2020 Formula Boinc Sprints

[Howdy]

KO stats eht rof sknahT

 

[StefanR5R]

Help me out, are we doing Einstein@home here, or Enigma@home?

 

[StefanR5R]

Back on Thursday, a handful of "Gravitational Wave search O2 Multi-Directional GPU" [GW] tasks slipped onto my computers. Here are results from GTX 1080Ti (Pascal, 3,584 shaders, 11 GB RAM) on Linux:

• running one GW job exclusively per card:
  ~11m25s...11m50s run time, ~85 % shader utilization

• running one GW job and one other random job per card (in hindsight, the parallel jobs were most certainly FGRP, but I wasn't there to watch):
  ~1h00m...1h15m run time, ~100 % shader utilization

In contrast, the change of "Gamma-ray pulsar binary search #1 (GPU)" [FGRP] run time from running 1 FGRP exclusively per card, to running it concurrently with another job (doesn't matter if FGRP or GW) is the expected less than double time on these cards.

Edit,
only two out of four computers showed this extraordinary increase in GW run time when they went from 1 to 2 jobs per card. I don't know why that is. OS, driver, and hardware are same between computers with and without this behavior; and WU names are all over the place. A possibility is that only a GW+FGRP mix on a card has this bad influence on GW run time (compared with GW+GW), but I would have to compare Thursday logs to find out what ran when precisely.

 

[Assimilator1]

Weird!

To anyone who was following my issues of E@H and my HD 7870 XT, it turned out to be 2 problems.
Yes 3 GB isn't enough vRAM for all the grav wave search WUs, but apparently Tahiti GPUs also seem to lack some sort of h/w function (I'm guessing) that means it crunches them at a really slow rate, hence timing out on the lower frequency range WUs. In the E@H forums they tested and discovered this about a year ago! Can't they block certain apps from sending WUs to incompatible GPUs?

Other things I noticed about E@H (GRPBS WUs), the RX 580 is about 33% faster than my HD 7870 XT(DS), RX 580 does them in roughly 584s, the HD 7870 XT roughly 868s, and my i7 4930k mostly 30,000 - 36,000s! lol (a few took just 10-11,000s though, all 693 credits each, weird. Downloaded CPU tasks in error, didn't know they had them! Something new??).

If you want POINTS, say for a Sprint, one should only be running the Gamma-ray pulsar binary search #1 (GPU) . Gravitational wave points are horrific in comparison.

Boy are you right! On my RX 580 GWS WUs get done in ~1550s & get 2000 credits each, the GRPBS take ~584s and get 3465 credits each! wth!?
I was going to re-enable the GWS app for my 580, but after seeing that I didn't bother! I have already got a bunch of those WUs, hmm........ tasks suspended .

 

[biodoc]

Late bunker drops and points by @StefanR5R (17,160,475 points) and @phoenicis (14,993,055 points) have moved us into 3rd place! Wow!

 

[Orange Kid]

Einstein@home Final

From tenth to third in two days, quite the finish
Great sprint all, take some time to cool off and relax.

1	XtremeSystems	25	159,967,684
2	The Scottish Boinc Team	18	83,804,507
3	TeAm AnandTech	15	72,196,527
4	UK BOINC Team	12	62,869,913
5	Gridcoin	10	57,251,041
6	BOINC@AUSTRALIA	8	44,429,192
7	Russia Team	6	41,916,630
8	Team China	4	39,370,556
9	Rechenkraft.net	2	33,842,404
10	SETI.Germany	1	25,705,181

