Discussion 2019 BOINC Pentathlon thread

[IEC]

Grrr... oh well, hopefully by this time next year I have a few Epyc rigs. Rome wasn't built in a day, after all.

 

[StefanR5R]

WCG stats went final at ~05:00 UTC. As far as I can tell they did not change significantly anymore from the last hourly stats at ~00:00 UTC. I.e., no major effects due to WCG's database shuffling at the day wrap.

----
Great show this year! There were several races which had really narrow outcomes.

For us it was a blessing to have several good friends with us this year. Which just had me thinking: The Pentathlon is expressly about the team skills of Mobilization, Communication, Strategy, Flexibility, Tactics. We had to be good in all of these to get this far, but in a way, Mobilization may have had the biggest effect for us this year. Looking at other teams, they sure all had a lot going in each of these 5 departments --- except for the overall winner of this year's Pentathlon (and one or another Pentathlon before) who simply excelled in Mobilization and could forgo the other skills pretty much entirely. :-)

----
Jeeper's final report in German: part 1, part 2
Edit, a preliminary translation:

...
5. TeAm AnandTech (4-9-4-4-4)
Previous year: 7
It was their best Pentathlon ever, and clearly so. They were able to improve constantly in recent years, and could celebrate for this reason. But a stale aftertaste remains as they narrowly missed Bronze several times, which is tragic. The overall medal was within reach but slipped shortly before the finish. All in all, probably the low Cross Country result put a spoke in their wheel.
...
The hectic time has ended now. Time to relax, review, prepare, and to ask questions.

• Will SETI.USA make a comeback?
• Can Rechenkraft.net excel again?
• Will TeAm AnandTech bring their deserved medal home?

 

[phoenicis]

Fantastic effort by all. Sooooo close.

 

[TennesseeTony]

We missed Marathon Bronze by 2,672,556 points. And thus those 10 points dropped us in the overall as well, knocking us down from 3rd to 5th. SG says: pschoefer: the WCG stats confirm that TC even won that Bronze on their own so those team hopping shenanigans were just a superfluous annoyance...

Ah, ok. freestman of Team China left the team to join Chinese Dream the day before, dumped his bunkers, then switched back to his home team (China) in the final hour. Sneaky!



It was one heck of a show all around, and I had a great time!

 

[StefanR5R]

Maybe not sneaky, but rather being a bit undecided about the extent of support of the teams freestman perhaps feels at home with. He has been going back and forth between both in the past.

Also, I suspect the dates which FreeDC and Boincstats are showing for team membership movements may be distorted by the infamous GDPR setting. Chinese Dream had WCG production during the entire Pentathlon considerably higher than what we are seeing e.g. here*:
https://boincstats.com/en/stats/15/user/list/8/0/5346
Hence I suspect freestman was with Chinese Dream all the time except for the last hour of Pentathlon.

----
*) even while taking the 7x conversion factor from boinc credits to WCG points into account

 

[crashtech]

This one wore me out. I think it might be the first time that I tried to run all five projects. I think we acquitted ourselves well, nobody expected a medal going in, in fact I think we were shooting for 7th, right? Way better than expected, except for Einstein.

 

[Howdy]

Special Thanks to the folks at [H] and OcUK for their help with this Pentathlon too. I believe that any of them from those 2 teams out produced me in 1 single drop of projects!!! (I'm used to that xii5ku guy smearing me across the stats pages normally )

 

[TennesseeTony]

So we learned two things at least from this Pentathlon:

• Our shiny 'new' fleet of 10/20 series Nvidia GPUs can't hold a candle to DP heavy old AMD cards.
• And, most of us could probably use some more RAM.

But I disagree with my own findings! Our shiny GPU fleet is in general far more effective. Usually. And in regards to memory, I forgot when I built the twin Ryzen 1700s to order more RAM, they have each been performing very well with 16 threads, and only 8GB of RAM (single channel too!).

So rare are these situations, where one needs ENORMOUS amounts of RAM, or Double Precision power, it's just not worth the cost to upgrade/have GPUs ready to swap out, just for an annual event.

Yeah okay, enough kidding. I have already ordered a 4x16GB kit to go in the threadripper. It needs more but even though 16GB 3200 sticks are the 'new normal', they are still kinda pricey.

And I know of at least two AMD R9-280X cards being purchased for the TeAm. lol.

Anybody else already making upgrades based on our weaknesses this year?

 

[Skillz]

My general rule of thumb when building new rigs for DC is 1GB of RAM per thread/core. Though, when I can afford it, I try to max out the RAM as much as possible. Makes running multiple VMs easier, too.

 

[StefanR5R]

Before I began building high core count computers, I always stuck as much RAM into my PCs and laptops as their boards/ chipsets/ RAM controllers would accept. Not because I had RAM hungry applications, but because I hated, and still hate, mass storage latency.

