SETI@Home Wow Event 2017

[ao_ika_red]

Well, 8th will be up for grasp, then. A bit hard to get ngb and kwsn guys but in 4 days everything can happen.
Unfortunately, since updating s@h tasks on Wednesday, my computer kept crashing while running s@h. It's fine with mw@h but no luck with s@h.

 

[StefanR5R]

I have put some Linux CPUs aside for experiments. I am pitting two identical Broadwell-EP machines against each other: One with stock CPU applications, another with lunatics' binary "setiathome_v8 8.05 r3345_avx_linux64". (Besides this one, lunatics also have SSE, SSE2, SSE3, SSE4.1, and SSE4.2 binaries which I am not trying.)

Interestingly, https://setiathome.berkeley.edu/apps.php lists only a Linux CPU application 8.00. But the server does in fact send not only "SETI@home v8 8.00", but also "SETI@home v8 8.05". When I have both of these aplication versions running, some CPUs run at their standard all-core turbo, and the others at the AVX all-core turbo.

This means that one possible result of this experiment can be that there is now no need to bother anymore with lunatics for Linux CPU applications. We will see.

I did not attempt to learn something about this at the setiathome.berkeley.edu forums. I browsed them during the SETI@Home Formula Boinc sprint and did not spot conclusive, up-to-date information.

Is somebody reading here who is privy to SETI@Home's application version arcana?

 

[ao_ika_red]

In my opinion, Lunatics lib is more towards to older cpu users and especially to those that have no avx instruction support. At first, I tried to use Lunatics' because of SSE4.2 advantage of AMD CPU. But, once I've learned that s@h developer didn't support it anymore, I revert to stock lib from s@h. I'm compelling to try Lunatics' avx lib, but due to recent fiasco in my computer, I will hold it until s@h wow and FB sprint event end.

 

[StefanR5R]

........................ today ........... yesterday .......... -2 days .............. total
6. kwsn ...........1.21 M ............ 1.20 M ............ 1.17 M ............ 12.8 M
7. ngb ............ 1.18 M ............ 1.16 M ............ 1.12 M ............ 12.8 M
8. TeAm ........ 1.20 M ............ 1.14 M ............ 1.01 M ............ 10.8 M
9. ita .............. 0.55 M ............ 0.57 M ............ 0.58 M ............ 10.4 M
10. afr ............ 1.00 M ............ 0.98 M ............ 0.91 M ............ 10.2 M

With 2.6 days left, our ranking looks stable.
I took today's points at 12:00 UTC and doubled them for PPD.

A third of the Italian's total credits apparently results from a bunker built before the event by their top user. Hence the disparity between their rank and their PPD.

 

[StefanR5R]

Here is the comparison of stock vs. lunatics CPU applications on Linux, measured on dual E5-2690v4 hosts after a bit over 1.5 days (hence still with an appreciable amount of validation lag).

I am omitting Astropulse results due to the rarity of WUs, and am only showing Multibeam results.

Linux host with stock CPU applications:

RAC = 5,100
SETI@home v8 8.00 x86_64-pc-linux-gnu:

562 consecutive valid tasks
18.0 GFLOPS
min / max / average of latest 100 valid tasks:

28 / 12,600 / 9,000 s run time
0.3 / 141 / 95 credits per task
760 / 27,100 / 1,450 PPD per thread (need runtime-weighted average, calculated below)​

sum of latest 100 valid tasks:

897,000 s run time
9,550 credits
920 PPD per thread, 51,500 PPD per host​

SETI@home v8 8.05 i686-pc-linux-gnu (note, 32 bit application):

114 consecutive valid tasks
14.5 GFLOPS​

Linux host with lunatics CPU applications:

RAC = 6,700
SETI@home v8 8.00 x86_64-pc-linux-gnu (i.e. stock application, ran during the FB sprint in June):

102 consecutive valid tasks
17.5 GFLOPS​

SETI@home v8 (anonymous platform, CPU) = lunatics' MBv8_8.05r3345_avx_linux64:

764 consecutive valid tasks
23.6 GFLOPS
min / max / average of latest 100 valid tasks:

5 / 8,400 / 6,300 s run time
0.3 / 128 / 92 credits per task
840 / 7,900 / 1,360 PPD per thread (need runtime-weighted average, calculated below)​

sum of latest 100 valid tasks:

633,000 s run time
9,170 credits
1,250 PPD per thread, 70,000 PPD per host​

The stock host is at this moment running only tasks with the better performing SETI@home v8 8.00 x86_64-pc-linux-gnu version. Most of its cores are clocked at the all-core AVX turbo, while some cores are at the non-AVX turbo at least some of the time. The same is true for the CPU clocks of the lunatics-equipped host. (This mixed clocking is a benefit of Broadwell-EP. Haswell-EP pulls down all cores to the AVX turbo as soon as at least one core runs an AVX workload.) — This means that the stock v8 8.00 application is already using AVX, but not as effectively as the lunatics binary does.

Conclusion:

The lunatics binary is still better on Linux hosts for recent CPU models.* By a lot. (+36 % for E5-2690v4.)​

Oh, and I also pitted two dual E5-2696v4 against each other in the same manner. 72,500 PPD versus 93,100 PPD. (+28 % with lunatics for the 2696v4 which have moar cores than 2690v4 but same memory bandwidth.)

