11th Annual Folding@Home Holiday Season 3-way Race

[crashtech]

Thanks for the welcome, gents! I guess my real question was was I needed to do anything to enter the race other than to post my intent here and be folding on Dec 1.

Here's one of my machines, funny how it hasn't been utilizing the 5820K at all. I guess CPUs barely matter these days:

 

[TennesseeTony]

Yeah! Another 5820K user!

Actually, you haven't [edit:] the client hasn't yet downloaded a task for the CPU yet in that pic. The reason is 11. F@H prefers tasks with even numbers of cores/threads, is my limited understanding. You will want to reserve 2 threads per GPU anyway. Why it auto configures odd numbered CPU tasks is one of the F@H quirks.

So, how to change the CPU slot....uhm....Go to Configure/Slots and REMOVE the CPU, then re-add the CPU and instead of -1 threads tell it to use 10 threads, or some lesser even number.
Or, if you prefer you can put the CPU to use in medical projects (BOINC) that can't use GPU power, such as Rosetta, World Community Grid, others.

 

[crashtech]

So an inputted value of -2 would enable F@H to use 10 threads? I'll try that when I get home.

 

[Markfw]

I think just select 10 instead of -1

 

[iwajabitw]

As soon as my boinc cach has exhausted Ill get back over on this project. MW has been messed up for weeks as well as Seti,just have Asteroids and Rosetta to clear out.

 

[TennesseeTony]

So an inputted value of -2 would enable F@H to use 10 threads? I'll try that when I get home.

No no, like Mark said, if you want to use 10 threads input 10.

8 for 8, 4 for quad cores, etc. Don't forget to subtract one full core (or two threads if hyper-threaded like your system) for each GPU in use. Thus 10 threads instead of a full 12 for your system.

I imagine the only time you want to use an odd number here is if you only want to use a single core/thread. My understanding is that Stanford has to actually create these tasks ahead of time, and they program the most common tasks first, for the single/dual/quad cores, and then the quad cores with HT (8). Although they have updated the code many times over the life of the project, I seem to recall (don't quote me) that the original code was so effective from day one, that it's still in there somewhere, and it doesn't like odd numbers of processors/threads.

 

[Markfw]

And yes, the days of multi-core in FAH are gone. My 24 core server that has 20 cores enabled gets 82k, and thats a record, its usually like 40-60k. NY dual6234's only were getting 22k, so I turned off CPU for that server.

 

[crashtech]

So maybe a Celeron running a 980ti would be cost effective? I'll have to see how much CPU power each GPU actually requires.

Also, my farmhouse cellar days are long gone, so I have to get creative about locating the machines these days. I'm even thinking about stuffing a GPU into my NAS, although I understand running F@H on FreeNAS can be a challenge.

 

[TennesseeTony]

Make sure you get a fast as possible a dual core at least, one core will be fully utilized/needed 100% when folding. I tried with HT on and off, both times one core/thread was fully utilized (per GPU). Didn't really see an improvement with HT off, but due to the limited time I tested, and the highly variable score system of F@H it may not have been a valid test.

Also, for versatility, if you decided to run something else some day, the more cores/threads the better, even with just a single GPU. Milkyway for example, tasks run from a few seconds to a few minutes depending on your GPU and app_config file, so it's common to run 3-10 tasks per GPU, each with a tiny need for CPU. Other projects need quite a bit of CPU to go along with the GPU, but Folding is the only one I know of that 100% needs it's own core/thread.

I love my 980Ti's, but I will once again champion here for the 1070 (assuming you plan to purchase something), with similar or better performance, and 100w less power (TDP), and only a slight bump in price. For me this was the best deal due to shipping and taxes and brand/model preference.

 

[Markfw]

Since the 980ti refurb deals are now not good, and based on my utilization (see thread about generations) I second the 1070, or even (if you could afford it) the 1080. I got 1.75 million ppd with 2 1080's and only a total system draw of 425 watts.

 

[crashtech]

For now I found a good deal on a 1060 that will go into an office machine that is currently using only its IGP. It seems to be a decent deal. I am looking at ppd/$ more than ppd/watt in most cases for now. I know this is a somewhat short-sighted view, but our electricity is cheap and sourced predominately from hydro. I'd like to have well over 1M ppd online by Dec, we'll see.

Thanks for all the advice, it seems that some out there are rather light-lipped about it, competitive spirit I guess.

 

[TennesseeTony]

........when I get one or 2 of the 9208 or 9212 units, I have seen the ppd on one card as low as 300k ! That sure kills the total output.

