Last year I was running a triple core, a GPU, two linux systems, and I had a PS3 intermittently running. The 40-gig PS3 is a bit at my wife's mercy since she often leaves it in the DVD mode. I don't dare run my original 20-gig PS3 on folding because I already had to send it in once.
The triple-core died last year and took my GPU folding with it. Since then, I've upgraded my Athlon x2 linux box to an Athlon-II x3 and set it to running a3 units, I spent $2000 building the i7-920 system and set it to running VMWare with linux and the A3 SMP session. I tried to run my HD5870 on it, but it was too buggy and was heating up too much. I then took my Athlon X2, installed a copy of windows on it, and put my old HD4xxx series cards on it. My Pentium 4 2.53 is still running out in the garage, but it only puts out 200 points per day now and sucks a horrific amount of juice.
I don't do any OC because I like my systems stable. Same reason I gave up 4000 points per day not running the GPU folding on the 5870. Waiting for GPU3... I was getting nearly 6000 points per day without it until shortly before I swapped all my SMP systems over to running the new 6.29 client... Now I barely break 2500... That's not counting the potential 900points per day from the PS3 when my wife remembers to exit her DVDs instead of just changing the tv back to the satellite.
As for the hypothetical, I'm trying to get them to understand our position that we're not taking over a team, we're trying to get it back, and trying to define exactly what their opposition is...
Taking over a Team
Moderators: Site Moderators, FAHC Science Team
Re: Taking over a Team
I disagree. How do you attract new people to it? Psychologically speaking, it's far easier to get people to do something if you give them a goal, and group them up and pit them against someone else. Gamers especially. The team system is there to permit us to drag new members in via existing members. I was folding just for science, but then I decided to recruit some of my friends to the project as well. Most of them have stopped folding over the last year because the team name was no longer accurate... and the psychological setback of losing all of the work would probably make it hard for me to get most of them going again.Wrish wrote: Your folding contributions are to Stanford's project to further a biophysical science; it's only by their grace that you have a rudimentary team structure and stats to identify with. For other distributed projects, I see people start over with new names just to change their team tag. Me, I fold under team Default because there's no intrinsic teamwork in processing work units.
Ok, I'll bite. Exactly how have you been getting an i7 to do 29,000 ppd? I had it doing about 4500 ppd back on the 6.02 client, and I've had the GPU on it do about 4000 ppd, but both of them tapered off, and with A3 points it's only managing 1000ppd. (I don't think I've put in enough A3 units even with 2 clients doing them to make 10 so far... A1/2 units took my Linux box about a day to produce 1920 points, and it's down to about 600ppd now with running the new core with a passkey. It not always getting A3 units either even with the passkey. The last one it got was P2653.)Oh, and let me tell you about that i7. It draws 300-370W depending on the OC, cooling, and luck on the chip - so about 2 PS3's. It puts out 29,000 PPD, or 32 PS3s' worth. Folding stats have been like that over the years - inflating with Moore's Law.
Re: Taking over a Team
It's off-topic, but I'll let it pass this time because it's a bit of respite from the semi-hostile tone this thread has been taking. Look atkazriko wrote:Ok, I'll bite. Exactly how have you been getting an i7 to do 29,000 ppd? I had it doing about 4500 ppd back on the 6.02 client, and I've had the GPU on it do about 4000 ppd, but both of them tapered off, and with A3 points it's only managing 1000ppd. (I don't think I've put in enough A3 units even with 2 clients doing them to make 10 so far... A1/2 units took my Linux box about a day to produce 1920 points, and it's down to about 600ppd now with running the new core with a passkey. It not always getting A3 units either even with the passkey. The last one it got was P2653.)
-bigadv bonus point calculator and SMP2 bonus point calculator. These two calculators should allow you to get a fairly accurate idea of what bonus you will receive . . . .
Posting FAH's log:
How to provide enough info to get helpful support.
How to provide enough info to get helpful support.