From FreeDC

biodoc	5,207,895	4,819,815	1,732,500		11,760,210
crashtech	3,300,985	2,783,890	691,635		6,776,510
Lane42	1,347,885	1,233,540	852,390		3,433,815
10esseeTony	1,677,060	859,320	620,235		3,156,615
Icecold-Team Anandtech	985,970	689,315	544,205		2,219,490
Orange Kid	784,325	557,620	388,650		1,730,595
Howdy	679,140	218,295	76,230		973,665
phoenicis	582,120	259,875	58,905		900,900
Ken_g6 - TeAm Anandtech	232,848	231,462	159,731		624,041
[TA]Assimilator1	270,058	177,476	28,930		476,464
elawyn	106,395	144,930	187,530		438,855
Kiska	217,925	22,000	0		239,925
Ingleside	13,860	48,510	62,370		124,740
Alan J. Simpson	50,146	16,757	31,878		98,781
zzuupp	43,009	42,000	0		85,009
Jesper S. Rytke	18,693	34,000	24,000		76,693
Endgame124	12,375	125	313		12,813

 

[Assimilator1]

Nice dump!

Cheers for the stats OK ......err, btw, what do the 4 columns relate to?

 

[Howdy]

@Assimilator1

3rd day/ 2nd day/ 1st day =Total

 

[Assimilator1]

*sniff* no stats.

Is the Einstein sprint over? Op dates don't tally up.

 

[biodoc]

*sniff* no stats.

Is the Einstein sprint over? Op dates don't tally up.

Those were the final stats. The sprints only last 3 days + one day of bunkering of course since the sprint project is announced 24 hours before the start.

 

[lane42]

over

 

[Endgame124]

Finally picked up a wingman for a bunch of my work units. Only a little late for the sprint

 

[Assimilator1]

Damn! Send them a letter of complaint!

Ok, thanks guys. I realised that last night, so I turned off the WU tap to my 2nd rig.

 

[Orange Kid]

Moo! Wrapper is the sprint.

We all guessed wrong again.

 

[Fardringle]

Another GPU project? Bummer. Good for a lot of people, not for me.

 

[gizmo23]

Thought I'd join in a sprint this time. I went to the Moo! page but can't create an account because my reCaptcha response is wrong. Always. Tried about 10 times, it's always wrong, apparently. Looks like I'll sit this out and go back to F@H.

 

[StefanR5R]

Did you also already try to create the account through boincmgr?
I.e. use the "Tools -> Add project..." requester, select Moo! Wrapper, then keep "No, new user" selected, and enter e-mail address and your password-to-be.

 

[lane42]

preparing for Whiplash

 

[gizmo23]

Did you also already try to create the account through boincmgr?
I.e. use the "Tools -> Add project..." requester, select Moo! Wrapper, then keep "No, new user" selected, and enter e-mail address and your password-to-be.

Yes, that was the first thing I tried, but it doesn't set me up as a Team Anandtech member so I thought I'd better create an account first.

 

[gizmo23]

Yes, that was the first thing I tried, but it doesn't set me up as a Team Anandtech member so I thought I'd better create an account first.

Ah. got it. That first try created an account so I need to update that profile and then use it.

 

[Orange Kid]

Moo! Wrapper day one

1	XtremeSystems	25	119,115,561
2	The Scottish Boinc Team	18	82,732,162
3	TeAm AnandTech	15	44,025,476
4	UK BOINC Team	12	27,248,035
5	LITOMYSL Boinc Team Czech Republic	10	24,517,020
6	Planet 3DNow!	8	9,580,828
7	Rechenkraft.net	6	7,709,432
8	L'Alliance Francophone	4	6,035,656
9	Overclock.net	2	3,688,620
10	Crunching@EVGA	1	3,585,208

From Free DC

	Day 1
biodoc	12,243,840
crashtech	10,408,862
Orange Kid	3,753,392
Lane42	2,911,840
Kiska	2,795,520
Icecold-Team Anandtech	2,632,816
10esseeTony	2,487,666
xii5ku	1,711,432
waffleironhead	731,957
Ken_g6	725,872
Gizmo23	582,680
zzuupp	232,608

For those interested F1 racing is back with 10 races announced so far.
Two have been run.
Austrian Grand Prix circuit was run twice.