 

[crashtech]

Speaking of old AMD cards, last time I tried to use a 7970 in Linux, it didn't work out. Has that changed any? My plan for the year is to migrate to Linux slowly, add CPU cores that do AVX2 well, and keep my three spare 7970s ready to go in some outdated platforms just to run when they are needed. That's what that X79 board was doing when it died, hosting dual 7970's crunching Einstein. Later this year I might add better Nvidia cards for winter Folding, hopefully this year we'll find a willing opponent.

 

[Orange Kid]

We done good. Thanks to all and all our friends.
Gotta shoot for at least one medal somewhere next time.
Sorry to have had to bail a bit early.

 

[Markfw]

I was in 3 or 4 of the contests, and used everything I had (proof is the electric bill) I would love to add more hardware, but I can't afford to run everything I have now !

 

[Assimilator1]

If I'm a bit more on the ball next time, or someone PMs me or emails me , I'll stick my old 7970 back into my rig for any Einstein or MW race. It's stashed away atm, I was thinking of selling it, but I'll hang onto it .
How easy is it to crunch on a 2nd grx card with no monitor attached? (although, if it is a PITA, I could just hook up my 2nd monitor to it).

Tony
Re DP & old cards, I would have thought you'd remembered that from the MW benchmarks you gave .

 

[crashtech]

I have not found a list of which GPU projects can benefit from DP.

 

[Ken g6]

I have not found a list of which GPU projects can benefit from DP.

I think Einstein and MW is it. PrimeGrid used to have an app that benefited, but no longer.

 

[Markfw]

If I'm a bit more on the ball next time, or someone PMs me or emails me , I'll stick my old 7970 back into my rig for any Einstein or MW race. It's stashed away atm, I was thinking of selling it, but I'll hang onto it .
How easy is it to crunch on a 2nd grx card with no monitor attached? (although, if it is a PITA, I could just hook up my 2nd monitor to it).

Tony
Re DP & old cards, I would have thought you'd remembered that from the MW benchmarks you gave .

Hook the card up, get your BOINC working, and set it up to be monitored by BOINCTasks, then just unplug the monitor !

 

[StefanR5R]

So rare are these situations, where one needs ENORMOUS amounts of RAM, or Double Precision power, it's just not worth the cost to upgrade/have GPUs ready to swap out, just for an annual event.

Or, biannual.

With SETI.Germany's increasing difficulties to find projects which are ready and able to host the Pentathlon, it is not entirely unlikely that Einstein@Home (2019, 2017, 2015, 2013, 2010), LHC@home (2019, 2017, 2015), or yoyo@home (2019, 2017, 2015, 2013, 2012, 2011 - FFS) get elected again in 2021. So, we have got 23 months time to address our hardware deficits specifically as these projects are concerned. :-)

(Of course, SG could become desperate and revoke their rule that projects cannot return right at the next year. Or MilkyWay@Home (2010) could suddenly become stable and be elected in 2020...)

 

[Assimilator1]

Hook the card up, get your BOINC working, and set it up to be monitored by BOINCTasks, then just unplug the monitor !

Thanks Mark, you'll probably have to remind me in a years time .

 

[Howdy]

@Assimilator1 I use these, I found that some of my machines would not boot without a monitor connected. With the dummy plugs there has been no issues.

 

[Assimilator1]

Thanks for the link , but the plan is to put the HD 7970 as a 2nd card in my main rig, so booting up won't be an issue....... I hope!

 

[StefanR5R]

From May 5:

The Scottish Boinc Team has a great start. 71+ million points and a big lead. Let's see if they can hold it.

If their SUSA or XS guests withdraw out of TSBT before the 00:00 UTC mark, TSBT will lose all points made by those temporary guests.

(This is a quirk of WCG's stats keeping: Hourly stats work like all other BOINC projects. But WCG's daily stats are computed separately from a 00:00 UTC snapshot of user credits & user team affiliations.)

That would be a cruel thing to do.

Could also happen by accident, if those concerned don't know about this quirk. But at least one SUSA member should know about it precisely, as he used it in 2017 to temporarily inflate Team China's score.

Today, VietOZ claimed in the shoutbox, that the SUSA member was unaware of the WCG team scoring quirk at the time when he moved to Team China in 2017. (And then saw this effect, and withdrew before any further WCG production.) So on that count, I may have erred. And it makes sense. Out of possible motives for the team change --- such as (1.) a sympathy of this person for Team China and a wish for them to make it to #2 in 2017's City Run; or (2.) improving the distance between SUSA's and P3D's overall scores by 10 points, by pushing TC in front of P3D in City Run, thus increasing SUSA's chances for overall 1st rank, (3.) trying to confuse TC about their WCG standing by using a then mostly unknown quirk of WCGs scoring mechanisms; --- that 3rd item is actually the least likely one according to Occam's razor. (As for 1. or 2. or other possibilities, I have no recollection whether or not that was clarified afterwards.)

This is old stuff and long in the past, so sorry that I dug this out again. Just wanting to clarify that my previous perception of what happened there (the underlined part in the last quote) was quite possibly wrong, and something different was actually being attempted back then.

 

[ChristianVirtual]

A good reason for a method to signup at begin of a challenge as an individual via CPID and stay with the choice until victory lap