Re: Taking over a Team
Ahh, Bigadv is why. I didn't put that in place because it would require doubling the amount of ram and CPU cores needed for folding, making the system mostly unusable for anything else. (My wife doesn't let me buy hardware for folding unless I can dual-use it...)
Without that, it seems that according to extremeoverclocking's stats, it would take 650 days to make up the points we would lose... And the cheapest I could build a good quality i7 system for was closer to $1100 when I looked last, so buying it as a dedicated isn't an option...
Without that, it seems that according to extremeoverclocking's stats, it would take 650 days to make up the points we would lose... And the cheapest I could build a good quality i7 system for was closer to $1100 when I looked last, so buying it as a dedicated isn't an option...
Re: Taking over a Team
Seeing you average 4000 PPD over the last few weeks I would never have guessed that you're running an i7 plus a dual-core with HD48xx. Do you fold under a different name? Or not leave your systems on 24/7? Then again, the 650-day quote comports with 4k PPD. It looks like rectifying all the inefficiencies in your current hardware/OS setup may add way more net production than all the former teammates you were hoping to recruit from that site.
2) I don't understand why you're running an A3 in a Linux VM when there's a Windows client for A3 that runs faster/leaner and can use more than 4 threads unlike some VMs. You can configure it to run -smp 7, which shouldn't slow you down any more than -smp 4. That means a safe margin of Vcore above that necessary to sustain the OC so that you'll maintain that stability you value at stock, along with an increase in performance so that you'll feel the effects of FAH even less than you do now.
4) My 29,000+ PPD figure is from a fairly optimized/dedicated i7 assembled Nov. 2008, but nowadays the steppings are better and this result easier to achieve. It's an average C0-stepping 965 brought up to 4.1ghz (195 x 21) with 1.46 Vcore and third-party air cooling, with fairly aggressive RAM timings and a stripped down Ubuntu (GUI-less). I'm now at 29:23 TPF, which assuming 20 minutes of network overhead per unit translates into 29,885 PPD. My point is if I can do this on year-old hardware, the chances are high that it can be done on an i7 bought more recently. So much for depending on team mates for 900 PPD!
).
1) That's a lot of money for an i7-920! 4 GB of RAM today is a hair over $100. I assume you have 3 of 6 slots occupied. Adding 2x 2GB would let you start a VM with -bigadv. I'm now running a native -bigadv -smp 8 client with only 3 GB, and swapping on a 7200-rpm 320gb drive seems to slow a 2-day unit by only a few minutes.I spent $2000 building the i7-920 system and set it to running VMWare with linux and the A3 SMP session.
2) I don't understand why you're running an A3 in a Linux VM when there's a Windows client for A3 that runs faster/leaner and can use more than 4 threads unlike some VMs. You can configure it to run -smp 7, which shouldn't slow you down any more than -smp 4.
3) Turn off that silly P4 in the garage and use the juice to OC your i7 within safe levels.My Pentium 4 2.53 is still running out in the garage, but it only puts out 200 points per day now and sucks a horrific amount of juice.
That means a safe margin of Vcore above that necessary to sustain the OC so that you'll maintain that stability you value at stock, along with an increase in performance so that you'll feel the effects of FAH even less than you do now.4) My 29,000+ PPD figure is from a fairly optimized/dedicated i7 assembled Nov. 2008, but nowadays the steppings are better and this result easier to achieve. It's an average C0-stepping 965 brought up to 4.1ghz (195 x 21) with 1.46 Vcore and third-party air cooling, with fairly aggressive RAM timings and a stripped down Ubuntu (GUI-less). I'm now at 29:23 TPF, which assuming 20 minutes of network overhead per unit translates into 29,885 PPD. My point is if I can do this on year-old hardware, the chances are high that it can be done on an i7 bought more recently. So much for depending on team mates for 900 PPD!
This is true, but applies best to a robust team system, which FAH doesn't have. Unlike some other projects, nothing stops a stranger from submitting units under your team name w/o your permission, or even under your name for that matter (though Stanford can tell the difference, the public stats don't show it, and if you care about stats, you only care to the extent that the public can see, right?How do you attract new people to it? Psychologically speaking, it's far easier to get people to do something if you give them a goal, and group them up and pit them against someone else. Gamers especially.