• Valtteri BOTTAS
• Charles LECLERC
• Lando NORRIS

and

• Lewis HAMILTON
• Valtteri BOTTAS
• Max VERSTAPPEN

this week is the Hungarian Grand Prix, then Great Britain twice, Spain, Belgium, Italy twice and last Russia.

 

[Ken g6]

I'm highly suspicious of the Czechs. They stayed right with us for awhile, then dropped like a stone. I think they're still with us, but hiding in bunkers.

 

[StefanR5R]

Until Friday 01:00 UTC (03:00 CEST = Czech local time), Litomyšl dumped 18.5 M. Then their PPH converged to a steady output, and they made additional 6.9 M in the next 24 hours. This latter output is basically the same as their average output during the two months before the sprint.

The question remains how many hours they needed to produce the initial dump. It could have been from a speculative bunker. Task deadline is 8 days.

Edit,
their active users appear to have been bunkering speculatively for Moo! Wrapper before of each of the past few Formula BOINC sprints, according to their Boincstats 60-day records. If they did so, it may not have been multi-day bunkers though, but buffers of single client instances per host perhaps.

 

[StefanR5R]

Here is the 1st half of the sprint, top ten teams of the 1st league:



And this is how the teams fare compared to a year ago:

​	​ a) June 2019, final score​	​ b) July 2020, half-time score​	b / a​
XtremeSystems​	238.3 M​	157.8 M​	66 %​
The Scottish Boinc Team​	29.8 M​	102.3 M​	343 %​
TeAm AnandTech​	169.3 M​	61.6 M​	36 %​
UK BOINC Team​	35.6 M​	39.6 M​	111 %​
LITOMYSL Boinc Team Czech Republic​	42.5 M​	28.3 M​	67 %​
Planet 3DNow!​	5.2 M​	18.2 M​	350 %​
Rechenkraft.net​	71.1 M​	14.8 M​	21 %​
L'Alliance Francophone​	16.4 M​	9.3 M​	57 %​
Crunching@EVGA​	145.9 M​	8.7 M​	6 %​
Overclock.net​	7.0 M​	5.7 M​	81 %​
[H]ard|OCP​	149.2 M​	0.01 M​	0 %​

I didn't capture half-time stats at the 2019 Moo! Wrapper sprint, hence can't make a more direct comparison.
I added [H]ard|OCP to the table since a good deal of their GPU capacity moved over to UK Boinc Team in the meantime. Mentally add part of [H]'s last result to UBT's result.
TSBT and Litomyšl ran in League 2 in 2019. It's obvious that TSBT's motivation has increased since then. ;-)

 

[Orange Kid]

Moo! Wrapper day two

1	XtremeSystems	25	198,967,777
2	The Scottish Boinc Team	18	130,150,347
3	TeAm AnandTech	15	82,471,032
4	UK BOINC Team	12	65,038,229
5	LITOMYSL Boinc Team Czech Republic	10	32,778,292
6	Planet 3DNow!	8	27,570,588
7	Rechenkraft.net	6	23,460,512
8	Crunching@EVGA	4	16,000,720
9	L'Alliance Francophone	2	13,865,441
10	Overclock.net	1	9,093,876

From FreeDC

	Day 2	Day 1			Total
biodoc	5,522,072	14,453,424			19,975,496
crashtech	5,074,604	12,389,326			17,463,930
Icecold-Team Anandtech	3,266,592	3,353,400			6,619,992
Orange Kid	1,956,768	4,551,088			6,507,856
Lane42	1,972,632	4,434,376			6,407,008
10esseeTony	1,428,704	3,694,154			5,122,858
xii5ku	2,188,744	2,547,576			4,736,320
phoenicis	4,292,696	265,072			4,557,768
Kiska	1,029,600	3,335,736			4,365,336
Ken_g6	396,784	848,072			1,244,856
waffleironhead	373,640	847,661			1,221,301
zzuupp	369,136	290,728			659,864