• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

PrimeGrid challenge strategy thread (Comments? Suggestions? This *is* a democracy!)

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
While I've got you interested in the results of the year's first race, I'd like to discuss strategy for the rest of the year. :sneaky: Here are my thoughts on each race left:

15 March: Ides of March Challenge (PSP Sieve):
The single most important day of the year for the PrimeGrid challenge series. This year, single-day challenges more important than their length would indicate. If you crunch PrimeGrid only one day this year, get a 64-bit OS and crunch on this day! If you're stuck with a 32-bit OS, it looks like you can run 64-bit Ubuntu in VMWare!

22-30 April: Earth Day Challenge (321 LLR):
A long slog with long WUs. Each takes over a day. Let me know if you want me to remind you about these long slogs. Otherwise I might or might not do them myself, and will post stats only if I see someone else join too. The name is ironic, considering we're wasting so much electricity on finding primes.

16-21 June: PrimeGrid's Birthday/Summer Solstice Challenge (TRP Sieve):
Updated: A nice, calm sieve to kick off the summer. Very much like 321 Sieve below, but the WUs are a bit longer.

25 July: Full Moon Challenge (PPSE LLR):
The second-most important day of the year, and the second-best chance to find a prime. 64-bit OSes won't help here, but it's still a good chance to get ahead.

20-22 August: Dog Days of Summer Challenge (321 Sieve):
Alot like the Ides of March Challenge, but two days long, so less credit per day. 'Twould be nice if all the same people joined, for the duration, all with 64-bit OSes. :)

24 Sept-7 Oct: Calendula (PSP LLR):
A longer slog, if you can believe it, with longer WUs too. Same rules as with the Earth Day Challenge apply.

17 November: Leonids (SGS LLR):
The third-most important day of the year, and the third-best chance to find a prime. Almost identical to the Full Moon Challenge, but the WUs are a little longer and less likely to be prime.

18-21 December: Winter Solstice Challenge (PPSE Sieve):
This one's going to be interesting. And not just because I'm the main developer for the project. 32-bit OSes will get about 3/4 as much done as 64-bit OSes (assuming someone can compile BOINC on a 32-bit OS by then). Plus I may have one or more GPU apps by then. Hard to predict how this is going to go. But I'll keep you updated. :)

So my immediate goal is to get everyone who is running a 32-bit OS on a 64-bit processor to try installing a 64-bit OS, either directly, with Wubi, or in a VM.

But what do you all think? Are you more interested in multi-day-long WUs than I anticipated? Would you rather focus on 32-bit challenges? Let me know!
 
Last edited:

SlangNRox

Golden Member
Oct 9, 1999
1,010
1
76
I plan on participating in all of these as long as I'm around. I have no gpus that can do any work on ap26, but I do have wubi on my 64bit cpus.

Looks like a good anaylsis of all the races and I'm sure you'll keep us informed to when they are starting etc. Now we just need more participants.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
Thanks! (Both of you!)

That i5 on 64-bit should eat up the sieve WUs too. Nom nom nom! ;)
 

waffleironhead

Diamond Member
Aug 10, 2005
6,550
81
101
I will probably be along for all of them as well. I might not be pushing 24/7 during the summer ones, but I will be active.
 

VirtualLarry

No Lifer
Aug 25, 2001
51,900
6,871
126
seventeenorbust.com 's homepage said something about PrimeGrid and BIONC now being able to crunch SoB? I guess I don't get how PrimeGrid and SoB have combined now or something.

Also, can PrimeGrid crunch using GPUs? (Are they high-precision enough for prime crunching? I know some GPUs have non IEEE FPUs, although Fermi is supposed to be different in that regard.)

That might be something that would get me back into DC more, if I could repurpose my quad-gpu cruncher from doing F@H (currently doing nothing) to doing SoB via PrimeGrid.

I found this in the other primegrid thread:
There are. Some people want to do different projects. Some machines are better at some projects. For instance, 64-bit machines are better at sieve projects and AP26; and video cards can only do AP26 (for now).
So it looks like video cards can do some prime crunching after all?
 
Last edited:

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
seventeenorbust.com 's homepage said something about PrimeGrid and BIONC now being able to crunch SoB? I guess I don't get how PrimeGrid and SoB have combined now or something.
Basically, PrimeGrid had a client that could crunch SoB WUs; but BOINC can't check out work through the protocols SoB uses. They worked out a deal where PrimeGrid got a large, manually checked out range of WUs.

Also, can PrimeGrid crunch using GPUs?
Currently, as you discovered, PrimeGrid can only use GPUs on the AP26 application. Also, right now they can only use nVIDIA GPUs. (Edit: I edited the FP to include this.) I'm currently working on a sieve app for GPUs. Though it looks like no nVIDIA GPU will exceed the speed of a Core 2 Quad, I might be able to get it running on ATI eventually.

(Are they high-precision enough for prime crunching? I know some GPUs have non IEEE FPUs, although Fermi is supposed to be different in that regard.)
Interesting question. For my particular application, IEEE isn't high-precision enough! :eek: For my application, I need to fully represent up-to-62-bit numbers, and that takes an 80-bit floating point precision. So I'm using a slower algorithm that uses integers instead.
 
Last edited:

Rudy Toody

Diamond Member
Sep 30, 2006
4,194
387
126
I have been working on one badge: PPS sieve. Is this the same as the December challenge?

I will be there for the Ides of March challenge because I have only 64 bit OS on all rigs.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
I will be there for the Ides of March challenge because I have only 64 bit OS on all rigs.
Good to hear! This is looking like our biggest participation yet! :awe:

I have been working on one badge: PPS sieve. Is this the same as the December challenge?
You'd be surprised how difficult it is to answer this question.

