Boinc Pentathlon 2024

[Skillz]

2 hours into the Pent and these are the standings in both events running.

 

[StefanR5R]

I wish to run for SG this year, as a token of appreciation for the many many many spectacular events they have hosted over the years.

Apropos. Pentathlon disciplines start and end at 2AM local time for SETI Germany. Announcements are made at 2AM, 8AM, 2PM, or 8PM in their local time. Likewise, World Community Grid Birthday Challenge days were aligned at 2AM local time. Just the SETI Wow!-Event's start and end times were scheduled more conveniently for them at 6PM local time.

That is, besides spending a lot of thought, skill, experience and work hours on the preparation and execution of these contests, the organizers and admins also put themselves up to a sleep cycle which is not natural for most persons, and not complying very well with many other kinds of activities which people may engage in in their daily lives.

This just crossed my mind ~an hour ago when I had just woken up and spotted pschoefer's shoutbox comments related to the sprint start ~6 hours earlier.

 

[Markfw]

Stats as of 5:30 am PST for PG ( we are killing it !)


And for NF

 

[StefanR5R]

The rate of network connections between Pentathletes and the NumberFields server is evidently a bit much for the server or its internet link.

I wonder what will happen during the City Run at Ramanujan Machine, where tasks are less than half the size of NumberFields'. On the other hand, there won't be traffic from GPU hosts with Ramanujan Machine. (And City Run will obviously be concurrent with Javelin Throw, and later probably with Steeplechase too; but one or the other will reduce traffic at Ramanujan only if they are CPU-only too.)

 

[Skillz]

I don't know, hopefully RNMA has a better Internet connection than Numberfields.

On the other hand, tomorrow my parts should get here so I can fire up the 4P 8890v4 Intel rig I have sitting on the floor doing nothing. I'll probably put it on Primegrid since it's got 60MB of L3 cache on each CPU.

But anywway, here are some stats updates.

 

[Markfw]

I need to say this now. I was in the hospital earlier in the week for a while. In case it looks like my numbers drop, feel free to post a question, but that could be the reason if I have to go back in, since I can't babysit and miss a down computer. I had that happen today. Blew a circuit breaker. It took a 9554 64 core and a 7742 64 core off-line.

 

[StefanR5R]

Stay well Mark!
(And I am not saying that for sake of our Pentathlon scores.)

 

[Skillz]

Those of you who are going to start on the RNMA City Run in an hour make SURE you run the built-in BOINC CPU Benchmark BEFORE downloading any tasks. The project uses CreditScrew to calculate points which is heavily based on the CPU benchmark score. If you don't your credits will be less than 50% of what they could be.

 

[Orange Kid]

The 1st Javelin Throw has been announced:
Start: 10 May 2024 00:00 UTC
End: 11 May 2024 00:00 UTC

NFS@Home

 

[crashtech]

NFS seems the opposite of Ramanujan as far as getting tasks .

 

[StefanR5R]

Last night I posted about the low rate of work assignment of Ramanujan to SETI.Germany's forum thread. Pschoefer responded that they already had been speaking with them about it and that it RNMA said that they'll look into it.

Also last night, one hour after I went to sleep, my Internet link went down and stood down until I saw it in the morning. Needless to say that my Ramanujan buffers weren't sufficient to bridge those >5 hours connection loss. — I divided the hosts which I have at RNMA into many small instances each now and of course also made sure that they don't back off after unsuccessful work requests. :-(

I hate it when a contest devolves into a beg-for-work contest like this one (or into a beg-for-getting-results-reported contest like the 2022 Obstacle Run at Universe).

 

[Skillz]

Everyone starting to drop bunkers on Numberfields, including P3D. Do they have enough to catch us? Guess we'll find out.

@StefanR5R I've done the same on my hosts running RNMA as well. Seems to help keep the CPUs busier than before, but I'm afraid this will hurt the project with so many more requests. So I'm watching it to make sure it doesn't kill anything unintentionally.

 

[StefanR5R]

Everyone starting to drop bunkers on Numberfields, including P3D. Do they have enough to catch us? Guess we'll find out.

Might take more than a day to find out what they have. :-/

 

[Orange Kid]

The first discipline is complete. Congrats to all the teams.
The Sprint top three are

1
TeAm AnandTech
68 233 802

2
Planet 3DNow!
40 079 396

3
SETI.Germany
26 330 014

 

[TennesseeTony]

Say hello to my new grandchild!




