PrimeGrid Races 2017

[StefanR5R]

OK, 2133 c15 and 2400 c17 is virtually the same latency, and not much bandwidth difference. I may have a penalty due to registered DIMMs, if you have unregistered DIMMs.

OTOH, it's probably just WU variability.
16,400 points / 3h30m = 112,000 PPD
12,100 points / 2h50m = 102,000 PPD

 

[TennesseeTony]

......And Tony, I saw your post on overclocking and AVX. I have it overclocked to 3850, and its running 60c, same as before.

Go try it on one of your OC'd i7's.

 

[StefanR5R]

Fresh from my box with -t 7 (two workers per CPU):
16,500 points / 6h20m = 63,000 PPD ( x 2 = 126,000 PPD per CPU)

I guess I switch the -t 14 box back to -t 7 for a while, until /edit/ half a day before the end of the race.

 

[Markfw]

OK, 2133 c15 and 2400 c17 is virtually the same latency, and not much bandwidth difference. I may have a penalty due to registered DIMMs, if you have unregistered DIMMs.

OTOH, it's probably just WU variability.
16,400 points / 3h30m = 112,000 PPD
12,100 points / 2h50m = 102,000 PPD

My current unit is going to take 4 1/2 hours ATM.

No registered dimms or ECC

 

[crashtech]

I'm seeing some very contradictory results; the machine I thought had a problem is my top producer according to primegrid.com. The only explanation is that there is significant variability between different PSP WUs.

 

[TennesseeTony]

Finally got off my butt and turned off HT and set the config to run on all 14 cores, and it looks like the few tasks that have completed for me were in the 4.25 to 4.5 hour range.

I can't find any 16,200sec tasks though, in either pending or valid. It's been 45minutes.......shouldn't they show nearly instantly?

 

[Markfw]

Finally got off my butt and turned off HT and set the config to run on all 14 cores, and it looks like the few tasks that have completed for me were in the 4.25 to 4.5 hour range.

I can't find any 16,200sec tasks though, in either pending or valid. It's been 45minutes.......shouldn't they show nearly instantly?

Well, I am the new guy, but it looks to me like all you see is the one "super task" thats hours long At least thats all I see.

 

[Ken g6]

Finally got off my butt and turned off HT and set the config to run on all 14 cores, and it looks like the few tasks that have completed for me were in the 4.25 to 4.5 hour range.

I can't find any 16,200sec tasks though, in either pending or valid. It's been 45minutes.......shouldn't they show nearly instantly?

Timing results with this app_config file can be weird. If you started a task without the config file it seems to show the full CPU runtime as elapsed time. Tasks started after changing the config file should show up right, I think.

 

[StefanR5R]

I'm seeing some very contradictory results; the machine I thought had a problem is my top producer according to primegrid.com. The only explanation is that there is significant variability between different PSP WUs.

I have been mislead too all the time since I began testing PSP-LLR before the race, and well into the race. One cannot go by number of completed tasks, because tasks are variable. One cannot go by reported Run Time, because the run times listed at primegrid's web site are random garbage if -t > 1.

I think the only somewhat reliable measures are:

"pseudo run time" ~= "reported CPU time" / "thread count of the worker"
"yield of a WU" = "points of the WU once validated" / "pseudo run time"​

One cannot go by the RAC of each machine either, because one machine may have much more luck than the other in getting its pending WUs validated sooner than later. Validation is instantly if you are the wingman, or in weeks if wingman after wingman aborts or errors out or suspends.

Finally got off my butt and turned off HT and set the config to run on all 14 cores, and it looks like the few tasks that have completed for me were in the 4.25 to 4.5 hour range.

I can't find any 16,200sec tasks though, in either pending or valid. It's been 45minutes.......shouldn't they show nearly instantly?

They show up near instantly. But they show up with totally made-up Run Time. (Occasionally, they show up with correct run time.)

You may be able to identify WUs by the date & time when their completion was reported.

Timing results with this app_config file can be weird. If you started a task without the config file it seems to show the full CPU runtime as elapsed time. Tasks started after changing the config file should show up right, I think.

(edit)
After app_config was changed, tasks show up correct in BOINC manager's UI (a) if they were downloaded after the change, or/and (b) if BOINC was restarted after the change.

They will show up correctly at primegrid's web site, except for mostly incorrect run time. And this is entirely independent of when the change to app_config was made, i.e., even if app_config was not changed during the entire runtime of both the boinc client and the task.

 

[Ken g6]

Day 12.05 stats:

Rank___Credits____Username
9______6247716____xii5ku
53_____1293886___10esseeTony
157____424750_____Ken_g6
249____210771_____Orange Kid
270____177364_____zzuupp
304____127307_____SlangNRox
320____114337_____markfw
352____94000______GLeeM
392____67401______crashtech

Rank__Credits____Team
4_____31845633___Sicituradastra.
5_____14563201___Crunching@EVGA
6_____11575300___BOINC@Poland
7_____8757534____TeAm AnandTech
8_____7583742____Team 2ch
9_____7455260____BOINC@MIXI
10____6978907____Rechenkraft.net

Still looking like a good 7th place finish.

 

[crashtech]

Do you guys even need me now? I think I am too little too late. Maybe I'll leave my top two performers in, some of the rest aren't doing well at PrimeGrid at all.

