Buck
Storage? I am Storage!
I know it's not much, but the bartender finally got around to contributing a smidge.
Folding@Home
- Thread starter The JoJo
- Start date
That makes no sense. 70 3.0GHz P4's munching QMD's would produce around 28,000 per day. I find it absurb that we suddenly need to quadruple our output just to stay even. 10 additional 3.0 GHz P4's (Mixed QMD's and non-QMD's) to move our daily production from 9000 to 11,500 would get us on an upward climb in the ranks.
Bozo
Storage? I am Storage!
I haven't given up yet. Looking at ways to run Folding off line, then dumping the results to Stanford.
Bozo :mrgrn:
Bozo :mrgrn:
CougTek
Hairy Aussie
My Pentium 4 2.66GHz produces between 450 and 600 points per week and it's folding around 18 hours a day. I figured that a 3GHz Pentium 4 would give us around 600 points per week on average. Team Latvia adds 86000 points per week, compared to 42000 for us. 86K - 42K = 44K points per week for us to find if we don't want to be overtaken by them.P5-133XL said:
44000 / 600 = 73.33
I rounded it to 70.
What's wrong with my math?
P.S. I don't use hyperthreading because I read that what's important to the Pande group is that each unit is completed in the shortest possible time. While hyperthreading allows to compute more units, each unit in itself can only be done at roughly 70% of the speed that the processor would be able to do a WU without HT.
Bozo
Storage? I am Storage!
My P4 3GHZ running 24/7 produces 900 points every two days. That is with hyperthreading, running two instances of folding. If I run only one instance of folding the CPU usage is only 50% as shown in Task Manager. How do you get one instance of folding to run at 100%??
Look at my totals for the last few days. Zero on the 19, 1350 on the 20 at 6PM. That is two P4s at 3GHz, running 2 instances of folding each. The CPU is running at 100% in each PC.
Bozo :mrgrn:
Look at my totals for the last few days. Zero on the 19, 1350 on the 20 at 6PM. That is two P4s at 3GHz, running 2 instances of folding each. The CPU is running at 100% in each PC.
Bozo :mrgrn:
I have 3 P4 3.2GHz's with 512MB Ram running 24x7 and they each get around 450 points per day (The majority of my points, the rest of my machines are folding beta-units which tend to get junk points). They get continious QMD's (p1910, p1911, p1912)
I have them configured with a single F@H process with Hyper-threading turned on; running large WU's is turned on; and with the following parameters: -local -forceasm -service -advmethods -verbosity 9
If you are getting lots of QMD's then you are better off with a single instance of F@H. QMD's are memory bandwidth limited rather than CPU limited. If you run two instances, then they compete for the same BW and the two produce fewer points per day than a single. This advise only applies to QMD's and those only run on Intel processors.
If you run with Hyper-threading turned on then the task manager assumes that there are two processors even though the CPU is only simulating two processors. With a single instance of F@H running at 100%then task manager assumes that one CPU is running at capacity and one CPU is not: hense the 50% CPU usage. So with HT turned on the technical answer is that you can't get one instance to show up as 100%.
Now, just because the CPU is listed as only showing 50% does not mean that the CPU isn't operating at 100%. Hyper-Threading is only simulated dual processors and the 2nd simulated processor only operates if there is some parallel operation it can finish which is great if you have 2 dis-similar functions it can finish simultanously and that happens only 5-15% of the time and with two QMD's that 2nd process hurts the total points production as they compete for the same resource -- RAM. If you are doing non-QMD's then 2 F@H processes produce 5%-20% more points per day with the best results when you are running two totally different cores.
The F@H people have requested that with Intel chips that support hyper threading that they only operated a single process even though there is potential for higher points production. A single process with finish quicker and because the next generation of WU is dependant upon the earlier WU they can actually finish a project faster than when using the extra 5-20% speed gain of operating in parallel.
However, as long as they are issuing QMD's then you are better off with a single instance regardless because of their 100% point bonus and they actually run slower with 2 processes.
Hopefully, I have made myself clear about the general case. If not ask Questions.
Cougtek,
The first most glaring problem is that your computer is producing at far to slow of a rate. There's something wrong! I posted my F@H configuration above -- try duplicating it. Also, post your F@H log to see if I can see a problem like lots of Early End Units or some other error that is causing problems.