Edit:
*) @Kiska told me so in June but I hesitated at that time to install practically undocumented binaries from sources unknown to me.

 

[Assimilator1]

........................ today ........... yesterday .......... -2 days .............. total
6. kwsn ...........1.21 M ............ 1.20 M ............ 1.17 M ............ 12.8 M
7. ngb ............ 1.18 M ............ 1.16 M ............ 1.12 M ............ 12.8 M
8. TeAm ........ 1.20 M ............ 1.14 M ............ 1.01 M ............ 10.8 M
9. ita .............. 0.55 M ............ 0.57 M ............ 0.58 M ............ 10.4 M
10. afr ............ 1.00 M ............ 0.98 M ............ 0.91 M ............ 10.2 M

With 2.6 days left, our ranking looks stable.
I took today's points at 12:00 UTC and doubled them for PPD.

A third of the Italian's total credits apparently results from a bunker built before the event by their top user. Hence the disparity between their rank and their PPD.

Thanks for the stats Stefan , good to see we've gained a couple of places

 

[StefanR5R]

Funny, twelve hours ago I diverted xii5ku's resources from SETI entirely to MilkyWay, yet I received 100,000 SETI credits since then, of course due to ongoing validations.

I had >3000 WUs pending validation 12 h ago, now I am down to >2000 WUs pending. (And I have >2500 WUs downloaded yesterday, to be processed after I return from the MilkyWay tonight.)

 

[petrusbroder]

One of my crunchers lost the internet connection - dumped more than 150 WUs in less than 10 minutes. The result: more than 20K credits added and at least as much pending - if not more ... in about 12 hours I'll pass my best years production (530 K in 2015). All other computers have dumped whatever they had ...

 

[StefanR5R]

Me being new to Wow!-Event, its requirement that each user registers individually came across a bit strange to me at first. But just now it occurred to me that it very effectively levels the playing field with pseudo-teams such as Gridcoin. (Though I understand that inter team competition is not the central point of this event.)

On another note, while cosmos is vast beyond human comprehension, the bit of empty space which remains in front of the Taurus rocket ship is getting smaller and smaller by the hour. Makes me feel a little bit melancholic. (But at the same time relieved because my water loop is in dire need of maintenance now...)

 

[TennesseeTony]

I think next year I might buy a ticket on one of those ships in the rear. I mean, what if we're headed unknowingly headed to a bug planet like Klendathu? It'd be nice to have boots on the ground ahead of time, just in case.

 

[StefanR5R]

Team stats again.
Lots of unloading of bunkers to be seen in the rear view mirror, and 15 ¼ hours to go.

            total¹    today²  yesterday  -2 days
-------------------------------------------------
6.  kwsn   15.18 M    1.17 M    1.17 M    1.23 M
7.  ngb    15.07 M    1.09 M    1.11 M    1.21 M
8.  TeAm   12.91 M    1.19 M    0.87 M³   1.20 M
9.  afr    12.15 M    0.95 M    0.98 M    1.01 M
10. rkn    11.57 M    1.09 M    1.52 M⁴   1.08 M
11. ita    11.45 M⁵   0.50 M    0.54 M    0.57 M

¹) today at 12:00 UTC
²) today at 12:00 UTC, extrapolated to 24 hours
³) excused absence due to Formula Boinc
⁴) RKN uploaded a 450 k bunker yesterday (4500 WUs from just two cards working for 35 days according to their user) and overtook Italy shortly after that. They still have at least one bunker left and aim to overtake AF.
⁵) 4.3 M of those are from a bunker which their top user uploaded on the first day of the event.

 

[TennesseeTony]

A 35 day gpu bunker on a project that limits tasks....wow, dude is dedicated, lol.

 

[StefanR5R]

Two hours ago, RKN uploaded a 1.2 M bunker. Total credits at 17:00 UTC:

8.  TeAm   13.16 M
9.  rkn    13.02 M
10. afr    12.35 M

 

[ao_ika_red]

Two hours ago, RKN uploaded a 1.2 M bunker. Total credits at 17:00 UTC:

8.  TeAm   13.16 M
9.  rkn    13.02 M
10. afr    12.35 M

RKN's final sprint is worrying.

 

[TennesseeTony]

Great event everyone! We held 8th against the onslaught from Reckenkraft (spelling)!

Now we can all relax until 2200 UTC Wednesday, when we start the next FB Sprint (bunkering).

 

[Markfw]

FB ? help a newbie (if I can join, the electric bill is still killing me, it was 100f today)

 

[ao_ika_red]

All hail The TeAm!
I'll promise to come back with better GPU next year.

 

[StefanR5R]

We actually took a Triathlon on ourselves and did rather well in it.

There were only two teams which were better than us in at least 2 of 3 disciplines:
Czech National Team in PrimeGrid and MilkyWay,
SETI.Germany in PrimeGrid and SETI.

 

[TennesseeTony]

"FB" is short for Formula Boinc, Mark, sorry for the confusion.

Stefan, thanks for the comparison to other teams regarding our triathlon, that certainly puts things in an even better perspective!