We need help on Primegrid...

[Skillz]

We're trying to boost the sub project PPSE into the Top 5000 Primes so they'll be eligible for Tour de Primes next month. Any and all help would be greatly appreciated.

Last Saturday we needed 4.2 MILLION tasks completed to get PPSE into the Top 5k Primes. Right now we are down to 4 million tasks needed.

Since the T5K list is constantly growing the target to reach is constantly moving as well.

PPSE is an easy PG project. Low level 3 cache. Run them on physical cores only, no SMT or HT. Single threaded, Multithreaded isn't needed. 1task per physical core. Tasks complete in under 1 hour each on nearly any host, but under 10 minutes on really fast hosts like 7950X or 9950X.

 

[Landjunge]

I've put all cores on all machines to PPSE. Hope this helps.

 

[Skillz]

You rock my friend!

 

[JagDoc]

Run them on physical cores only, no SMT or HT.

My 7945HX@45W (2,5GHz) SMT on 10% more output.

16WU 1C   930.38      928.42    93.56     139k/Tag    FFT length 144K
32WU 1T 1,691.22    1,687.60    93.56     153k/Tag      FFT length 144K

16WU 1C 1,265.94    1,263.96    101.88    111k/Tag      FFT length 192K
32WU 1T 2,285.28    2,275.98    101.88    123k/Tag      FFT length 192K

 

[Skillz]

We have actually tested that it's only on zen4 that smt on does give a slight increase in PPD, but all our other testing with other CPUs doesn't. Including Zen5. It also uses a lot more power, but very little gain.

We've tested on multiple Xeons, Zen2, zen3 and Zen5 and none of them show any gains but instead of usually they lose PPD.

 

[thorsam]

I'll be at home in about 5 hours, then i switch the cluster cores.

 

[sompe]

Threadripper 1950x and Ryzen 7 5700G are working.

All tasks for computer 1342718

All tasks for computer 1077331

 

[JagDoc]

It also uses a lot more power, but very little gain.

Thank you for point out, without SMT power drop from 100W to 75W.

 

[Skillz]

I'll be at home in about 5 hours, then i switch the cluster cores.

You had a nice collection of hardware man! Thanks for helping PG out with PPSE.

We are around 3k or less digits to get PPSE into the T5K. We've been moving the goal much faster than anticipated! Thanks to all of P3D for throwing what you can at it.

I think we will have PPSE in the T5K for the PPSE Challenge later this month. If so, then any primes found during that time will go on the T5K list.

 

[xii5ku]

Seit Mitternacht sind PPSE-Primzahlen wieder groß genug, um in die T5k aufgenommen zu werden. Jetzt braucht das PPSE-Subprojekt allerdings noch weiterhin ein gewisses Level an Rechenleistung, damit PPSE auch noch bis Ende des Monats T5k-Größe hält.

Eine erste T5k-Primzahl hat auch schon Markfw/TAAT gefunden (PG/T5k) und eine zweite Scott Brown/ATP (PG/T5k).

Liste aller PPSE-Primzahlen in chronologischer Reihenfolge: link

 

[xii5ku]

Fortschritte der T5k-Liste und des PPSE-LLR-Projekts in den letzten Wochen, gemessen in Dezimalstellen der Primzahlen (T5k min vs. PPSE max):

Datum ....... T5k .............................. PPSE
14.12. ........ 618,192 (+2,303) ....... 608,289 (+843)
21.12. ........ 618,854 (+662) .......... 609,282 (+993)
28.12. ........ 621,335 (+2,481) ....... 611,291 (+2,009)
04.01. ........ 622,314 (+979) .......... 617,388 (+6,097)
11.01. ........ 624,040 (+1,726) ....... 624,491 (+7,103)

Quelle: https://www.primegrid.com/forum_thread.php?id=10031
Dort gibt es jeden Samstag Abend ein ASCII-Art-Diagramm dazu.

Inzwischen hat auch @Landjunge eine Primzahl gefunden, die in der T5k verewigt wurde. Glückwunsch!

 

[thorsam]

That means we are good enough, or do we need more power?

 

[Skillz]

We need to maintain it until Feb 1st. Also, the PG admins are trying to reach you. You seem to not have noticed you've got emails or PMs from them.

Basically you are finding top 5k primes but your account isn't set up to report them. If you want your primes auto reported to t5k then you need to update your settings on PG to have them reported for you. So you get credit for finding them.