Next, your math is in error - Team Latvia is only one team. Yes, if they pass us then it drops us down one place, but if we pass someone else then we will go up one place. Look at all the teams ahead of us and behind us and you will see that to keep even all we have to do is increase out point count from 9000 per day (56,000 per week) to around 11,000 per day (77,000 per week). If we can exceed 11K then our ranking will actually go up. Now the other problem with your math is that we are actually producing aprox 60,000+ points per week rather than 42,000: See SF stats summery and Team Overtake to see who we are overtaking and who is overtaking us[/url]
The second problem is that your P4 is producing an abnormally low amount of points. You should be producing around 300-350 per day not 450-600 per week. If what we need is 2000 points per day then we need about 7 CPU's and more if some of them are not getting QMD's which get a 100% bonus.
I have them configured with a single F@H process with Hyper-threading turned on; running large WU's is turned on; and with the following parameters: -local -forceasm -service -advmethods -verbosity 9
If you are getting lots of QMD's then you are better off with a single instance of F@H. QMD's are memory bandwidth limited rather than CPU limited. If you run two instances, then they compete for the same BW and the two produce fewer points per day than a single. This advise only applies to QMD's and those only run on Intel processors.
If you run with Hyper-threading turned on then the task manager assumes that there are two processors even though the CPU is only simulating two processors. With a single instance of F@H running at 100%then task manager assumes that one CPU is running at capacity and one CPU is not: hense the 50% CPU usage. So with HT turned on the technical answer is that you can't get one instance to show up as 100%.
Now, just because the CPU is listed as only showing 50% does not mean that the CPU isn't operating at 100%. Hyper-Threading is only simulated dual processors and the 2nd simulated processor only operates if there is some parallel operation it can finish which is great if you have 2 dis-similar functions it can finish simultanously and that happens only 5-15% of the time and with two QMD's that 2nd process hurts the total points production as they compete for the same resource -- RAM. If you are doing non-QMD's then 2 F@H processes produce 5%-20% more points per day with the best results when you are running two totally different cores.
The F@H people have requested that with Intel chips that support hyper threading that they only operated a single process even though there is potential for higher points production. A single process with finish quicker and because the next generation of WU is dependant upon the earlier WU they can actually finish a project faster than when using the extra 5-20% speed gain of operating in parallel.
However, as long as they are issuing QMD's then you are better off with a single instance regardless because of their 100% point bonus and they actually run slower with 2 processes.
Hopefully, I have made myself clear about the general case. If not ask Questions.
Cougtek,
The first most glaring problem is that your computer is producing at far to slow of a rate. There's something wrong! I posted my F@H configuration above -- try duplicating it. Also, post your F@H log to see if I can see a problem like lots of Early End Units or some other error that is causing problems.
Next, your math is in error - Team Latvia is only one team. Yes, if they pass us then it drops us down one place, but if we pass someone else then we will go up one place. Look at all the teams ahead of us and behind us and you will see that to keep even all we have to do is increase out point count from 9000 per day (56,000 per week) to around 11,000 per day (77,000 per week). If we can exceed 11K then our ranking will actually go up. Now the other problem with your math is that we are actually producing aprox 60,000+ points per week rather than 42,000: See SF stats summery and Team Overtake to see who we are overtaking and who is overtaking us[/url]
The second problem is that your P4 is producing an abnormally low amount of points. You should be producing around 300-350 per day not 450-600 per week. If what we need is 2000 points per day then we need about 7 CPU's and more if some of them are not getting QMD's which get a 100% bonus.
CougTek
Hairy Aussie
I use the graphical client. I see no option to edit the parameters within it.
When my current unit finishes, I'll switch to the console client (which I used before) and see if your parameters can change something to my stats.
When my current unit finishes, I'll switch to the console client (which I used before) and see if your parameters can change something to my stats.
Bozo
Storage? I am Storage!
All the P4 boxes that I'm running have at least 1GB of RAM.
All the boxes that are running F@H are running QMDs.
I have large WU's turned on.
On average it takes about 30 minutes to complete a step on the box running two instances of F@H, with Hyper Threading.
