How to break cores down after BIGADV

The most demanding Projects are only available to a small percentage of very high-end servers.

Moderators: Site Moderators, PandeGroup

Re: How to break cores down after BIGADV

Postby Grandpa_01 » Wed Dec 31, 2014 2:23 pm

The best thing to do would be to limit atom size before Jan 1, regardless of testing or not. The only thing the current scheme is going to accomplish is angering donors with large MP machines when the PPD drops from 500K to 75K PPD, On v6 I can get 360K running a3 smp on a rig that gets 990k PPD on bigadv on v7 that same rig will drop to 120k on the a4 WU's, The only thing that v7 current scheme / PPD will do is promote getting a % of the rigs shut down, or converting to v6 in order to avoid a4 WU's as much as possible.

I think if it were me I would try and be proactive rather than reactive (they already know the problem exist). :ewink:
Image
2 - SM H8QGi-F AMD 6xxx=112 cores @ 3.2 & 3.9Ghz
5 - SM X9QRI-f+ Intel 4650 = 320 cores @ 3.15Ghz
2 - I7 980X 4.4Ghz 2-GTX680
1 - 2700k 4.4Ghz GTX680
Total = 464 cores folding
User avatar
Grandpa_01
 
Posts: 1757
Joined: Wed Mar 04, 2009 7:36 am

Re: How to break cores down after BIGADV

Postby rhavern » Wed Dec 31, 2014 3:06 pm

Grandpa_01, I bet this request would be worth posting at https://www.reddit.com/r/foldingathome/ as well as here.
Folding since 1 WU=1 point
ImageImage
User avatar
rhavern
 
Posts: 484
Joined: Mon Dec 03, 2007 8:45 am
Location: UK

Re: How to break cores down after BIGADV

Postby bruce » Wed Dec 31, 2014 10:18 pm

Grandpa_01 wrote:The best thing to do would be to limit atom size before Jan 1, regardless of testing or not. ...


I'm not sure what you're saying. Atom size cannot be limited. Either they want to solve a small protein or a large one. Assignments can be restricted based on whatever information the project owner believes will work. They already do that. Projects with ~500 atoms don't seem to be assigned to anything above a Quad (or is it restricted to Duo or Uni?).

Should a protein with, say 33597 atoms be restricted to less that 32 cores or should it be restricted to 24 or to 16 or less? It depends on whether you believe the minimum number is 100 atoms per core or 150 or 200.
bruce
 
Posts: 22616
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: How to break cores down after BIGADV

Postby Joe_H » Wed Dec 31, 2014 10:23 pm

I also do not see any reason for a Jan 1 deadline, bigadv is scheduled to end at the end of the month on the 31st. It also is not clear if the current projects will end on that date, or just revert to point awards based on normal benchmarking without the current bigadv bonus. In any case, that is probably something best taken up on the subreddit.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Joe_H
Site Admin
 
Posts: 4533
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Re: How to break cores down after BIGADV

Postby 7im » Wed Dec 31, 2014 10:36 pm

From the News Blog: https://folding.stanford.edu/home/revised-plans-for-bigadv-ba-experiment/

2) The BA experiment will permanently end on January 31, 2015. On that date, the servers will be set to accept only and we will have no plans for future BA WUs. This would allow donors to continue to use their machines and recoup more of their investment...


Looks like the servers just stop assigning WUs at some time at the end of the month.
How to provide enough information to get helpful support
Tell me and I forget. Teach me and I remember. Involve me and I learn.
User avatar
7im
 
Posts: 14648
Joined: Thu Nov 29, 2007 4:30 pm
Location: Arizona

Re: How to break cores down after BIGADV

Postby Grandpa_01 » Thu Jan 01, 2015 12:06 am

bruce wrote:
Grandpa_01 wrote:The best thing to do would be to limit atom size before Jan 1, regardless of testing or not. ...


