FormulaBOINC

[Ken g6]

Where do you find that info? Cause I can't on FB's thing

Oh, I finally figured it out. That's the overall Marathon stats (League 2), not the Sprint stats.

 

[Orange Kid]

Yes those are the Marathon stats, I figured theSsprint stats are what they are and always there. The best part is the points you get in the sprint at the end is added to you total in the marathon and stays till the end of the year.
I thought it would be interesting to see how much we might move up overall in our league on each project.

 

[Kiska]

Oh, I finally figured it out. That's the overall Marathon stats (League 2), not the Sprint stats.

Oh.......
WTF!

So I've done a simple analysis of my own results: https://docs.google.com/spreadsheets/d/1vjSdPk-RaCdR3V3jVE6wzfrGhKQsciEZmXc9Sho_B7I/edit?usp=sharing in the form of a really large table

 

[crashtech]

Ugh, we are having a power outage!

 

[waffleironhead]

Here are the sprint stats.
http://formula-boinc.org/sprint.py?sprint=1&lang=en&year=2017
Sprint 1, league 2

 

[StefanR5R]

Also my new build is coming together nicely, just got my $500 motherboard

My 2P boards are similarly laid out. I have them in midi tower cases with perforated top; CPU tower coolers oriented to blow upwards. Avoids blowing the exhaust of one CPU onto the cooler of the other. But this top exhaust works so nicely only because (a) there is no GPGPU in these servers which would heat both CPUs from below, and (b) there is no desk or other obstruction above the cases which could cause a short circuit from the cases' top exhaust to their front and back intakes.

@StefanR5R You can analyse my host results
https://docs.google.com/spreadsheets/d/1vjSdPk-RaCdR3V3jVE6wzfrGhKQsciEZmXc9Sho_B7I/edit?usp=sharing

But you already analyzed the heck out of it yourself... Well, except for the yield; let's see --
E3-1230v3:
ECM ....... 18.5 hours (27 %) ......... 1810 credits (41 %) ......... 98 credits/hour
OGR ....... 49.0 hours (73 %) ......... 2640 credits (59 %) ......... 54 credits/hour
total ....... 67.5 hours (100 %) ....... 4450 credits (100 %) ....... 66 credits/hour

Good that your X3323 only worked at ECM so far (40.6 hours, 3043 credits, 75 credits/hour), and not on OGR.

Ugh, we are having a power outage!

Argh!

 

[Orange Kid]

The Sprint Stats

YOYO@Home Marathon Top 10

The Start
1 Overclock.net................... 25 ........... 9,573,188
2 XtremeSystems ............... 18 ............ 2,189,004
3 AMD Users ...................... 15 ............ 1,751,334
4 Ars Technica ........ ........... 12 ........... 1,685,996
5 meisterkuehler.de Team .. 10 ........... 1,134,844
6 The Scottish Boinc Team .. 8 ............... 859,399
7 team.se ............................. 6 ................ 627,370
8 TeAm AnandTech ............. 4 ................ 621,026
9 Sicituradastra. ................... 2 ................ 556,376
10 Team Norway ................. 1 ................. 510,937

Day One
1 Overclock.net................. ..25 ..............10,432,940
2 XtremeSystems ...............18 ...............2,189,004
3 AMD Users ......................15 ...............1,776,897
4 Ars Technica ...................12 ...............1,726,447
5 TeAm AnandTech ...........10 ...............1,648,561
6 meisterkuehler.de Team ..8 .................1,444,511
7 The Scottish Boinc Team .6 ..................861,159
8 team.se .............................4 .................642,232
9 Sicituradastra. ...................2 .................557,696
10 Team Norway .................1 ..................530,117

Day Two
1 Overclock.net ..................25 ................10,942,116
2 TeAm AnandTech ...........18 .................2,369,665
3 XtremeSystems ...............15 .................2,189,004
4 AMD Users ......................12 .................1,782,530
5 Ars Technica ...................10 .................1,745,222
6 meisterkuehler.de Team ...8 .................1,740,835
7 OcUK - Overclockers UK ..6 .................1,060,864
8 The Scottish Boinc Team ..4 ...................862,039
9 team.se ..............................2 ..................647,011
10 Sicituradastra. .................1 ...................558,025

I don't see us getting to number one but should have a comfortable number two.
OcUK - Overclockers UK have moved from 20th at the start to 7th

 

[TennesseeTony]

23 hours remaining in the race! It's been exciting so far!

