Info PrimeGrid Challenges 2024, sieve-free edition

[Ken g6]

It's 2024 and we're all sieved out. Only prime-finding projects this year!

	Date	Time UTC	Project(s)	Best on	Challenge	Duration
1​	23-26 January​	01:00:00	PPSE	CPU	Hilbert's Birthday Challenge	3 days
2​	20-25 March​	03:14:00	ESP	CPU	"Einstein" Anniversary Challenge	5 days
3​	14-21 April​	12:14:00	TRP	CPU	JUICE's Birthday Challenge	7 days
4​	12-19 June​	19:00:00	GFN-19	GPU	PrimeGrid's 19th Birthday Challenge	7 days
5​	8-13 August​	08:08:00	Factorial Primorial	CPU?	(Tentative) International Cat Day Challenge	5 days
6​	22-29 September​	12:44:00	GCW	CPU	Autumnal Equinox Challenge	7 days
7​	25-30 October​	06:12:00	SR5	CPU	Cascade Day Challenge	5 days
8​	13-20 November​	07:00:00	PSP	CPU	World Kindness Day Challenge	7 days
9​	10-20 December​	18:15:00	GFN-21	GPU	Ada Lovelace's Birthday Challenge	10 days

And here's that possibly useful guide to cache sizes I found last year:

 

[StefanR5R]

5​

8-13 August​

08:08:00

Factorial Primorial

???

(Tentative) International Cat Day Challenge

5 days

From what I understand, ??? = CPU.

I've just created a new project, "PRST testing" to test new prime searching program - PRST.

PRST is a new prime testing program, written from scratch by Pavel Atnashev, author of LLR2. This is still a work in progress, but it's quite stable and basic types of tests are already implemented. The thing most important to us is that PRST can test factorial and primorial primes with fast double-check and Gerbicz-Li error detection.

(source)

The "PRST testing" project is over. All tests has been processed.

In this project we double-checked PrimeGrid PRPNet work on factorial and primorial primes. About 0.4% of original tests were invalid, but no missing primes were found.

Eventually, search for new factorial and primorial primes will be restarted on PrimeGrid.

(source)

 

[mmonnin03]

Factorial Primorial is GPU only on PGFN and just started.

Message boards : Factorial / Primorial

 

[Ken g6]

Factorial Primorial is GPU only on PGFN and just started.

Message boards : Factorial / Primorial

That's the sieve for the project. Not the project itself.

 

[Skillz]

I've been running the project on a few of my GPUs in hopes to get enough done so it can be moved to PG soon.

edit
I asked on Discord, the new project is definitely going to be CPU only when it starts on PG.

 

[Ken g6]

More news on GPUs: The AP27 project will continue searching for an AP28. So GPUs will continue to have something to do on PrimeGrid besides GFN.

 

[StefanR5R]

Data allocation sizes, checked with a Haswell CPU (FMA3) with the ranges of candidates from https://www.primegrid.com/server_status_subprojects.php as of today:

• PPSE-LLR (January challenge) ............. 0.9375 … 1.125 MB
• ESP-LLR (March challenge) .................. 18.75 … 20.0 MB
• TRP-LLR (April challenge) ..................... 10.0 … 12.0 MB
• GFN-19 (June challenge) ...................... 10.1 MB
• FPS/PRS (August challenge) ................ not active yet
• GCW-LLR (September challenge) ........ 18.75 … 24.0 MB
• SR5-LLR (October challenge) ............... 7.0 … 9.375 MB
• PSP-LLR (November challenge) ........... 20.0 … 25.0 MB
• GFN-21 (December challenge) ............. 40.5 MB

Keep in mind that data allocation sizes are creeping up all the time at LLR based projects, but rarely change at GFN projects. Also note, most GPUs are more efficient than most CPUs at the two GFN challenges; the rest are CPU-only challenges.

 

[StefanR5R]

Data allocation sizes […] as of today:

…are listed at the "Edit PrimeGrid preferences" web page too. :-O

 

[waffleironhead]

Just under 7.5 hours till the next challenge.

"PrimeGrid will celebrate David Hilbert's 162nd birthday with the inaugural challenge for the 2024 challenge series: a 3 day PPSE challenge from January 23 01:00 UTC to January 26 01:00 UTC. David Hilbert was a German mathematician and one of the most influential mathematicians of the 19th and early 20th centuries. His work made the first foray into the foundations of mathematics, from proof theory to reverse mathematics, and Hilbert's 8th Problem, the Riemann Hypothesis, remains one of the most important unsolved problems in mathematics, underpinning results from number theory to cryptography."

 

[Ken g6]

And we're off! (For the past 30 minutes or so.)

 

[Skillz]

I had a rolling start. Wasn't home when it started, so I started my stuff several hours earlier. Of which I discovered a few hosts wasn't getting tasks (after I left home) but managed to get them fixed relatively quickly. So far everything of mine seems to be going good except the heat they're dissipating now. Had to open the doors a little bit wider for these tasks compared to Universe@home tasks haha

 

[StefanR5R]

I just came home from work and found that my garbage internet connection broke down 6½ hours ago and didn't automatically reconnect. Two boinc clients ran out of work at some point, the others still had some left.