).Re: Taking over a Team
Been busy recently, haven't had time to update this. We worked something out that will at least shave our time to get back up to normal by several months, and in exchange I dedicated my I7 to bigadv for a month...
I don't overclock. Ever. People may call me crazy, but it's just a policy of mine... (I actually tried to bend it for this month pushing it to a 3000mhz as long as I didn't need to up the voltage, but it locked up as soon as I applied the settings. I kicked it back to 2667...)
Anyway, the total changes I made were...
Upgraded my Athlon x2 4800+ system to an AthlonIIx3 running at 2.9ghz, disabled CNQ, and now it's putting out just short of 3000ppd. ($150.
Put together the Athlonx2 for my wife to use, put in an old HD4670 card, it was getting 1000ppd. (Free, all spare parts.)
I couldn't leave it at that though, so...
Upgraded the wife's computer to an AthlonX2 5800, installed a $99 GT9800 card, upgraded ram to 4gigs. Total price, about $230 Pushes system from 1000ppd to 4000ppd, with improved performance playing Spore.
Took the parts that this freed up and upgraded the P4 in the garage to the X2 4800, it's 32 bit OS still, so only getting about 500ppd.
Work let me use a couple of X2 5200's, so that adds about 1300ppd.
Turned on two of my PS3's for another 1800ppd.
So it drifts between 23k and 26k depending on what work units it gets...
Now that I've done all this optimization, it's inspired at least one of my team mates to tune up his systems with cheap upgrades like that too...
I wasn't running both. I had the i7 running A1 units under a VM, and the HD5870 was turned off because it was too unstable. I did find inefficiencies, but mostly in the linux boxes. I spent $400 additional in hardware recently and gave up the use of my i7 for a month, and am up to 25k per day.Wrish wrote:Seeing you average 4000 PPD over the last few weeks I would never have guessed that you're running an i7 plus a dual-core with HD48xx. Do you fold under a different name? Or not leave your systems on 24/7? Then again, the 650-day quote comports with 4k PPD. It looks like rectifying all the inefficiencies in your current hardware/OS setup may add way more net production than all the former teammates you were hoping to recruit from that site.
It's 6 gigs of ram, the HD5870, and top of the line for every component in it... I wasn't trying to make a FAH dedicated machine, it's my main gaming PC... It's a 1tb drive as well. The main problem with running bigadv was that VMware goes absolutely bonkers trying to run bigadv units for some reason. Says it can't access the drive, I think it's running windows out of file handles or something. The system is currently running bigadv on native linux at least for a month here.1) That's a lot of money for an i7-920! 4 GB of RAM today is a hair over $100. I assume you have 3 of 6 slots occupied. Adding 2x 2GB would let you start a VM with -bigadv. I'm now running a native -bigadv -smp 8 client with only 3 GB, and swapping on a 7200-rpm 320gb drive seems to slow a 2-day unit by only a few minutes.I spent $2000 building the i7-920 system and set it to running VMWare with linux and the A3 SMP session.
At the time, I didn't know the windows A3 client even existed. ChelesaOilman pointed me at it, and once I'm done with this one-month challenge I'll switch to using it.2) I don't understand why you're running an A3 in a Linux VM when there's a Windows client for A3 that runs faster/leaner and can use more than 4 threads unlike some VMs. You can configure it to run -smp 7, which shouldn't slow you down any more than -smp 4.
I can't get rid of it without having a replacement for its role in my house. It's the DHCP/DNS/Django server for the house, and it's rack mounted and always running, unlike some of my other systems that get periodically rebooted. I actually did replace it though. It's now running a Athlon X2 4800 running two clients and getting around 500 ppd. Eventually I'll upgrade it to Ubuntu 64-bit and it should get more like 800ppd on A3.3) Turn off that silly P4 in the garage and use the juice to OC your i7 within safe levels.My Pentium 4 2.53 is still running out in the garage, but it only puts out 200 points per day now and sucks a horrific amount of juice.