I disabled Hyper Threading on one box and fired up one instance of F@H. CPU usage is at 100%. It takes 14 minutes to complete a step.
Hmmmm More points per day!
Bozo :mrgrn:
All the boxes that are running F@H are running QMDs.
I have large WU's turned on.
On average it takes about 30 minutes to complete a step on the box running two instances of F@H, with Hyper Threading.
I disabled Hyper Threading on one box and fired up one instance of F@H. CPU usage is at 100%. It takes 14 minutes to complete a step.
Hmmmm More points per day!
Bozo :mrgrn:
CougTek said:
Drop the -service parameter for the console unless you are running the console as a service. If you are going to operate F@H as a service then you will need to edit the registry to add the parameters. The alternative is to use a program like FireDaemon (what I use) to run F@H as a service and then you add the parameters in one of FireDaemons fields during setup.
CougTek
Hairy Aussie
Unfortunately, FireDaemon is no longer free. 35$ per computer isn't worth it for me at present.P5-133XL said:
Ah, but the free versions still exist, they are simply no longer supported by the original company. I did a google search and came up with: FireDaemon-Lite-1_6-GA.exe This will work for one service. If you need more than one, then you would have to get the pro version and that has always cost money.
CougTek
Hairy Aussie
Thanks. I figured that it was still floating around but this afternoon, I wasn't in the mood to look for it.
Will Rickards
Storage Is My Life
I'm running the console as a service, two in fact for each of the processors.
Should I be using another method to run them as a service?
I thought once you setup the parameters through the UI it always used them?
My OS is Windows XP and processor is AMD X2 3800+
Should I be using another method to run them as a service?
I thought once you setup the parameters through the UI it always used them?
My OS is Windows XP and processor is AMD X2 3800+
Instructions for manually adding parameters when the console is operated as a service:
1/ Shut down that instance through the Services manager snap-in
2/ Run regedit and go to key HKLM\SYSTEM\CurrentControlSet\Services\<service name> and edit the entry ImagePath, adding whatever flags you wish to the end. Exit regedit.
3/ Re-open the Services manager snap-in, and verify that your changes appear there, and then restart the service.
NOTE: Registry editing can be dangerous if you do it wrong, although the above is pretty straight forward. Use this procedure at your own risk!
I much perfer using FireDeamon to accomplish the same as the native service capability. There are now 3rd party installers, if you want to use them: 3rd Party Tools
1/ Shut down that instance through the Services manager snap-in
2/ Run regedit and go to key HKLM\SYSTEM\CurrentControlSet\Services\<service name> and edit the entry ImagePath, adding whatever flags you wish to the end. Exit regedit.
3/ Re-open the Services manager snap-in, and verify that your changes appear there, and then restart the service.
NOTE: Registry editing can be dangerous if you do it wrong, although the above is pretty straight forward. Use this procedure at your own risk!
I much perfer using FireDeamon to accomplish the same as the native service capability. There are now 3rd party installers, if you want to use them: 3rd Party Tools
Bozo
Storage? I am Storage!
Every work unit helps
If not for the team, surely for the defeat of cancer.
Bozo :mrgrn:
If not for the team, surely for the defeat of cancer.
Bozo :mrgrn:
CougTek
Hairy Aussie
Forget hyperthreading, I've just find out that my Pentium 4 505 doesn't support it. I've set my client as a service and I'm curently crunching a QMD unit. We'll see the results.
The only flag set in the registry is -svcstart. I've rebooted my system a few times since yesterday and every time, when I check the fahlog, it enables SSE3 instructions when it restarts. So I don't think I'll need to add the -forceasm flag as it seems to do it automatically. When I ran the -config, I told it to seek for +5MB units and to ask for newly released units. This is like adding -advmethods. I'll monitor my fahlog file often for the rest of the week, but I don't think that adding -forceasm and -advmethods to the registry will be necessary.
Thanks for the help.
The only flag set in the registry is -svcstart. I've rebooted my system a few times since yesterday and every time, when I check the fahlog, it enables SSE3 instructions when it restarts. So I don't think I'll need to add the -forceasm flag as it seems to do it automatically. When I ran the -config, I told it to seek for +5MB units and to ask for newly released units. This is like adding -advmethods. I'll monitor my fahlog file often for the rest of the week, but I don't think that adding -forceasm and -advmethods to the registry will be necessary.