 

[Ken g6]

Day 1 stats:

Rank___Credits____Username
5______3004823____[TA]Skillz
7______2904434____crashtech
9______2374034____w a h
13_____1820418____xii5ku
15_____1550049____cellarnoise2
37_____545293_____biodoc
38_____535493_____mmonnin
52_____389282_____waffleironhead
70_____293074_____Orange Kid
71_____274024_____markfw
137____117807_____Skivelitis2
169____76546______Ken_g6
220____38381______SlangNRox
223____38030______johnnevermind
228____36244______ChelseaOilman
230____32144______Fardringle
361____5058_______kiska
387____2802_______geecee

Rank__Credits____Team
1_____14130638___TeAm AnandTech
2_____10266601___Antarctic Crunchers
3_____9054420____Czech National Team
4_____7429948____SETI.Germany

We've also found quite a few primes already! Congratulations to:

@Skillz 3 primes
@cellarnoise 2 primes
@crashtech 2 primes
@emoga 1 prime
@Skivelitis2 1 prime

 

[Orange Kid]

Thanks for the stats update.
Looks like I will need to increase my core count in the coming year.
Funny that my 3950x is out producing my 5950x.

 

[waffleironhead]

Thanks for the stats update.
Looks like I will need to increase my core count in the coming year.
Funny that my 3950x is out producing my 5950x.

Are you just comparing task count on your user page, or ppd? It looks like my 3700x is outperforming my 5700g. Not sure why either. The 5700g has a clock advantage and higher ipc.

 

[StefanR5R]

I just came home from work and found that my garbage internet connection broke down 6½ hours ago and didn't automatically reconnect. Two boinc clients ran out of work at some point, the others still had some left.

Almost all of the other clients ran out of work too while they uploaded and didn't request new work because of too many uploads in progress. It was late yesterday and I did not bother to switch over to new client instances which would fetch work immediately.

And while I was asleep, the internet link went down again and did not reconnect itself, 4½ hours ago. The Epyc 9554P ran out of work of course; I've got just a single client instance on it for now and ~1100 PPSE tasks are but a small buffer for it.

This internet service provider is complete garbage. But as I mentioned at another occasion, the locally available alternatives offer much lower specs (useless to me because the upload speed of my current link is already inadequate sometimes for the current amount of computers here) or/and want considerably more pay.

Connection loss without automatic recovery happens about every other month or so, but it stands to reason that this is more likely to happen during a PPSE challenge with heavy use of the connection due to the high frequency of server accesses that the clients need to perform.

It's possible that FTTH is going to be built here next year.

 

[cellarnoise]

Almost all of the other clients ran out of work too while they uploaded and didn't request new work because of too many uploads in progress. It was late yesterday and I did not bother to switch over to new client instances which would fetch work immediately.

And while I was asleep, the internet link went down again and did not reconnect itself, 4½ hours ago. The Epyc 9554P ran out of work of course; I've got just a single client instance on it for now and ~1100 PPSE tasks are but a small buffer for it.

This internet service provider is complete garbage. But as I mentioned at another occasion, the locally available alternatives offer much lower specs (useless to me because the upload speed of my current link is already inadequate sometimes for the current amount of computers here) or/and want considerably more pay.

Connection loss without automatic recovery happens about every other month or so, but it stands to reason that this is more likely to happen during a PPSE challenge with heavy use of the connection due to the high frequency of server accesses that the clients need to perform.

It's possible that FTTH is going to be built here next year.

Contact the Fiber company and let them know you are interested! That may motivate them to focus on your area first?

 

[StefanR5R]

Contact the Fiber company and let them know you are interested! That may motivate them to focus on your area first.

The timeline for this city district is already fixed. Roadside construction will start later this year, works within the buildings which are owned by 'my' housing cooperative are scheduled for next year. The only thing which I have to find out yet is whether or not this very house, which is somewhat detached from the other blocks of the cooperative, is subject to the same schedule.

Edit

Almost all of the other clients ran out of work too while they uploaded and didn't request new work because of too many uploads in progress. It was late yesterday and I did not bother to switch over to new client instances which would fetch work immediately.

And while I was asleep, the internet link went down again and did not reconnect itself, 4½ hours ago. The Epyc 9554P ran out of work of course; I've got just a single client instance on it for now and ~1100 PPSE tasks are but a small buffer for it.

During recovery from this second network outage, two more client instances on another host ran out of work too. --- As long as most hosts still have some work left = keep computing, the recovery = uploading the whole involuntary bunker takes almost as long as the network outage took.

From this follows: Deeper buffers (by means of more client instances, or by means of client source code modification for more than 1000 runable tasks) won't really help. For steady state computation of PPSE with as many CPUs as I've got, I need the internet link to be up for least 50% of the time! And for somewhat reasonable buffer sizes, downtimes should be no longer than three hours individually; which doesn't sit well with a natural sleep cycle nor with being away for work.