I don't overclock. Ever. People may call me crazy, but it's just a policy of mine... (I actually tried to bend it for this month pushing it to a 3000mhz as long as I didn't need to up the voltage, but it locked up as soon as I applied the settings. I kicked it back to 2667...)
The i7 is getting about 16k dedicated at stock speeds... I use stock cooling as well.4) My 29,000+ PPD figure is from a fairly optimized/dedicated i7 assembled Nov. 2008, but nowadays the steppings are better and this result easier to achieve. It's an average C0-stepping 965 brought up to 4.1ghz (195 x 21) with 1.46 Vcore and third-party air cooling, with fairly aggressive RAM timings and a stripped down Ubuntu (GUI-less). I'm now at 29:23 TPF, which assuming 20 minutes of network overhead per unit translates into 29,885 PPD. My point is if I can do this on year-old hardware, the chances are high that it can be done on an i7 bought more recently. So much for depending on team mates for 900 PPD!
Anyway, the total changes I made were...
Upgraded my Athlon x2 4800+ system to an AthlonIIx3 running at 2.9ghz, disabled CNQ, and now it's putting out just short of 3000ppd. ($150.
Put together the Athlonx2 for my wife to use, put in an old HD4670 card, it was getting 1000ppd. (Free, all spare parts.)
I couldn't leave it at that though, so...
Upgraded the wife's computer to an AthlonX2 5800, installed a $99 GT9800 card, upgraded ram to 4gigs. Total price, about $230 Pushes system from 1000ppd to 4000ppd, with improved performance playing Spore.
Took the parts that this freed up and upgraded the P4 in the garage to the X2 4800, it's 32 bit OS still, so only getting about 500ppd.
Work let me use a couple of X2 5200's, so that adds about 1300ppd.
Turned on two of my PS3's for another 1800ppd.
So it drifts between 23k and 26k depending on what work units it gets...
Now that I've done all this optimization, it's inspired at least one of my team mates to tune up his systems with cheap upgrades like that too...
Re: Taking over a Team
While I applaud your upgrades and reconfigurations - that PPD jump must be gigantic from your point of view - do know that the most efficient way of spending <$250 is on folding graphics cards, and the most efficient way of spending >$600 is dedicated i7s qualified for bigadv (you can reduce costs further by using a P55 platform).
When you swap back to Windows/A3, your i7 should put out ~10k PPD untouched. It will be interesting to adjust settings for your usage pattern with things like turning off core affinity and running with -smp 7.
The HD5870 is 4k+ PPD. Being able to stabilize that existing hardware is an asset. You've read up about environment variables and have been using them for the HD4670, right? Unlike the HD4670, the 5800 series can usually be undervolted (but you may need to underclock slightly).
Overclocking is not as simple as turning up a dial and watching for a few moments to see that it doesn't blow up. There are many articles ranging from novice to veteran about overclocking the i7. Among at least two dozen i7 installations (mostly 920s) I've not come across one that couldn't break 3.2 GHz at its stock voltage. Whether I can maintain that record by adjusting only a single dial, though, is another story, since only on the $1000 extreme editions is there even a dial to increase just the processor frequency.
Search kakao for my forum name and you'll see what performance I extract out of similar aggregate hardware on my home network, nothing of which was purchased with folding in mind:
i7 965 w/ 3 GB @ 4+ Ghz
Q9550 w/ 4 GB @ 3.8 GHz
E7500 w/ 4 GB @ 3.2 GHz undervolted
GTX280 @ 1429 MHz undervolted
P7450/HD4650 4 GB Vaio laptop, stock everything
Basically, I use the laptop as my workstation when possible, and I place the GTX280 on the E7500. The Q9550 serves most of my on-demand capacity since I don't have to restart to get into Windows. It has a current uptime of >1 month (HFM reports 93 consecutive SMP units w/o a fail except for the core 2.17 switchover) and hasn't blue-screened in the half year since finalizing those clocks. I've been OC'ing on DIY assemblies since 2006, though folding since only about 2 months ago.