Thanks for the help.
Hah. I was just thinking about that and was going to ask...
I unearthed an evil PC Chips motherboard over the weekend that actually came free (yes, they're giving these things away!) with the 1.6MHz Duron processor that's running in my main system. So ... if anyone has an old Duron processor lying around that they'd like to get rid of, please let me know.
I might be able to trade a nice Pentium Pro motherboard, CPU, and some lovely Fast Page Mode RAM for it.
I unearthed an evil PC Chips motherboard over the weekend that actually came free (yes, they're giving these things away!) with the 1.6MHz Duron processor that's running in my main system. So ... if anyone has an old Duron processor lying around that they'd like to get rid of, please let me know.
I might be able to trade a nice Pentium Pro motherboard, CPU, and some lovely Fast Page Mode RAM for it.
.
Cover your eyes, Tannin!
It's a PC Chips M811 motherboard. According to the manual, it includes a "VIA KT266A + VT8235 chipset" and it "accomodates AMD Athlon processors with FSB speeds up to 133MHz." Does that help?
I did actually get this board up and running using my Duron processor at one point, but disk access was absolutely psychotic (first time I've ever used that description for an IDE controller). I have an old Promise original "FastTrack 100" pseudo-RAID controller somewhere. Does anyone know if you can just use a card like that as a standard IDE controller for an HDD and DVD drive?
Cover your eyes, Tannin!
It's a PC Chips M811 motherboard. According to the manual, it includes a "VIA KT266A + VT8235 chipset" and it "accomodates AMD Athlon processors with FSB speeds up to 133MHz." Does that help?
I did actually get this board up and running using my Duron processor at one point, but disk access was absolutely psychotic (first time I've ever used that description for an IDE controller). I have an old Promise original "FastTrack 100" pseudo-RAID controller somewhere. Does anyone know if you can just use a card like that as a standard IDE controller for an HDD and DVD drive?
Pradeep
Storage? I am Storage!
Had to remove it from quite a few machines. But I have now returned, with I think five P4 3.6GHzs, running the big mama unit.
BTW there is now a beta 5.04 client, that allows you to specify -advmethods all the time under advanced options. Also an option to allow you to define how much available memory it reports back to HQ (in case you still want large units, but not the huge ones).
BTW there is now a beta 5.04 client, that allows you to specify -advmethods all the time under advanced options. Also an option to allow you to define how much available memory it reports back to HQ (in case you still want large units, but not the huge ones).
CougTek
Hairy Aussie
Pradeep, Groltz and Bozo are engaged in one Hell of a war for the 7th spot (soon for the 6th as Jake no longer participates). I'm ungry for pop corn every time I look at the stats.
Groltz
My demeaning user rank is
CougTek said:
I don't consider it a war.
Unlike some people, my units are about 90% earned from my two home computers. I don't have the opportunity to go around my workplace Borging high-powered machines to fold for me (at least anymore). I just let my two home machines do what they can. Most of the rest comes from an A30p Thinkpad.
Bozo
Storage? I am Storage!
90% of my units also come from my home machines. I have to shut down the boxes occasionally to keep my electric bill under control.
Hmm, maybe solar cells...........
Bozo :mrgrn:
Hmm, maybe solar cells...........
Bozo :mrgrn:
Looking at our stats, some observations.
Welcome Andy - our newest folder.
i has drasticly increased his folding production this past week.
Cougtek after reconfiguring his client is doing much better at producing team points.
After a long absense, Bartender is again folding.
The race for position 6,7,8 seems to be shaping up to be a real horse race between Bozo, Mourn, and Pradeep with Pradeep a long-term favorite. However, at their production levels they are actually competing for 2,3, and 4.
f_bostrom must have put more CPU capability online: His production has increased 50% in October.
Overall, October has been a good month as I suspect that several people have started warming their houses using surplus computer heat rather than keeping them off, so that they don't have to air-condition.
Welcome Andy - our newest folder.
i has drasticly increased his folding production this past week.
