That's truly stupid. I've got several machines (C2D) here in the office being under my control. They could easily do 2500ppd together during our working time. But it's not an option to have them running 24/7, though.
Does anyone know when there'll be a SMP client for Windows, which supports checkpointing? I'd like to add a few points to my team...
Rant: Why's there still no SMP client w/ checkpointing?
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Hey Mexell,
I was not aware that the Windows SMP client did not already do that.
There is another option that may not be as convenient, but that may even generate more points for you.
You can run VMware under windows, and then run Linux in the virtual machine, and fold using the Linux SMP client. If you have 64-bit hardware, you can use 64-bit Linux, and the 64-bit Linux client.
I was not aware that the Windows SMP client did not already do that.
There is another option that may not be as convenient, but that may even generate more points for you.
You can run VMware under windows, and then run Linux in the virtual machine, and fold using the Linux SMP client. If you have 64-bit hardware, you can use 64-bit Linux, and the 64-bit Linux client.
At least my SMP installation saves the progress once every 1%. It worked when I ran one instance of SMP on my Quad, and it works when I run two instances of it (installed into separate directories).
There's some problems with shutting down a SMP client with two instances running. Doing so may corrupt one or both of the work units in progress. Because of this, I cannot shut them down properly (CTRL+C) but instead I reboot the computer without shutting either client. That way I don't lose the work done.
When I was running just one instance, I could CTRL+C safely. It may still be possible to shut down a client from CRTL+ALT+DEL and kill the processes forcibly, not letting them to try proper shutdown, and thus avoid corruption, but avoiding rebooting Windows.
mexell shouldn't worry about with what I just wrote as these issues didn't surface until I started running several SMP clients on same computer, simultaneously (with FAH Affinity Changer). (My PPD increased almost about 40%. I'm now finishing two work units every 22 hours. So I consider it worth the tweaking.)
There's some problems with shutting down a SMP client with two instances running. Doing so may corrupt one or both of the work units in progress. Because of this, I cannot shut them down properly (CTRL+C) but instead I reboot the computer without shutting either client. That way I don't lose the work done.
When I was running just one instance, I could CTRL+C safely. It may still be possible to shut down a client from CRTL+ALT+DEL and kill the processes forcibly, not letting them to try proper shutdown, and thus avoid corruption, but avoiding rebooting Windows.
mexell shouldn't worry about with what I just wrote as these issues didn't surface until I started running several SMP clients on same computer, simultaneously (with FAH Affinity Changer). (My PPD increased almost about 40%. I'm now finishing two work units every 22 hours. So I consider it worth the tweaking.)
Last edited by whiic on Wed Feb 20, 2008 9:21 am, edited 1 time in total.
The thing is, even if there's a technical solution for my problem (that you and me find acceptable), that's no practical solution. (meaning: it has to run in a way that it doesn't get into the focus of someone else than me or that it disturbs my stressed real estate agent colleagues at their work 
). Seems like I will have to wait a bit...
). Seems like I will have to wait a bit...Hi again,
Although not as simple and convenient as you would like, or as they should have to keep their volunteer help happy, but if you set the priority of virtual machines to low, then the users should not be impacted, and they will get the CPU when they need it, and spare cycles and cycles after hours would still fold. (Not really spare, as it will keep the machine running constantly which will use more power, keep it hotter, etc.)
Also, you can then close the VM console, and folding can continue pretty invisibly unless you look in task manager.
I have not tried it yet, but I think that you could even connect to each of the virtual machines to check on the status remotely.
Although not as simple and convenient as you would like, or as they should have to keep their volunteer help happy, but if you set the priority of virtual machines to low, then the users should not be impacted, and they will get the CPU when they need it, and spare cycles and cycles after hours would still fold. (Not really spare, as it will keep the machine running constantly which will use more power, keep it hotter, etc.)
Also, you can then close the VM console, and folding can continue pretty invisibly unless you look in task manager.
I have not tried it yet, but I think that you could even connect to each of the virtual machines to check on the status remotely.
Hi all, it's been years since I have folded, but for some reason last night I got the bug to start up again. I initially made the mistake of installing the regular GUI client, found FahMon thanks to another thread and saw after 4 checkpoints it was averaging 155ppd on my old S939 Opteron 185 (2.6Ghz dual core). I looked up my stats and saw that I was at 9860 points or something, so I figured maybe if I at least ran for a couple weeks I could get my score over 10,000 as some sort of milestone.
I then found all the threads about the SMP client and I figured at the very least that would mean a lot better utilization than having the GUI running and just grabbing one core. That's when I realized that the SMP client produces way higher PPD, like 1080ppd showing after my system ran all night.
So as dual and quad core processors continue to gain popularity, is it likely that they will eliminate the multi-core premium, or bring it in line with the single-core units? I could understand if I was making 1.8x the PPD running the SMP on a dual core, but from 155 to 1080 is a 7X increase, which is hillarious to me. Is there any talk of it being eliminated, or is it more likely they'll keep it this way to provide an incentive that everyone upgrade their computers to multi-core CPU's?
I then found all the threads about the SMP client and I figured at the very least that would mean a lot better utilization than having the GUI running and just grabbing one core. That's when I realized that the SMP client produces way higher PPD, like 1080ppd showing after my system ran all night.
So as dual and quad core processors continue to gain popularity, is it likely that they will eliminate the multi-core premium, or bring it in line with the single-core units? I could understand if I was making 1.8x the PPD running the SMP on a dual core, but from 155 to 1080 is a 7X increase, which is hillarious to me. Is there any talk of it being eliminated, or is it more likely they'll keep it this way to provide an incentive that everyone upgrade their computers to multi-core CPU's?
-
GentleGiant
- Friend of SPCR
- Posts: 34
- Joined: Sat Oct 16, 2004 3:21 pm
- Location: Socorro, New Mexico, USA
It's not checkpoints, but deadlines that are the problem
The Windows SMP client (or core, really) does checkpoint in pretty much the same way as the conventional single-thread client.
What's likely the problem is the current short deadline (4 days) on SMP units. A dual core machine folding 8 hours/day will miss the deadlines. My 2 Ghz iMac C2D runs 24 hours/day and is just 2.6x the minimum speed to make deadlines.
If you're only running 1/3 the day and since the Windows client runs about 20% slower than the Mac OSX & Linux clients --- you'll never complete a unit on time.
When the SMP client leaves "beta" status the deadlines should get relaxed, but until then only the standard CPU client can work with machines that aren't 4 cores or folding pretty much full time
What's likely the problem is the current short deadline (4 days) on SMP units. A dual core machine folding 8 hours/day will miss the deadlines. My 2 Ghz iMac C2D runs 24 hours/day and is just 2.6x the minimum speed to make deadlines.
If you're only running 1/3 the day and since the Windows client runs about 20% slower than the Mac OSX & Linux clients --- you'll never complete a unit on time.
When the SMP client leaves "beta" status the deadlines should get relaxed, but until then only the standard CPU client can work with machines that aren't 4 cores or folding pretty much full time