The dog ate my 4P.

Both my machines are down I was thinking about making the g34 a starcraft server. shame i don't play starcraft..
 
 

I've been wondering how much of the total decline in active clients in the last six months or so is due to core 78 going EOL.  I believe projects that used that core stopped getting sent out in late August.  I'm not sure just how widespread this has been in the past, but it seems like over the years there have been a number of IT people and the like setting up FAH on large numbers of computers in their purview.  They were probably installing the uniprocessor 6.23 client, or something earlier, and likely haven't touched those installations in a while, if ever.  Since at this point the only way to get uniprocessor work is through core a4, and that requires at least client 6.34, all of those set-and-forget clients would be sitting idle, as would many people's old, repurposed hardware that they've had running uniprocessor work in a basement or something.
 
I don't have a good sense of just how many clients this would be, however.  Does tens of thousands sound unreasonably high?  I certainly don't think it's responsible for the entirety of the drop, and I'm not sure how that would figure into the declines of the last month or so.  If all clients are being counted as active within 50 days, that would correlate with machines being shut off in late November.  Perhaps the holidays had something to do with it, starting with Thanksgiving, but it seems like a lot of clients, and in that scenario they wouldn't have been turned on once the holidays were over, which seems a little unusual.

One problem with the end of uniprocessor theory is that, when I plotted Windows active cores/active CPUs, that figure has actually fallen a bit over the last couple of months.  It's currently sitting at around 1.81, and the highest it got was a hair over 2.0 in early August, when the total number of clients overall was near its most recent peak.  This also suggests there are still a lot of uniprocessor clients around (which is possible with 6.34 or newer, as many a4 projects can run on one core), or that there are issues with the stats reporting.
 
I've seen some speculation that the recent severe drop in reported active AMD GPUs (from around 5000 in early November to a low of 192 on Jan. 4, now at 365) might be the servers reporting them as Nvidia Fermi's.  If that's the case, then the sum of those two has ticked up after a lull during the holiday season.  I realize that there are some people who have moved to cryptocoin mining, but nearly all of them seems a bit unlikely.  Or maybe I'm just naive.  Does anyone know if there was some major happening in any cryptocurrencies in October or November that would explain this?  I know LiteCoin got a big surge of interest, but I'm not sure when that started.

Cutting the bottom out of the 4p community does nothing to address the SMP work not being done. It only serves to alienate an entire groups of people who invested a lot of good knowledge, coin and time in the name of science. They are not looking at the bigger picture. Flipping the switch on these machines only turns them off. It does nothing to increase SMP work being done and shrinks their willing donor pool rather significantly. 
 
Its a lot like saying we are having a walkathon, no a marathon, no a triatholon and then scratching your head wondering where everyone went.
 
 
 
 

Bill, which GPUs are you referring to?  The ones that I know of that were EOL'ed were the x1900's and the ATI 2000-4000 series.  
 
I wasn't involved in the folding community when the first happened, but from what I understand they were EOL'ed pretty much without any forewarning, and they were still pretty new to boot.  That's really bad, but they do seem to have learned from that; it seems like every EOL or major change since then has had at least a 2 months' heads-up.  There's obviously still debate about whether 2 months is enough lead time for a major change, or more on the order of 6 months, but I doubt they'll ever effect a change with no heads-up again.
 
Maybe I heard wrong, but I thought the x1900's deprecated because they could only run DirectX and DX couldn't reliably produce accurate results.  That's a situation I'm glad I was never in charge of.  It seems like the two options there are to upset everyone who owns one by ending it, or continuing to have them do garbage work.  The latter seems like it'd be more harmful to the project's reputation; even if people are really unhappy with donor relations, I don't recall anyone seriously questioning whether the work turned in was scientifically valid or useful.
 
Regarding the 2000-4000 cards, they were fairly old at the time (the 4000 series was around 4 years old when AMD core 11 finally ended), and none of them had the capability to run the new cores they wrote.  I think there were only a couple of projects left for them, and I doubt anyone at PG really wanted to keep using obsolete code; that's not a very attractive pitch to researchers interested in using your system.  I'm not sure this situation compares to the present dilemma as well as the x1900 one does.

Have one of my P4 BA Crunching PrmGrid at least for the next 120 Hours.
 

cokeman54
 (as of today) 193,288 others contribute to finding cures for some of life’s most threatening illnesses. This is the number in question and if you click on it, you get a breakdown. The important part is "(as of today)" this is the changing number that declines with each update. "IF" they count each core as an "other" then take into consideration the number of 4P's shut down or moved to other DC projects, then this decline makes sense since that announcement. This number has gone up and down in the past but not on a scale of the magnitude that it is now doing. It has got their attention but response was not what everyone wanted to hear. The thread at FF is now 41 pages and this thread is 18 pages, 4 weeks later, and no major changes from PG (except more blog posts +1).

Because the number tracking needs to continue.
 
192,909 10:35 PM CST 1/13/2014

Better yet let PG buy there own BA hardware and no more public help for BA.
Every donor then runs the same SMP and GPU projects. Then there's no need to
squabble about points difference between projects or what hardware gets what.
Problems solved.

The 5% is a very wide open and vague number. 
 
If they had more folders folding smp.. that 5% margin gets wider if they have less it gets thinner.
 
I don't know the solution that they would like, but in my book you want more SMP work to get done, the solutions pretty simple. Sweeten the pot for SMP. They worry about the numbers instead of making those numbers guide the direction of the work. The points are the tool to use to get the work done they want. I'll give a prime example as to why people don't like SMP work.
 