Thanks for the stats, Orange Kid!

 

[StefanR5R]

23 hours remaining in the race!

This reminded me to have another look at the task queue of each of my PCs and sort out some pending tasks with a suspiciously long runtime. But it's slightly unnerving that the estimated runtime is so unreliable.

Thanks for the stats, Orange Kid!

Indeed!

 

[Kiska]

This reminded me to have another look at the task queue of each of my PCs and sort out some pending tasks with a suspiciously long runtime.

Now my server has approx. 350k lines in the event logs!
Also I'll update my spreadsheet in a bit

 

[StefanR5R]

Related to my difficulty to bunker properly before the race, and to my worries about estimated time to completion: From what I just read on the internet, I figure that the BOINC scheduler continuously monitors estimated and actual time to completion, and maintains an according duration_correction_factor in client_state.xml on a per-project basis (but not per subproject or by application). The factor can also be looked up at the project's web site, account page, "computers on this account", at each computer summary page.

Of course when I started with Yoyo entirely fresh, my first task downloads had to go on with some guesswork, as there hasn't been a duration_correction_factor calculated yet. On the other hand, I suspect that Yoyo sometimes sets extremely misleading estimated task runtimes, with the result that BOINC scheduler extrapolates a less than helpful duration_correction_factor. I currently got the following DCFs (rounded):

i7-4960X oc ....... 0.6 (This one is loaded only partially, so maybe this factor is OK.)
i7-4900MQ ........ 6.6
E3-1245v3 ......... 1.1
i7-6950X oc ....... 1.9
2x E5-2690v4 .... 1.2
2x E5-2690v4 .... 1.3

 

[crashtech]

Everything I am getting from yoyo says it's going to take 2+ days to complete at this point, which isn't very helpful.

 

[Ken g6]

Everything I am getting from yoyo says it's going to take 2+ days to complete at this point, which isn't very helpful.

Me too. Looks like I'm getting tasks with big points, but even bigger compute times, like this. Maybe if you cut down the number of cores being used it'll run faster?

 

[crashtech]

Me too. Looks like I'm getting tasks with big points, but even bigger compute times, like this. Maybe if you cut down the number of cores being used it'll run faster?

It would be interesting to know if the tasks are strictly single threaded. If so, the only thing that would speed them up is more clockspeed or maybe disabling HT.

 

[StefanR5R]

Everything I am getting from yoyo says it's going to take 2+ days to complete at this point, which isn't very helpful.

Here it got this extreme only on my laptop yet. All newly fetched tasks had this alleged 2+ days estimated runtime. (Probably would take a lot less, but perhaps not until before the end of the sprint.)

So I rebooted into HT-off, fetched "2+" days long ECM tasks again. Aborted and switched to OGR, got "2+" days OGR tasks. Switched to Muon and got lucky to fetch 8 tasks which allegedly take 4, 7, or 10 hours. Only a few minutes into these tasks, their progress bar jumped to 66.666 % or 33.333 % respectively and is now stuck there.

I sure hope that my two 2P boxes remain somewhat productive before the end of the sprint. Both are currently stuck on ecm_ru_* tasks with 10....17 hours elapsed runtime and alleged 40.000...60.000 % progress. Beyond these, they still have got a few ecm_ru_* tasks in the queue with "estimated" 3 hours runtime.

It would be interesting to know if the tasks are strictly single threaded. If so, the only thing that would speed them up is more clockspeed or maybe disabling HT.

Windows 7 task manager shows ECM and Muon as single-threaded applications. Edit: i.e. in the task table, with "Threads" column added to the view.

I too think that disabling HT (on Windows) or simply running on only 50 % of the logical CPUs (on Linux) would help reduce runtime per task.

On my laptop, further reducing the # CPUs used helps too, because its processor clock speed is governed by cooling capacity.

 

[crashtech]

I won't be able to disable HT on several of the machines, since they are headless, but I might play with the ones that have easy access to a monitor!

 

[StefanR5R]

If they are running Windows, then simply running at 50 % of logical CPUs works too to avoid hyperthreading, to a degree. AFAIU it depends on the Windows version and especially on the power management scheme how effective this is. Windows process scheduler is blissfully unaware of hyperthreading, but power management compensates for it if "core parking" is enabled, from what I understand. I don't recall whether it is the "performance" or "balanced" scheme which enables the pseudo-awareness of HT by means of core parking.

