PrimeGrid Races 2017

[GLeeM]

Thanks for the stats, Ken

 

[Orange Kid]

Thanks for the stats updates
We'll get'em next one!

 

[StefanR5R]

Must I be the first one to point out that our team rank is prime?

As promised, xii5ku is being consigned to community service at WCG now. I sentenced him to 10 days crunching OpenZika on all of the nodes which he used during the PG race.

 

[geecee]

As always, thanks Ken for organizing and updating us.

Surprisingly, the old X6 did ok crunching PG. Just uses a lot of power.

 

[crashtech]

I missed this one, but come March 10th I'll have 12 Westmere cores ready to add to our team totals.

 

[StefanR5R]

I'll see if I can produce as many WCG OpenZika credits as TennesseeTony's GCW Sieve credits from the Newton Challenge (converted 7:1 of course).

Done. Total from Sat, Jan 15 to Fri, Jan 27: 14,723,857 WCG points = 2,103,408 BOINC credits

 

[StefanR5R]

Here's the (tentative) list of all races for this year:

[ IMG ]

I get a 404 instead of the image. Here is the list in ASCII.

    Date             Time UTC   Project(s)   Challenge                               Duration
---------------------------------------------------------------------------------------------
1   3-13 January     18:00:00   GCW-Sieve    Isaac Newton's Birthday Challenge       10 days
2   10-25 March *)   12:00:00   SoB-LLR      Year of the Fire Rooster Challenge      15 days
3   7-22 April       12:00:00   PSP-LLR      Mathematics Awareness Month Challenge   15 days
4   12-13 June       00:00:00   SGS-LLR      PrimeGrid's Birthday Challenge           1 day
5   20-23 August     18:00:00   GCW-LLR      Solar Eclipse Challenge                  3 days
6   3-8 September    18:00:00   321-LLR      Number Theory Week Challenge             5 days
7   18-23 October    00:00:00   TRP-LLR      Diwali/Deepavali Challenge               5 days
8   17-20 November   12:00:00   GFN-15       Pierre de Fermat's Birthday Challenge    3 days
                                GFN-16
                                GFN-17-Low
9   18-21 December   16:28:00   PPS-Sieve    Winter Solstice Challenge                3 days

Edit, March 6:
*) Races # 2 and 3 postponed, as noted by Ken g6 below
Edit, April 2:
Race # 3 is a go, as originally scheduled.

 

[Ken g6]

In case you were wondering, PrimeGrid races are indefinitely delayed. The next race will not start on March 10. April is likely also out. It's hard to say beyond that.

Maybe we should try FormulaBOINC racing instead?

 

[Orange Kid]

Well, that is really the only time I run PrimeGrid
I guess I'll focus more on the FB 3day sprints. It breaks up the monotony of just running a project here and there. Gives a little excitement. Guess I need a life

 

[Ken g6]

Time to bump this thread for the next race, in less than a week.

 

[StefanR5R]

Interesting: LLR can apparently be run in multithreaded mode.

• race announcement
• multithreading how-to
• Core i5/i7 Haswell/Broadwell/Skylake PSP-LLR results:
  • Haswell Core i5: 4 single-thread jobs vs. 1 quad-thread job: throughput ratio 1:1.015
    
  • Broadwell Core i5: 4 single-thread jobs vs. 1 quad-thread job: throughput ratio 1:0.94
    
  • Skylake Core i7: 4 single-thread jobs vs. 1 quad-thread job: throughput ratio 1:1.17
    A Skylake Core i5 scales similarly.
• Xeon 2683v3 (Haswell-EP 14 cores) PSP-LLR results:
  • 2 seven-thread jobs vs. one 14-threads job(?): throughput ratio 1:0.94 in terms of CPU time
  • throughput ratio in terms of real time not measured due to reporting difficulties
  • throughput with single-threaded tasks not posted?

It's a trade-off between multi-threading overhead and multi-processing overhead.

Edit: I posted 2P Broadwell-EP results in the PrimeGrid CPU thread.

 

[StefanR5R]

Engaged two humble 4-core Haswells with PSP-LLR now.

Against logic, I decided to leave the Broadwell-EPs at Einstein for a little more.

 

[Orange Kid]

Got a few cores going.

 

[Ken g6]

This one was an early start for me, but I got a few cores going too.

 

[Kiska]

Unfortunately I don't think I'll be able to contribute any cores for this race, since its all on Einstein right now

 

[Ken g6]

Unfortunately I don't think I'll be able to contribute any cores for this race, since its all on Einstein right now

This is a 15-day race. The current FormulaBOINC race will end before this one does. (So will the next FormulaBOINC race.)

 

[StefanR5R]

OK, 4x14 cores went from Einstein to Sierpiński.

Edit:
Also, a 10-core CPU which is currently carrying 6 GPU feeder tasks for Einstein, is now running PSP-LLR on the remaining 4 cores like so:

<app_config>
   <app>
      <name>llrPSP</name>
      <max_concurrent>2</max_concurrent>
      <fraction_done_exact/>
   </app>
   <app_version>
      <app_name>llrPSP</app_name>
      <cmdline>-t 2</cmdline>
      <avg_ncpus>2</avg_ncpus>
      <max_ncpus>2</max_ncpus>
   </app_version>
</app_config>

 

[Orange Kid]

One day down and only two to go till I finish some WU's. These things are as slow as molasses.