This is when i first started folding
 
http://forums.evga.com/FindPost/1898555
 

Re:Broken item in que? Tuesday, March 26, 2013 5:02 AM 

That one finally finished, got a massive 10k points from a 3 day project woohoo,,,   the new one (8082) seems much more better, 3.85% into it TPF less then 3 min PPD 15k and climbing
 
I am going to stop running them on this box after these and  do a fresh windows install, tweek the cpu/gpu to a mild CPU/GPU OC with a different drive in and see if it runs better over the weekend. I do use this one as my main box so its not folding 24/7.

On my 680 Rig I Fold the 3 680's and 8 Cores for SMP only to support the SMP backlog.
In my current 770 *Now Crunching after I get the Step-up completed I will do the same as my 680 Rig.
I don't really count the low PPD and Credit I get from any SMP.
If you look at Crunching you get only 90 - 100 credits on some projects that run for 1.5 to 2 hours each.
The one item I find out of place is running any SMP on a Server with 24 or more cores, and maybe even on a Server with 16 Cores.
 

Xavier Zepherious
I still think you need to break them up into two or three groups and have different point categories and leaderboards
then you have  a BA race , SMP race and GPU race this would entice any team or individual to do all 3 (or for some to specialize)

+1

When I get home, well you know I'm gonna be I'm gonna be the guy checking the count today.
192,271 1/14/2014 10AM CST.
 
If we go off his "5%" statement with this number that would mean he is looking for 9000+ Bigadv systems.  I call this a lie.  I want him to put the actual number of Bigadv systems they want.  I highly suspect it is VASTLY smaller than this.  Maybe dropping a zero.
 

I do wonder if the current slate of SMP work would be sufficiently conducive to satisfying one of the original reasons for the QRB.  For those who don't remember, or weren't around (I don't know who was or wasn't since I joined this team less than a year ago), PG was facing an issue where many SMP WUs wouldn't scale well with core count.  At the time, that meant that going from 2 to 4 cores didn't get you twice the PPD since the SMP core at the time couldn't use those 2 additional cores very well (4 cores was still somewhat rare).  People found that they got higher PPD from running two SMP clients concurrently, but PG wasn't particularly happy about that since it meant that each WU being processed in this manner was being returned in more time than if it were processed by itself.  
 
This was about when they started emphasizing that they prefer fewer WUs returned faster rather than more WUs returned slower.  In the years since then, however, they've put out SMP cores that scale a whole lot better, but not to bigadv numbers of cores .  My 4P has only done three regular SMP projects: 6098, 6099, and 7808.  The first two used core a3, while the latter used a4.  The first two got about 4.5x the base points vs. my 3770k running at 4.3 GHz, while the latter got 6x.  I think a nice dedicated 3770k box could have been put together when it was new for around $750 (no GPUs to make the comparison simpler; if my quick estimates are off feel free to put in your estimate).  The price of a 4650 4P has gone down since I bought mine; I paid around $4500, though I probably could have been a bit more frugal and done it for $4000, and right now you might be able to do it for $3500.  Throwing out my build price, you're looking at 4.6-5.3 times the cost for 4.5-6 times the points.  
 
I'm a little unclear on how much power that 3770k would be pulling from the wall on its own.  150W?  200W?  So the 4P would be drawing 4-5x the power.  If it's 4x, then maybe some people justify going that big with the power savings over 5 3770k boxes.  If it's 5x, then you're looking at pretty even scaling.  Would any of that be enough to motivate people to build an Intel 4P?  Maybe AMD 4Ps would look more attractive, but I don't have any SMP figures on those.
 
I also wonder if 2 32-thread SMP slots would fare better PPD-wise than 1 64-thread one.  If so, then at the top end you'd have the same problem they invented QRB to eliminate in the first place.  Maybe a better-scaling core would be the solution, though.

cokeman54
For those that may not have seen this: Top

---

Report this postReply with quote

Re: Change in BA requirements

by VijayPande » Tue Jan 14, 2014 8:17 pm

7im wrote:anyone claiming to know why the client count dropped without a 4 page explanation of all the variables just doesn't have a leg to stand on. Could be people don't like V7. Could be the crappy economy. Could be everyone is moving to mobile devices. Could be an alien invasion. Could be GPUs. Could be an AMD stats glitch. Could be the increasing cost of electricity. Could be Global warming, so they turned off their FAH heaters.

Regarding the client count: we have had a large company donating computer time anonymously and that donation time naturally ran its course (they don't care about BA, etc). That covers about ~30,000 CPUs or so. It's unfortunate timing that that ended around the new year, coincidentally with some of the rough server backend issues we had and the BA discussion.

I am hoping that this group will let us publicly acknowledge their contribution soon as what they've done (and the work we've been able to do on those machines) has been pretty exciting for us.
Prof. Vijay Pande, PhD
Departments of Chemistry, Structural Biology, and Computer Science
Chair, Biophysics
Director, Folding@home Distributed Computing Project
Stanford University


VijayPandePande Group Member Posts: 2561191,497 at 01/15/14  12:14 AM CST

Page1 in a very large font would simply say "7im"
Page2 again in large enough font a link to the f@h forums
Page3 more of the same but with a link to the blog subtitled "vague at best"
Page4 some nonsense copy pasted about global warming for comedic affect.

Done

I was going to post that myself in the FF but figured I'd just upset 7im with my thoughts.  I was also going to mention the major drop over the July 2013 numbers but then again... why bother anymore?

I think the point is Tex (at least from me) I'm pretty much done posting at FF.  No good comes of it and those FAH Fanatics tend to wear blinders so I'll just cease and desist and figure it all out when it affects me.  Right now I'm just shooting for a new WCG badge on the Mapping Cancer Markers.