Cougtek after reconfiguring his client is doing much better at producing team points.
After a long absense, Bartender is again folding.
The race for position 6,7,8 seems to be shaping up to be a real horse race between Bozo, Mourn, and Pradeep with Pradeep a long-term favorite. However, at their production levels they are actually competing for 2,3, and 4.
f_bostrom must have put more CPU capability online: His production has increased 50% in October.
Overall, October has been a good month as I suspect that several people have started warming their houses using surplus computer heat rather than keeping them off, so that they don't have to air-condition.
I'm running inconsistent at the moment until my new machine takes place. I have a part that needs an RMA< and then I'll have my original XP 2200+ back online, and then the new A64 3200+. Right now I'm experimenting with win Vista and F@H. So far it's running good. And damn, that A64 Venice core runs cool with the XP-90C. 88 Degrees F at full load, or 77 at idle.
Bozo
Storage? I am Storage!
I have computers all round me at work. Most are idle 90% of the time. Unfortunately, only one is connected to the internet. For months I've been trying to figure out a way to get some folding done. ( I despise cancer) I think I have found a way.
1) All the computers must be on a network that connects to a computer with internet access. (internet sharing is locked out on our corp boxes.)
2) Copy the folding client exe to a folder on the slave box. Make sure the folder is shared out.
3) Create a shortcut to the exe and put it in the same folder as the folding client. Add 'switches' if desired.
4) Go to the master box (with internet accesibility) and drill into the slave box and the folding folder through the network.
5) Double click on the shortcut. The program should start and you can set up the config file.
6) Once the slave folder is folding, stop the client.
7) Return to the slave box ( I use VNC ) and start the client with the shortcut.
You must monitor the clients progress. When it has finished, shut down the client and the run it from the master. When it has uploaded the work and downloaded a new work unit, stop the client on the master and then start it on the slave.
Seems to be working on a single box for me. Time will tell
Bozo :mrgrn:
1) All the computers must be on a network that connects to a computer with internet access. (internet sharing is locked out on our corp boxes.)
2) Copy the folding client exe to a folder on the slave box. Make sure the folder is shared out.
3) Create a shortcut to the exe and put it in the same folder as the folding client. Add 'switches' if desired.
4) Go to the master box (with internet accesibility) and drill into the slave box and the folding folder through the network.
5) Double click on the shortcut. The program should start and you can set up the config file.
6) Once the slave folder is folding, stop the client.
7) Return to the slave box ( I use VNC ) and start the client with the shortcut.
You must monitor the clients progress. When it has finished, shut down the client and the run it from the master. When it has uploaded the work and downloaded a new work unit, stop the client on the master and then start it on the slave.
Seems to be working on a single box for me. Time will tell
Bozo :mrgrn:
CougTek
Hairy Aussie
1. Who's Andy? Adcadet already has two accounts in our team.
2. Doug, does it mea I won't be able to take back my 11th spot?
3. Clown. You're giving yourself a lot of trouble. Don't risk your job for F@H.
2. Doug, does it mea I won't be able to take back my 11th spot?
3. Clown. You're giving yourself a lot of trouble. Don't risk your job for F@H.
CougTek
Hairy Aussie
Funny I am. "Does it mean" is what my fingers were trying to type.CougTek said:
Pradeep
Storage? I am Storage!
Oooh, what's this the FedEx man has delivered. Eight P4 3.8GHz machines. Dell end of quarter madness.
I'll be gentle. I won't bring them all up at once
I'll be gentle. I won't bring them all up at once
Coug, we can fight for that spot, I don't know what will happen.
My dell is running 24x7, but it doesn't crank out that many units with its poky celeron. I definitly run all my work units on home PC's. None are coming from work. I don't want to risk my job for the F@H effort.
My dell is running 24x7, but it doesn't crank out that many units with its poky celeron. I definitly run all my work units on home PC's. None are coming from work. I don't want to risk my job for the F@H effort.
CougTek
Hairy Aussie
My scary 2.66GHz Pentium 4 against your Athlon XP 2200+ and a pair of 2GHz Athlon 64 cores. What a fight it will be...
I wish Pradeep's delivery man would pass by my door too.
I wish Pradeep's delivery man would pass by my door too.