I'm not sure what you're saying. Atom size cannot be limited. Either they want to solve a small protein or a large one. Assignments can be restricted based on whatever information the project owner believes will work. They already do that. Projects with ~500 atoms don't seem to be assigned to anything above a Quad (or is it restricted to Duo or Uni?).

Should a protein with, say 33597 atoms be restricted to less that 32 cores or should it be restricted to 24 or to 16 or less? It depends on whether you believe the minimum number is 100 atoms per core or 150 or 200.


All of the a3 WU's appear to scale ok on larger MC rigs which are all or most all 77K atoms or larger most or all of the a4 WU's do not scale well at all on the larger MC rigs so I am saying if the server see's 48 cores or larger = 77K atoms or greater = a3 or however they want to set it up. :wink:

The individual folders can do that themselves by running Linux v6 of F@H since it currently rarely if ever receives a a4 WU
User avatar
Grandpa_01
 
Posts: 1757
Joined: Wed Mar 04, 2009 7:36 am

Re: How to break cores down after BIGADV

Postby bruce » Thu Jan 01, 2015 12:13 am

Grandpa_01 wrote:The individual folders can do that themselves by running Linux v6 of F@H since it currently rarely if ever receives a a4 WU


It sounds like the PG needs to make it work for V7 so the individual doesn't have to do it himself. If that isn't clear in the subreddit communications, it should be.
bruce
 
Posts: 22616
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: How to break cores down after BIGADV

Postby tiwake » Sat Jan 03, 2015 9:57 pm

I have a headless 64 core AMD system I built a while back, will be interesting to see what it does once the cutoff happens. After reading more stuff about this in general it seems like it shouldent be a problem but might be?
Image
User avatar
tiwake
 
Posts: 10
Joined: Sat Feb 04, 2012 12:59 am

Re: How to break cores down after BIGADV

Postby Nathan_P » Sat Jan 03, 2015 11:11 pm

It may only be a problem if you run the v7 client, even then you may still be OK. I haven't run any tests myself on v7 on Linux, all I do know is that my v6 Linux boxes (32/48 threads) only ever pick up core a3 work - which is better suited for our large core count rigs.
Image
Nathan_P
 
Posts: 1442
Joined: Wed Apr 01, 2009 9:22 pm
Location: Jersey, Channel islands

Re: How to break cores down after BIGADV

Postby Leonardo » Sat Jan 03, 2015 11:37 pm

I have a headless 64 core AMD system I built a while back, will be interesting to see what it does once the cutoff happens.
It's fairly simple to see what will happen. Regardless which client you run, you will need to set the client to download SMP work units. (Sorry if I've not understood your post well.) The SMP work units (non-Big Advanced), which I test on my systems occasionally, scale very well to engage all cores, which are 48 CPU cores per system. I'm assuming the efficiency and scaling, as the the cores heat up to the same level and the systems consume nearly the same power as when processing BigAdv work units.

Edit: I forgot to mention. The referenced systems are running client v6 in Linux.
Image
User avatar
Leonardo
 
Posts: 597
Joined: Tue Dec 04, 2007 5:09 am
Location: Eagle River, Alaska

Re: How to break cores down after BIGADV

Postby bruce » Sun Jan 04, 2015 12:56 am

So it sounds like everybody is saying
tiwake wrote:... it shouldent be a problem but might be.


I wish I knew why Core_A5 was needed at all. (...if Core_A3 works fine with high thread-counts). What changed?
bruce
 
Posts: 22616
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: How to break cores down after BIGADV

Postby Joe_H » Sun Jan 04, 2015 1:11 am

My recollection was that Core_A5 was further optimized for large, high thread-count processing of WU's. Bigadv did start on Core_A3 before the release of the A5 core, that change was what restricted bigadv to Linux after being possible on both Linux and OS X.
Joe_H
Site Admin
 
Posts: 4533
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Previous

Return to SMP with bigadv

Who is online

Users browsing this forum: No registered users and 1 guest

cron