 

[StefanR5R]

@crashtech, in a race, nothing is too little, too late, right? Well, PrimeGrid LLR races are a bit different though. I had an Ivy Bridge-E in it except for the duration of the Formula Boinc sprint. I have taken this IVB-E out of the PrimeGrid challenge now because the heat/noise/stability issues were too severe, considering its comparably poor yield in PSP-LLR.

So I'd personally say, any core which isn't Haswell or later (if Intel; in case of AMD: which isn't Ryzen), and/or isn't top-notch cooled (if Intel), and/or is desirable to have in another project that is important to one personally (be it in a CPU project, or as a GPU feeder), can surely be excused.

It also doesn't feel like MOABs are waiting to be dropped on us in this race.

And lastly, the next two races (Formula, Pentathlon) are right around the corner.

 

[crashtech]

I got the better cooler on the 4590 last night, so it's very happy with all four cores doing PrimeGrid. And one of the Westmere boxen that has no GPU seems to be doing satisfactorily, believe it or not. Seems turning off HT and putting all twelve cores to work on one instance does well enough, albeit rather inefficiently. Those two will be my representation. I thought the Skylake would be good, but it's apparently massively crippled by the dead memory channel.

 

[StefanR5R]

@Ken g6, thanks for the stats.

Spoiler: user graph

I still haven't added BOINC@Poland to the graphs. They are holding their 950 k PPD.

Spoiler: team graphs

 

[TennesseeTony]

Thanks for the stats, Ken, and graphs, Stefan. After the last FB Sprint, I had dropped to 136th place. Not too bad of a comeback, but moving up from here is getting slower and slower. Hoping for a top 50.

 

[StefanR5R]

If the dozen guys before you with ~2 M points currently were at it constantly during these 12 days, that'd be 170 k PPD, against your current 420 k (250 k more per day). That should be a top 40, right? (An effective top 10 in my eyes though, with what you did at Formula.)

 

[StefanR5R]

Tasks tended to get longer, and continue to do so. Credit per task increases proportionally: 10 k, 16 k, 23 k.
[Michael Goetz, April 17], [Michael Goetz, April 19]

 

[TennesseeTony]

Certainly good to know why after changing to 14 non-HT cores, that my task times were getting longer compared to the first few that finished in 4.5 hours. Here is a 23k point task, E5-2683v3, 14 core MT sample from WUSS. It shows 8 hours run time, and I believe that to be accurate:

 

[Ken g6]

Day 13.08 stats:

Rank___Credits____Username
9______6882296____xii5ku
48_____1689291___10esseeTony
152____487726_____Ken_g6
233____243797_____Orange Kid
242____229338_____markfw
280____181844_____crashtech
281____177364_____zzuupp
285____176162_____SlangNRox
286____175967_____GLeeM

Rank__Credits____Team
4_____34408983___Sicituradastra.
5_____16014450___Crunching@EVGA
6_____12393056___BOINC@Poland
7_____10243788___TeAm AnandTech
8_____8250710____Team 2ch
9_____8081591____BOINC@MIXI
10____7661377____Rechenkraft.net

I'm not sure if we can catch Poland, but no one's catching us.

 

[StefanR5R]

Spoiler: user graph

I'm not sure if we can catch Poland, but no one's catching us.

Spoiler: team graphs

Poland's PPD are down to 0.79 M. If it stays there the next two days, we would need 0.79 + 2.15 / 2 = 1.87 M PPD to catch them. Don't we?

 

[crashtech]

At least we'll have them eyeing their rear-view mirror, so to speak!

 

[Ken g6]

Day 14.06 stats:

Rank___Credits____Username
7______7539288____xii5ku
43_____2093276___10esseeTony
156____511017_____Ken_g6
201____332122_____markfw
223____291804_____Orange Kid
233____263611_____crashtech
269____208926_____GLeeM
281____193190_____zzuupp
301____176162_____SlangNRox

Rank__Credits____Team
4_____36610819___Sicituradastra.
5_____17346324___Crunching@EVGA
6_____13391860___BOINC@Poland
7_____11609400___TeAm AnandTech
8_____8883819____Team 2ch
9_____8741972____BOINC@MIXI
10____8372086____Rechenkraft.net

Looks like we won't catch Poland, but Mark won't catch me either.

Some of you haven't been in many PrimeGrid races, so the "usual end-of-race announcment" may be new to you:

At the Conclusion of the Challenge

We would prefer users "moving on" to finish those tasks they have downloaded, if not then please ABORT the WU's (and then UPDATE the PrimeGrid project) instead of DETACHING, RESETTING, or PAUSING.

ABORTING WU's allows them to be recycled immediately; thus a much faster "clean up" to the end of a Challenge. DETACHING, RESETTING, and PAUSING WU's causes them to remain in limbo until they EXPIRE. Therefore, we must wait until WU's expire to send them out to be completed. Since these tasks have a 21 day deadline, this causes up to a three week delay in determining the final challenge standings. Your help is greatly appreciated. Thank you!

 

[Orange Kid]

Thanks for the updates.
Looks like I'll only get two more done before the end

 

[StefanR5R]

I already got comfortable at individuals' rank 9. But suddenly I am finding me in a tie with Poland and ad astra.
They and I will probably continue to take turns at ranks 7-9 until the last minute.

Spoiler: graphs again

 

[Markfw]

If I had time to reconfigure the other 5 CPU boxes, I could have made a bigger dent. Thats 2 boxes, and 22 cores, thats 332, 000 in 2 days ! (or was it 3 for me)