• 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.

Why timeless WU on slow systems?

M

marty9876

[H]ard|DCer of the Month - February 2006
Joined
Jun 11, 2003
Messages
4,906
I've just seen this a lot and it's got me wondering- why do so many folks run timeless WU's on slower systems? 233-700's ect.

If the system is not on 24/7 then I totally understand.

But is the system is on 24/7 why not run normal WU's? Does a 40 point (making these point values up, bear with me) timeless WU take 1/4 the time a 160 point normal WU takes? Or would a 140 normal WU take ~1/3 the time giving you a slight bonus? I know the big WU's have major bonuses, but 80 days w/o a point from a boxen is along time.

Or is it just to get more regular output from a boxen? I've never ran a timeless WU, and some of those normals for 70-241 points take 75% of the time a 600 pointer big WU did for me. I should have switch all my boxen to big blocks way before I did, I'm basically getting the points from 20 boxen that I was on 35 boxen running 50/50 big/normals. Yea, live and learn.

I'm going to be setting up a bunch of slow boxen hence the question.

Any input?
 
marty9876 said:
I've just seen this a lot and it's got me wondering- why do so many folks run timeless WU's on slower systems? 233-700's ect.

If the system is not on 24/7 then I totally understand.

But is the system is on 24/7 why not run normal WU's? Does a 40 point (making these point values up, bear with me) timeless WU take 1/4 the time a 160 point normal WU takes? Or would a 140 normal WU take ~1/3 the time giving you a slight bonus? I know the big WU's have major bonuses, but 80 days w/o a point from a boxen is along time.

Or is it just to get more regular output from a boxen? I've never ran a timeless WU, and some of those normals for 70-241 points take 75% of the time a 600 pointer big WU did for me. I should have switch all my boxen to big blocks way before I did, I'm basically getting the points from 20 boxen that I was on 35 boxen running 50/50 big/normals. Yea, live and learn.

I'm going to be setting up a bunch of slow boxen hence the question.

Any input?
Click to expand...

Even at 24/7, your slower systems will run out of time. At that point you have wasted days/weeks of computer time. It's happened to the one slow box I have.
 
Hito Bahadur said:
Even at 24/7, your slower systems will run out of time. At that point you have wasted days/weeks of computer time. It's happened to the one slow box I have.
Click to expand...

I thought basically Stanford would not give you anything you could not finish per their benchmarking progress on core startup.
 
Your absolutley right! If the box is completing the WU on time I would let it run on normal.
When a box finnishes a WU...and you have the verbosity-9 on it tells you how much time it had left to complete it.
I have an old celeron that runs @850 set to regular WU and it completes them with no trouble.
Use the verbosity-9 flg and it will tell the tale for you!

 
Considering my slowest system does not have SSE, it's not a good idea for me to run Gromac or Amber core proteins. Thus, the reason for the timeless work units. I would also hate to see how long it would take for me to run normal WUs on a 266 mhz PII. There are probably some WUs it would get that it probably wouldn't finish in time.

I have run timeless and regular on my PIII 700 at work and find that the timeless units do not give as many points per day as normal units do. But that can also be a difference because it has SSE and can chew through the Gromac and Amber core WUs a lot faster than a non-SSE system.

I also have a T-Bird 1.33@1.4 borged that I run timeless on for two reasons. It's not on very much and in some cases may not make deadlines and it's also on dial up and I don't get to the box but maybe once every week or two to send the units back it has completed. The ability to queue up to 10 timeless WUs also helps.

 
Marty,
I agree with SmokeRngs on the reason for running timeless on those boxen. One is very slow (8 years old) and the other isn't on much... those are the only reasons to run timeless in my opinion. Anything greater then 400 or 450 if all it's doing is folding 24/7 then it shouldn't have any problem with even the big 600 getting them in within the deadlines. I mean its got 82 days to finish those. Thats nearly 1 day per frame. The slowest box I have doing big pointers is an AthlonXP running at 1.6GHz. It only takes 50 minutes per frame on them. So even if a 400MHz boxen was truely 1/4 of its speed thats still under 4 hours per frame, taking it a little under 17 days to finish the WU. The last time I was in math class, 17 was a hell of alot less then 82. Judging from the ones I've seen the 300+ pointers take about 55-60% of the time a 600 pointer does.
Yes, its a long time inbetween WUs. The choice is just wether or not you want to see daily output from it or not. Oh and congrats on 200K

 
I have a Morgan core 1 Ghz Duron that runs the 600 pointers. It finishes them in just under a week. I would not get that kind of output from regular or timeless units as far as points are concerned. It is a Morgan core, though, so it has SSE. I would not suggest running 600 pointers on something that does not have SSE personally.

 
SmokeRngs, I've pointed this out to Marty already but, Thunderbird core Athlons and all Durons all have Enhanced 3DNow! instructions. They contain 18 of the same registers as SSE. If you use the -forceasm switch FAH will use those instructions. I'm not sure if it provides the same boost SSE does but it does utilize them. If you look in the logs it'll say its using 3DNow optimizations instead of SSE.