 

[Ken g6]

Day 1.2 stats:

Rank___Credits____Username
11_____169145_____xii5ku

Rank__Credits____Team
5_____570957_____BOINC@Poland
6_____398897_____Czech National Team
7_____204596_____The Knights Who Say Ni!
8_____169145_____TeAm AnandTech
9_____157951_____US Navy
10____67306______Special: Off-Topic
11____65936______PrimeSearchTeam

One day down and only two to go till I finish some WU's. These things are as slow as molasses.

And that's why it's a 15-day race. That's also one reason why @StefanR5R's multithreading post is useful. Multithreading may be easier than juggling.

 

[StefanR5R]

Multithreading is also the major (if not only) reason why there are more than 0 credits in the day 1.2 stats.

By the way, these WUs are not only looong, their runtimes are also untypically variable for PrimeGrid's standards. They seem to be bimodally distributed. This confused me already during the past week when I tested PSP-LLR multicore scaling.

 

[TennesseeTony]

I'll be joining later, so be ye not dismayed at my absence.

 

[Ken g6]

Day 2.125 stats:

Rank___Credits____Username
7______697292_____xii5ku
142____22926______Ken_g6

Rank__Credits____Team
4_____1969519____Czech National Team
5_____1589216____BOINC@Poland
6_____1187362____BOINC@MIXI
7_____720219_____TeAm AnandTech
8_____641068_____US Navy
9_____636052_____The Knights Who Say Ni!
10____623846_____Team 2ch

Everybody but Stefan and me using slow computers and not multithreading?

By the way, I turned on multithreading on my laptop yesterday, and it only ran one existing WU at a time. I had to change the config file to turn it back off temporarily.

 

[Orange Kid]

Everybody but Stefan and me using slow computers and not multithreading?

Yes.
Can I switch to multicore in the middle of a WU?
Is this for real cores or can virtual cores be included? IE: my I5 as two cores or four?

 

[Ken g6]

Yes.
Can I switch to multicore in the middle of a WU?
Is this for real cores or can virtual cores be included? IE: my I5 as two cores or four?

Switching in the middle of a WU seems to finish it on only one core. HT doesn't help either.

 

[StefanR5R]

By the way, I turned on multithreading on my laptop yesterday, and it only ran one existing WU at a time. I had to change the config file to turn it back off temporarily.

I am not sure anymore, but perhaps the same (under-utitilized host) happened to me during my experiments with this feature on the Xeons before the challenge.

Right now all is fine:

4 single-thread tasks on a 4C CPU,
2 dual-thread tasks on another 4C,
2 dual-thread tasks plus 6 e@h feeders on 10C (PrimeGrid forced to 2 tasks max as posted above, by means of the max_concurrent tag),
1 four-thread task plus 3 e@h feeders on 6C/HT (PrimeGrid forced to 1 task max),
14 dual-thread tasks on 2x14C,
4 seven-thread tasks on 2x14C.​

I will revisit how the two 2x14s do in terms of PPD after one or two more days and reconfigure as indicated. (One 4C and the 6C will have to leave PrimeGrid during the week due to their noise.)

Note, omit max_concurrent in the app_config.xml if you do not want to specify a CPU limit in this way. (I guess Ken g6 is aware of that, just mentioning it for others who, like myself, do not deal with app_config.xml's day in, day out...)

Can I switch to multicore in the middle of a WU?

I switched once in the following way:

• Had four single-threaded tasks running on a 4C,
• wrote C:\ProgramData\BOINC\projects\www.primegrid.com\app_config.xml,
• exited boincmgr and had it shut down the tasks at this point,
• restarted boincmgr.

This resulted in two of the four tasks continuing from their previous progress percentage, but now running dual-threaded instead of single threaded. And the other two tasks sat there waiting for their turn, and they were also continued dual-threaded without losing previous progress. To be sure, I later looked these tasks up at the PrimeGrid web site, and they were marked valid (3) or pending (1), not invalid or error.

Is this for real cores or can virtual cores be included?

I tested hyperthreading only with single-threaded tasks, and it was detrimental even on Linux (which I assume to deal better with hyperthreading than Windows). My guess is that it is detrimental with multi-threaded PSP-LLR tasks too. (edit) Unfortunately, to prove or disprove it by measurement takes a long time. But from what I read elsewhere, HT is generally discouraged with LLR.

IE: my I5 as two cores or four?

(edit) That would be two cores on i5-6200U, 2 single-threaded workers or 1 dual-threaded worker with HT off, or with HT on with total utilization forced to 50 %. Or double as many with HT on at 100 %, but possibly performing worse overall.

I switched hyperthreading off in the BIOS of all my machines, except in the two Einstein@Home feeders. (Which will soon leave E@H for other endeavors.)