In other words, for me, PPSE is comparably light on computation and heavy on networking == yet another project which is no fun at all for me. Initially I was looking forward to this challenge because PPSE doesn't need more cache than Zen 2 has got. (I have a few Zen 2 cores...) But now I can't wait for this contest to be over, so that I can return to something with normal networking requirements, like MilkyWay.

 

[StefanR5R]

Overall challenge stats, reimplemented by DeleteNull of SETI.Germany based on Eudy Silva's work:

• http://www.rechenaugust.de/boinc/2023_Challenge_Series_Current_Standings_Teams.html
• http://www.rechenaugust.de/boinc/2023_Challenge_Series_Current_Standings_Individuals.html
• http://www.rechenaugust.de/boinc/2024_Challenge_Series_Current_Standings_Teams.html
• http://www.rechenaugust.de/boinc/2024_Challenge_Series_Current_Standings_Individuals.html

Another site with season stats (and per-challenge stats), which was made earlier by ComradeTurtle of team Storm:

• https://pg.comradeturtle.dev/ (may not work in some browsers)

 

[Orange Kid]

Are you just comparing task count on your user page, or ppd? It looks like my 3700x is outperforming my 5700g. Not sure why either. The 5700g has a clock advantage and higher ipc.

Just looking at the Tasks returned last 24 hours. That shows the real task not the double check tasks, if I understand correctly.
Thing1L = 3950x
Thing2L = 2700x
Thing3L = 5950x
Thing4W = 5900

Tasks returned in the last 24 hours​

Sub-project	Host	Threads	Tasks	Firsts	First percentage	Send/receive duration (avg/min/max)	Elapsed time (avg/min/max)	CPU time (avg/min/max)
PPSE (LLR)	Thing3L	2​	1007​	1007​	100.00​	17,219 / 13,428 / 20,751​	1,095 / 887 / 1,241​	2,159 / 1,743 / 2,444​
PPSE (LLR)	Thing2L	2​	329​	329​	100.00​	24,445 / 15,796 / 55,927​	1,542 / 1,314 / 1,806​	3,023 / 2,582 / 3,536​
PPSE (LLR)	Thing4W	2​	815​	815​	100.00​	16,901 / 9,972 / 19,371​	970 / 800 / 1,083​	1,146 / 793 / 1,400​
PPSE (LLR)	Thing1L	2​	1199​	1199​	100.00​	28,595 / 12,138 / 65,817​	966 / 811 / 1,125​	1,895 / 1,585 / 2,188​

 

[StefanR5R]

Strictly speaking, you can't just go by tasks/day (or by # of concurrent tasks / average task durations) when random tasks are involved. The workunits vary to some degree. Though if you average over a lot of tasks (and that's certainly the case with a thousand tasks) which were fetched by both hosts during the same period of time, then the error of measurement due to workunit variations is reduced. This error mostly vanishes if you take the credit of each workunit into account.

I followed your links and get this credit/time from the 50 most recent valid main tasks:
3950X: 75.9 average credit / 964 s average elapsed time = 0.0788 average points/s per task
5950X: 76.7 average credit / 1083 s average elapsed time = 0.0709 average points/s per task

Does the 3950X have a higher power limit perhaps?

BTW, you are running 2-threaded tasks. I suspect that 1-threaded tasks are better, using either 100% or only 50% of the logical CPUs. I haven't tested this myself, but couldn't give projections for 3950X/5950X anyway because my CPUs have lower per-core power budgets, therefore may have a different sweet spot.

 

[Orange Kid]

Don't really have any answers as the machines are 1300 miles away from me at the moment.
I do know they are basically the same. Motherboard, ram, power same. Different GPU'S though. That may be enough to create the difference along with the CPU. A more power hungry GPU with the 5950x.

 

[mmonnin03]

Woo I found a prime today. I saw waffle also had some.

 

[Ken g6]

Day 2 stats:

Rank___Credits____Username
6______6836950____[TA]Skillz
7______6080913____crashtech
10_____4680664____w a h
12_____3955672____xii5ku
14_____3450191____cellarnoise2
36_____1164663____biodoc
37_____1090027____mmonnin
55_____763107_____waffleironhead
74_____595880_____Orange Kid
102____379065_____markfw
112____326331_____ChelseaOilman
142____238132_____Skivelitis2
177____154582_____Ken_g6
214____100572_____SlangNRox
222____92555______johnnevermind
224____90505___10esseeTony
261____59577______kiska
292____37969______Fardringle
404____8356_______geecee

Rank__Credits____Team
1_____30015552___TeAm AnandTech
2_____21497202___Antarctic Crunchers
3_____20129102___Czech National Team
4_____18666365___AMD Users

@cellarnoise are you a furnace? 'Cause you're on fire! Primes found today:

cellarnoise2 : 3 primes!!!
@Skillz: 2 primes!!
@mmonnin03: 1 prime!

 

[cellarnoise]

The office / basement area where most of the puters are located is nice and toasty