Let's see...first of all, PPS and PPSE are both searching for Proth numbers, which are of the form K*2^N+1. Even K's are ignored because they're duplicates (6*2^4+1 == 3*2^5+1), and 1 has already been searched (they're called Fermat numbers.)

So, PPS is searching K=3-1199, while PPSE is searching K=1201-9999. My PPSE sieve is faster than what's being used on PPS now - but only for large ranges of K like PPSE uses. So there's been some interest in merging PPS into PPSE. If it were possible, it would make doing the PPS range basically free.

One problem. PPSE sieve has so far tested factors up to around 125 trillion. PPS sieve has already tested all factors up something like 7000 trillion. :eek:

Another problem. PPS is searching N's from 1 to 5 million. PPSE has only searched N's from 1 to 2 million. :\

What is going to happen, I don't know.
 

RobertE

Senior member
May 14, 2005
419
0
0
What will be the best way to not get work until the start of the challenge and to dump as much as possible before the deadline? The start and stop times are when I'm normally at work.

I'm thinking I can restrict network access until the start of the challenge, but what about at the end? Is boinc smart enough to know that it's going to not have net access after x time and try to dump stuff before then?
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
Is boinc smart enough to know that it's going to not have net access after x time and try to dump stuff before then?
I doubt it. I'd suggest setting your queue level low. Make sure that at any given time the WUs in your queue will complete at least one or two WU-lengths before the end of the race. (Sieve WU lengths are generally constant.) Once your queue has enough for almost the entire rest of the race, set the queue length to 0, and it will upload any complete WUs before fetching new ones.

Be aware that, once you set queue length to 0, BOINC probably won't communicate at all until the WUs remaining in the queue finish. But if you see you've gone overboard the first day, when you get home you can abort some un-started WUs.

Technically you could just leave queue length at 0 the whole time, but that could lose more work if the server has problems. The server's less likely to have problems with this challenge than the last one, but nothing's guaranteed.
 

RobertE

Senior member
May 14, 2005
419
0
0
Turned up the heat on the i5. Now going through the PSP WU's at just over 17 minutes each. As long as she stays stable, should be good to go for the challenge. :cool:
 

biodoc

Diamond Member
Dec 29, 2005
5,869
1,377
136
Turned up the heat on the i5. Now going through the PSP WU's at just over 17 minutes each. As long as she stays stable, should be good to go for the challenge. :cool:
17 minutes is great! My Q9950 @3.4GHz has a crunch time of just under 20 minutes. My Q6600s @3.0 GHz are doing 23 minutes.

Gotta get me one of those "i" rigs one of these days!:)
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
Yeah, sieves seem to produce less heat than LLR challenges in general, so I can overclock my Q9400 to 3.2GHz instead of just 3. :)

That's also true of AP26 - though not of the December challenge.
 

Rudy Toody

Diamond Member
Sep 30, 2006
4,194
387
126
Ken_g6, do you have a bash script that would start PrimeGrid at the right time for the challenge?

And perhaps, another one to force an upload at the deadline?
 

RobertE

Senior member
May 14, 2005
419
0
0
Ken_g6, do you have a bash script that would start PrimeGrid at the right time for the challenge?

And perhaps, another one to force an upload at the deadline?
For the start, I'm going to drain my cache down and suspend network communications until the right time. As for the end, I'm hoping I can get home in time to force the uploads unless someone knows a better way.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
15,483
2,528
55
As far as uploads go, it's probably best to set several at jobs to run boinccmd. Both at and boinccmd exist on both Linux and Windows, but at is run differently on each.

On Linux, if you were in UTC:
echo "boinccmd --project http://www.primegrid.com/ update" | at 17:00 16.03.10
echo "boinccmd --project http://www.primegrid.com/ update" | at 17:30 16.03.10
echo "boinccmd --project http://www.primegrid.com/ update" | at 17:45 16.03.10
echo "boinccmd --project http://www.primegrid.com/ update" | at 17:55 16.03.10
echo "boinccmd --project http://www.primegrid.com/ update" | at 17:59 16.03.10

On Windows, if boinccmd.exe is in "c:\program files\boinc", run:
at 17:00 /next:Tuesday c:\progra~1\boinc\boinccmd.exe --project http://www.primegrid.com/ update
at 17:30 /next:Tuesday c:\progra~1\boinc\boinccmd.exe --project http://www.primegrid.com/ update
at 17:45 /next:Tuesday c:\progra~1\boinc\boinccmd.exe --project http://www.primegrid.com/ update
at 17:55 /next:Tuesday c:\progra~1\boinc\boinccmd.exe --project http://www.primegrid.com/ update
at 17:59 /next:Tuesday c:\progra~1\boinc\boinccmd.exe --project http://www.primegrid.com/ update
 
Last edited:

RobertE

Senior member
May 14, 2005
419
0
0
Figured out how to get windows task scheduler to call boinccmd as ken posted, so all go to go at the end of the challenge.

Doing some quick math here.
i5 plows through a WU in roughly 1050 seconds.
86400 seconds in a 24 hour period
82 WUs per core over that 24 hour period = 328 wu total
29.56 credits per wu = ~9695 credits for 24 hours of the challenge.

So, bring it on :cool:

Now do I keep her at 3.6ghz or go for more :sneaky:
 
Last edited:

RobertE

Senior member
May 14, 2005
419
0
0
Ok, here's my plan, does this seem like it will work?

Currently running 100% Einstein with a little cache. Going to set it to no new work before going to bed. In the morning before leaving for work, I'm going to suspend Einstein, flush whats done and suspened network activity until a minute or two after the challenge starts.

At the close, I've got windows task scheduler set to run Ken's script above every 10 minutes starting at 1 pm my time (1 hour before challenge closes).

I think that should work, any thoughts or changes I should make?
 

ASK THE COMMUNITY