I couldn't resist, the price is for the components alone (or less?), thanks to Intel's/Motherboard Mfg'rs overzealous power pumping scheme to beat AMD.

https://www.newegg.com/msi-infinite-rs-13nui-419us/p/N82E16883151331?Item=N82E16883151331

 

[Markfw]

Say hello to my new grandchild!

View attachment 98567


I couldn't resist, the price is for the components alone (or less?), thanks to Intel's/Motherboard Mfg'rs overzealous power pumping scheme to beat AMD.

https://www.newegg.com/msi-infinite-rs-13nui-419us/p/N82E16883151331?Item=N82E16883151331

AIO on an OEM system ???? Well, the CPU won't do much in DC but that 4090 sure will !

What did you have to give for that. ? OOS newegg does not show the price.

 

[StefanR5R]

Might take more than a day to find out what they have.

Things calmed down a lot not long after I wrote that. As far as I can tell from logs, the NumberFields server could be reached very well again during the last ~8 hours of the sprint.

Edit,

the combined credit of the Pentathlon teams in this year's Sprint was 11.5% higher than last year at the same project

Could have turned out less, given the general trend to higher electricity prices and other costs of living. But didn't.

 

[TennesseeTony]

What did you have to give for that. ? OOS newegg does not show the price.

$2650 plus tax

I relentlessly questioned the big/little cores, and they gave me their word they would give me absolute hell, when trying to multi-thread.For single threaded tasks they pleaded the 5th, but seem to be okay on the Rammin'Jammin' machine.

 

[StefanR5R]

the combined credit of the Pentathlon teams in this year's Sprint was 11.5% higher than last year at the same project

Could have turned out less, given the general trend to higher electricity prices and other costs of living. But didn't.

PS, AFAIU NumberFields workunits differed between last year and this, but I don't know if PPD changed due to that.
But more importantly, the 2023 sprint was concurrent with GPU projects, and CPU bunkering for the sprint was concurrent with Javelin Throw.

Year-over-year turnout in the NumberFields sprint of the respective top 5 teams:
...................................................... 2023 .................... 2024
TeAm AnandTech ...................... 60.2 M ................ 68.2 M (+13 %)
Planet 3DNow! ........................... 40.0 M ................ 40.0 M (+0.2 %)
LinusTechTips_Team ................ 34.5 M ................ 19.6 M (−43 %)
SETI.Germany ............................ 14.4 M ................ 26.3 M (+83 %) <---- the @TennesseeTony effect! :-)
Rechenkraft.net + NoTEAM ...... 14.0 M ................ 16.5 M (+18 %)
Czech National Team ................ 13.3 M ................ 19.4 M (+45 %)

 

[Skillz]

But also last year the Obstacle course (Marathon) was a crap show and thus most people didn't put their CPUs on the project or at least fully on the project due to issues. This year PG is the Marathon and it's got no problems running CPU to its knees.

So the fact that two, stable CPU projects are running simultaneously and we still did better than last as far as points concerned is definitely worth mentioning.

 

[TennesseeTony]

SETI.Germany ............ 14.4 M ........... 26.3 M (+83 %) <---- the @TennesseeTony effect! :-)

I subtracted my contribution and they still won Bronze by over 1M points. pschofer's sneaky 2M last hour bunker was risky, lol

 

[StefanR5R]

(projects concurrent with Sprint)

But also last year the Obstacle course (Marathon) was a crap show and thus most people didn't put their CPUs on the project or at least fully on the project due to issues. This year PG is the Marathon and it's got no problems running CPU to its knees.

True! 2023 Obstacle and 2024 Marathon are quite different, not just because of the projects and servers involved, but also regarding the contest tactics.
(I don't think I have proper notes about ECM to dare to come up with an estimation of how many FLOPS the teams put at ECM then vs. are putting at CUL now...)

 

[Ken g6]

Steeplechase is SiDock. Which is not even a GPU project.

 

[StefanR5R]

GPU projects timeline:
2017 – Einstein
2018 – PrimeGrid (and Asteroids)
2019 – Einstein
2020 – Amicable Numbers (and NumberFields)
2021 – PrimeGrid and Einstein
2022 – PrimeGrid and Einstein and SRBase %-O (and NumberFields)
2023 – PrimeGrid AP27 and Einstein (and NumberFields)
2024 – (just NumberFields)

________
Foot note: Asteroids@home and NumberFields@home both have GPU-accelerated applications with lower PPD/W than their CPU-only applications.

 

[Skillz]

This is good for us since we're much stronger on CPUs than we are on GPUs by a good bit.

Not that we're weak on GPUs by any means, but we do have more CPUs than GPUs I believe.