one time. It pulls 8-9 minutes per WU with only one client.
Is this efficient? Is there a caveat I'm missing?
I just double-stacked clients on my 2.1GHz Athlon XP 1700, and it gets the same time (13 minutes) pulling 2 of the same (p544) WUs at one time. It pulls 8-9 minutes per WU with only one client.
Is this efficient? Is there a caveat I'm missing?
0
Comments
How do I switch between 3dnow and SSE, and which is better? Removing the -forceasm flag?
If they are both running -advmethods and -forceasm then I am not sure why the 2.1 is doing 2 WU's the in the same time it would take the 1.53 to do 1. Not sure on this one at all
Are you running console or GUI on your machine? What about your brother's machine? 3.25b client?
My 2.1 is completing 2 WUs at the same time it takes for the 1.53 to do one.
Console on both, firedaemon service.
//EDIT:
At the height of my folding power, I cranked out about 700 points a week. If it's higher with this experiment..Well...Hooray.
//EDIT 2:
Idle priority.
Windows can run a CONSOLE at varying priorities if it is set as a service. The graphical client is fixed priority and is a tib slower than a console set for maximum priority that leaves machine usable. Console is set to be low priority service at default. IT CAN be set to totally take over an XP box, so do not set GREATER than negative 2 in the config filefor the console client-- note here that I am using priority when saying greater than and the number you set in console config is SIGN INVERTED to Windows and thus negative 15 will give a box that loads the system applet main pane in 15 minutes on a P4 2 gig box. You want a priority that is closer to 0 than the default but I would step slowly and let the user (you?) determine what is acceptable unless you want to load the Command Console (used to be DOS Command Prompt) and take 8-10 min for that to load and then use the DOS editor with a 45 second-1 min delay per keystroke entry if the service happens to get accidently set to negative 15 in the config file.
I do not have XP on the box to check exact name and path woudl be unique to me except for name but it is in the client directory and possibly in a subfolder (mine was, I run Linux so often that I run Linux folding whatever is sent under my own name and such that I get about 1.5 K every 13 days right now becuase I get a lot of complexes and some genome followthrough projects and about 20-25% Tinkers of molecule structure in fluid type-- if you want to see the probable no-reject for a 2 GIG P4 box and know that typically a -forceasm on a P4 box call degrades throughput by 10-12% in widnows console and in Linux console and that helps, I will toss that in-- it likes -advmethids calls in both Widnows and Linux, and this is what I SEE. I am user jdii1215 on folding if you want to see frequency and project numbers for your own info over 6 weeks time,and take into account it is a 2.26 GHz P4 running folding at about 90% of total 24\7 time with ONE instance running.).
So, you ARE getting close to double efficency, but can get more out of the CONSOLE by 10-30% with a little bit of XP slowdown and can get a lot more(like console at 1.6 times current throughput) if you want to literally have a folding only box on this box.
Depends on how image intense your work is and how much you game, test priorities versus other use with a familiar game running after a reboot so the service comes up at new priority and take baby steps with numbers for priority going toward NEGATIVE for more to folding adn less to the rest of XP with the console set as service. I do not know that FireDaemon will let you set priority as finely as a manual setting will for maximum throughput tuning with use decent-- AFAIK, it will NOT do so within 10% of what you CAN get on that box.
Running 2 clients would be a lil bit more efficient than 1. There are seconds where the client writes data to the HDD and it not working on something. Then the other client takes over for a bit. I dont think that would make up for the speed increase you are seeing.
Is yours running leaner, more efficient than his? Background programs? Kazaa? Anti-Virus?
If you're running two clients on a TBred, which is already slightly slower than a Barton, you're crippling F@H even more because now two processes share the cache. Does the amount of time saved while 1 process writes to disk really make up for the time lost because of splitting the cache?
Example: If it took 10 minutes per frame running one instance, it took 22 minutes per frame running 2 instances. So I was basically losing 1 minute per frame (x2) by running 2 instances.
//edit: that was running tinkers, maybe gromacs are more efficient...
His is a 1700+ and the other pretty much has to be a TBred or Pal since the 2500+ was the first Barton and its faster then 1.53 or 1800+ speeds.