 

[Kiska]

Here is what credit looks like for ~22 hrs of work: http://www.rechenkraft.net/yoyo/result.php?resultid=39153225

 

[TennesseeTony]

I guess I was fortunate to have actually stock-piled 5 days worth of pre-race tasks. I aborted hundreds per Xeon this morning, and still have plenty to reach the finish line.

 

[StefanR5R]

Switched to Muon and got lucky to fetch 8 tasks which allegedly take 4, 7, or 10 hours.

Yeah right. In fact, these tasks take 19+ hours.

So regarding (a) excessive actual runtimes, (b) utterly wrong estimated runtimes set by the program in new tasks, the subprojects ECM, OGR, Muon (and probably Evo) are equal offenders.

 

[StefanR5R]

The 2P boxes got all their 23 h tasks done.

 

[Orange Kid]

YOYO@Home Sprint Top 10

Day Three - Final
1 TeAm AnandTech ..............25 ......2,527,748
2 Overclock.net .....................18 ......1,892,312
3 OcUK - Overclockers UK ...15 ......1,393,241
4 meisterkuehler.de Team ....12.......1,020,843
5 Crunching@EVGA .............10 .........436,598
6 Ars Technica .......................8 ............56,318
7 BOINC.SK ...........................6 ............36,576
8 US NAVY ............................4 ............35,559
9 Team Norway ......................2 ...........23,966
10 team.se ..............................1 ...........22,784


YOYO@Home Marathon Top 10

The Start
1 Overclock.net................... 25 ........... 9,573,188
2 XtremeSystems ............... 18 ............ 2,189,004
3 AMD Users ...................... 15 ............ 1,751,334
4 Ars Technica ........ ........... 12 ........... 1,685,996
5 meisterkuehler.de Team .. 10 ........... 1,134,844
6 The Scottish Boinc Team .. 8 ............... 859,399
7 team.se ............................. 6 ................ 627,370
8 TeAm AnandTech ............. 4 ................ 621,026
9 Sicituradastra. ................... 2 ................ 556,376
10 Team Norway ................. 1 ................. 510,937

Day One
1 Overclock.net................. ..25 ..............10,432,940
2 XtremeSystems ...............18 ...............2,189,004
3 AMD Users ......................15 ...............1,776,897
4 Ars Technica ...................12 ...............1,726,447
5 TeAm AnandTech ...........10 ...............1,648,561
6 meisterkuehler.de Team ..8 .................1,444,511
7 The Scottish Boinc Team .6 ..................861,159
8 team.se .............................4 .................642,232
9 Sicituradastra. ...................2 .................557,696
10 Team Norway .................1 ..................530,117

Day Two
1 Overclock.net ..................25 ................10,942,116
2 TeAm AnandTech ...........18 .................2,369,665
3 XtremeSystems ...............15 .................2,189,004
4 AMD Users ......................12 .................1,782,530
5 Ars Technica ...................10 .................1,745,222
6 meisterkuehler.de Team ...8 .................1,740,835
7 OcUK - Overclockers UK ..6 .................1,060,864
8 The Scottish Boinc Team ..4 ...................862,039
9 team.se ..............................2 ..................647,011
10 Sicituradastra. .................1 ...................558,025

Day Three
1 Overclock.net ....................25 .................11,510,398
2 TeAm AnandTech .............18 ...................3,128,419
3 XtremeSystems ...............15 ...................2,189,004
4 meisterkuehler.de Team .12 ...................2,160,964
5 AMD Users .......................10 ...................1,788,510
6 Ars Technica .....................8 ...................1,763,110
7 OcUK - Overclockers UK ..6 ...................1,458,217
8 The Scottish Boinc Team .4 ......................863,315
9 team.se ...............................2 .....................657,123
10 Sicituradastra. .................1 .....................558,659

Well done TeAm
We took First place in the Sprint and moved up 6 spots overall !!!
Thank everyone and two weeks till the next one

 

[TennesseeTony]

Wow. We gained 39 points from just one race!

 

[Ken g6]

And now for the bad news. The next PrimeGrid race overlaps the next two FB races, the first partially and the second completely. Let's hope at least one is a GPU race!

 

[TennesseeTony]

Or, we could hope one of the two FB Sprints is PrimeGrid, as it is an eligible project.