When you swap back to Windows/A3, your i7 should put out ~10k PPD untouched. It will be interesting to adjust settings for your usage pattern with things like turning off core affinity and running with -smp 7.
The HD5870 is 4k+ PPD. Being able to stabilize that existing hardware is an asset. You've read up about environment variables and have been using them for the HD4670, right? Unlike the HD4670, the 5800 series can usually be undervolted (but you may need to underclock slightly).
Why am I not surprised? If you never get your feet wet how can you have any experience.I don't overclock. Ever. People may call me crazy, but it's just a policy of mine... (I actually tried to bend it for this month pushing it to a 3000mhz as long as I didn't need to up the voltage, but it locked up as soon as I applied the settings. I kicked it back to 2667...)
Overclocking is not as simple as turning up a dial and watching for a few moments to see that it doesn't blow up. There are many articles ranging from novice to veteran about overclocking the i7. Among at least two dozen i7 installations (mostly 920s) I've not come across one that couldn't break 3.2 GHz at its stock voltage. Whether I can maintain that record by adjusting only a single dial, though, is another story, since only on the $1000 extreme editions is there even a dial to increase just the processor frequency.Search kakao for my forum name and you'll see what performance I extract out of similar aggregate hardware on my home network, nothing of which was purchased with folding in mind:
i7 965 w/ 3 GB @ 4+ Ghz
Q9550 w/ 4 GB @ 3.8 GHz
E7500 w/ 4 GB @ 3.2 GHz undervolted
GTX280 @ 1429 MHz undervolted
P7450/HD4650 4 GB Vaio laptop, stock everything
Basically, I use the laptop as my workstation when possible, and I place the GTX280 on the E7500. The Q9550 serves most of my on-demand capacity since I don't have to restart to get into Windows. It has a current uptime of >1 month (HFM reports 93 consecutive SMP units w/o a fail except for the core 2.17 switchover) and hasn't blue-screened in the half year since finalizing those clocks. I've been OC'ing on DIY assemblies since 2006, though folding since only about 2 months ago.
Last edited by Wrish on Fri Mar 26, 2010 1:06 am, edited 1 time in total.
Re: Taking over a Team
The one thing I'd say about OC'ing, is it reallllyy pays to get a third-party heatsink for it.
Beside that, you say you've got top of the line components for it... why have you skimped @ the CPU cooling?
No offence to Intel, but the stock cooler for the 1366 socket is horrendous; in a case with high ambient temperatures, this is exacerbated. Even at stock speeds you'll notice a far lower temperature using a high quality heat sink (air) like that from Noctua (good AND quiet) or Thermaltake etc. (my stock cooler is now quite a nice paperweight)
Even if I hadn't overclocked my 920 to 3.8 I'd have still got my third-party heatsink. People will tell you that the i7 copes with high temperatures well, but I get uneasy about having them above 70-75 degs. You'll also benefit from quieter operation. It doesn't take much of a load at stock for the fan to start going hammer and tongs!
Beside that, you say you've got top of the line components for it... why have you skimped @ the CPU cooling?
No offence to Intel, but the stock cooler for the 1366 socket is horrendous; in a case with high ambient temperatures, this is exacerbated. Even at stock speeds you'll notice a far lower temperature using a high quality heat sink (air) like that from Noctua (good AND quiet) or Thermaltake etc. (my stock cooler is now quite a nice paperweight)
Even if I hadn't overclocked my 920 to 3.8 I'd have still got my third-party heatsink. People will tell you that the i7 copes with high temperatures well, but I get uneasy about having them above 70-75 degs. You'll also benefit from quieter operation. It doesn't take much of a load at stock for the fan to start going hammer and tongs!
Re: Taking over a Team
This topic is about taking over a Team, not about hardware or overclocking. I'm closing it. I should have done that earlier.
We have a forum for that sort of discussions.
We have a forum for that sort of discussions.
Posting FAH's log:
How to provide enough info to get helpful support.
How to provide enough info to get helpful support.