What about the 9210s? 10 hour tasks and they drop the 1070 estimate down to 370k PPD.

Edit: Not sure I've seen a 9208 or 9212. I only look when I notice the PPD estimate to be less than 1.3M though. (pair of 1070's, for those just joining the conversation)

 

[Markfw]

Right now I have a 9208 on a 980ti and its doing 600k. I think the extra cuda cores helps it. and I have a 9210 on a 980 doing 256k. That card usually does 400-500k

I have a 9205 on my 290 (worst card I have) doing 283k, which for it is upper average. Its the 9212 and 9208 that are the worst.

Bottom line is that some units kill ppd worst on different cards, The 1080's hate them, and the 980ti's slow down on them.

edit: the 9210 is at 10:30 eta, and its 15% right now. So it looks like the 1070 does better than a 980 on this unit.

 

[TennesseeTony]

16.6% now with 8h8m ETA on the 1070.

Is it bad form to pause and find/delete the WU? Of course it is...unless during time of war/December.

Edit: Enable hidden folders: C:\Users\user name\AppData\Roaming\FAHClient\work\ then delete the whole offending folder (usually 00 or 01 or 02) in Windows, just in case anyone was wondering...

 

[Markfw]

Really hard to delete WU's I think I tried it once, and was too difficult. But your 1070's are doing better than my 980;s (at least in this one case). Maybe we should keep record of each type of card, and the PPD on each wu ? or too much work ? no payback ?

edit: as far as deleting bad WU's, with 11 cards I would be doing that all day every day

edit 2, I started a spreadsheet on tracking ppd by card and unit. Tony, send me actual unit and ppd from the FAHControl for your 1070;s

 

[TennesseeTony]

I'll be presenting the data in the following format: PRCG ##### followed by the Estimated PPD. I'll edit this post as new data is available. Generally speaking, the PPD is just a snapshot, but I'll try to ensure the task is at least 10% done, in hopes that the PPD has stabilized somewhat.

10496/658292
11707/668525

EDIT: Actually let's create a thread for this similar to what Assimilator1 does for MW.

 

[salvorhardin]

I'm in, I have the same username in f@h as I do here.

 

[GLeeM]

I'm in, GLeeM = GLeeM

 

[Markfw]

Guys, those that are joining in this race, since we are also competing with other teams, it would be good (I recommend) that you fire up your boxes 2 weeks in advance to work out any problems. I have spent the last 4 weeks doing so, but I have a few more boxes to tweak.

Lets kick some hiny !

 

[BrainEater]

Let's rock !!!!



If there is anything HWC can do to help , let us know !!

 

[PCTC2]

I'm still contemplating. I don't have any machines to borg at the moment so it won't be like my crazy PPD of old. If I had $20,000, I could probably do some awesome PPD with 32 GPUs and 2x bigADV clients running.

 

[Markfw]

I'm still contemplating. I don't have any machines to borg at the moment so it won't be like my crazy PPD of old. If I had $20,000, I could probably do some awesome PPD with 32 GPUs and 2x bigADV clients running.

I only have 11 GPU's, and I am only running one of my 3, 24 core CPU's, and its 6 million ppd ! And you could do it for only like $5000 or so

 

[Smoke]

http://folding.stanford.edu/home/

petrusbroder tells me that he will be back in position to handle the upcoming races by November 6th.

I've made a small addition to his initial post giving "new" folders info on getting started and joining our TeAm #198.

 

[Unolocogringo]

Skimmed over this thread a little.
The simple way to dump a work unit is to delete the slot and then re add it.
Works a charm and is fast.

Now on to more SERIOUS business.!!!
Trash talk time.
I wanted to come by and apologize in advance for kicking your butts in our upcoming race. We take our Folding VERY seriously.
I or should I say we will follow your RULES. ALL underhanded recruiting ,sandboxing,Polite jests etc... are to considered Fair Game.
Good Luck( like it will do you any good)

 

[Markfw]

Skimmed over this thread a little.
The simple way to dump a work unit is to delete the slot and then re add it.
Works a charm and is fast.

Now on to more SERIOUS business.!!!
Trash talk time.
I wanted to come by and apologize in advance for kicking your butts in our upcoming race. We take our Folding VERY seriously.
I or should I say we will follow your RULES. ALL underhanded recruiting ,sandboxing,Polite jests etc... are to considered Fair Game.
Good Luck( like it will do you any good)

so wat team are you referring to ? we have Tomshardware and the Canadians to race. and whats you username in F@H ?