 
I know they do. I've checked it out on the old T-Bird system my mom has and also when it had an older Duron in it. The 3DNow enhancements do not come close to what SSE provides, though. On an older client I think I was able to get my XP1800 to run either SSE or 3DNow and there was a sizeable difference in the amount of time it took to process a frame of the same protein on a Gromacs core. If the CPU doesn't have SSE, running 3DNow is a good bit faster than nothing, but does not stack up to what you can do with an SSE core.

A couple of weekends ago, I think there was a power outage at my job and my work system restarted. Unfortunately I forgot to edit the registry to add the -forceasm switch to the client since it's installed as a service. I noticed when I got back to work that it was taking way too long to process the WU it was chewing on. I looked at the log and noticed it wasn't running SSE. I quickly made the necsary changes and it was speeding along again (well, as much as a PIII 700 can).

It's just scary how much the SSE boost is faster than nothing or 3DNow. I have no proof for this, but I think one reason SSE is faster is because of the support for it. There just has been no support for 3DNow for the most part. I believe in some cases 3DNow was actually better than SSE. I do not know of where to find the proof for that so don't take it as fact.

I apoligize for any spelling and/or grammatical errors. I didn't get any sleep last night and I should have already passed out by now. I'm now going to head home and get what sleep I can.

 
SmokeRngs said:
A couple of weekends ago, I think there was a power outage at my job and my work system restarted. Unfortunately I forgot to edit the registry to add the -forceasm switch to the client since it's installed as a service. I noticed when I got back to work that it was taking way too long to process the WU it was chewing on. I looked at the log and noticed it wasn't running SSE. I quickly made the necsary changes and it was speeding along again (well, as much as a PIII 700 can).
Click to expand...

If I may ask, what is the registry edit to add the -forceasm switch to the client installed as a service? I just setup a new Athlon XP 2500+ for my parents and installed the client as a service, but I could not find where to add the -forceasm switch. Any help?
 
This where the edit is and the flags I use.
C:\Documents and Settings\Mayhem\My Documents\Folding@home Files\FAH502-Console.exe -svcstart -forceasm -advmethods -verbosity 9
Run regedit go to HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\FAH@C:+Documents and Settings+Mayhem+My Documents+Folding@home Files+FAH502-Console.exe

Its the image file that you modify and put your switches into, there is a space between each switch.
Hope this helps!
Good luck.

 
Schmave said:
If I may ask, what is the registry edit to add the -forceasm switch to the client installed as a service? I just setup a new Athlon XP 2500+ for my parents and installed the client as a service, but I could not find where to add the -forceasm switch. Any help?
Click to expand...

What I normally do is just do a search in the registry for "-svcstart" . This should bring you to the value you need to add to. Just open the entry up that has the path to the folding executeable. Depending on the directory you have folding installed to it should look something like this C:\Folding@Home\FAH502-console.exe -svcstart . Once the entry is open, all you have to do is add -forceasm to the end of it and the next time the service is started it will start with the -forceasm switch. It's actually rather simple to do.

If you do a search through some of the threads here, you may find a better explanation on how to find the correct registry entry. My description is not the best but these steps work well for me.

 
Some numbers.

If running without any boost = 1 Point per hour then adding 3DNow+ boost will up that to around 1.5 and adding SSE boost will up it to around 2 points per hour.
Stanford benchmarks on a 2.8 Ghz Intel system and records the number of day it took.
The multi this figure by ten and add two days for the first deadline.
So in theory a 280 Mhz system running 24/7 with SSE boost could return every work unit with 2 days to spare and get full points.
With only 3DNow boost that ups to around 450 Mhz and with no boost at all it will be closer to 550-600 Mhz.

Luck.......:D
 
Wait...so let me get this right...SSE boost is not on by default?

And if not...all I need to do is add -forceasm to the command line of the graphical client to enable it, correct?
 
Its on by default the first time you start F@H up.
But if F@H shuts down incorrectly (power cut, system crash, etc) then the next time it starts up it will be off.
So you need to shut F@H down and restart it to enable SSE boost again.

To get around that problem just add the -forceasm switch so it works everytime.

Luck.......:D
 
Ok, I think I just answered my own question...sort of.

I've got a p2 400 256/ram boxen running normals (p2 no SSE right)

Gromac core 155 pointer (p874_p53dimer874) (all per EM3)
100 frames
5:11:15/frame
.299 PPH
7.176 PPD
50.232 PPW

Wholly molly slow as hell. Boxen if basicallt folding 98% of the time. XP box.

Man I can't wait to get 40 of these boxen, I'd be putting up 280 points per day. Rock on....not.

Now I'm scared to check the p2 450....

But I'll make the deadlines by a good 10 days....
 
I would highly suggest running timeless units on that. My PII 266's usually get around 50-68 points per week on the timeless. If you ran the timeless on those boxen, you would up your points production from them a good bit since what you are now getting I'm gettin on a 266.

 
You must log in or register to reply here.
Back
Top