Info PrimeGrid Challenges 2024, sieve-free edition

[waffleironhead]

Is it overclocked to 12.3 GHz and submerged in a vat of liquid nitrogen? Dang that's quick!

Sits at about 4.8ghz. Should be able to do 3 units a day. They really make my 5700g and 3700x look pathetic.

 

[emoga]

It's been two years and four months since the last ESP prime was discovered. Will anyone uncover one in this challenge?

 

[TennesseeTony]

It's been two years and four months since the last ESP prime was discovered. Will anyone uncover one in this challenge?

Yes. 2 will be discovered actually. One for me, and one for you. I've been watching ESPN and have discovered I am actually a psychotic. I even went so far as to go buy a lottery ticket.

Big primes do seem to be FEW and FAR between these days, don't they? Lot's of AP's left to find though.

******************

Anyone know which PG projects can benefit from AVX-512?

 

[emoga]

Yes. 2 will be discovered actually. One for me, and one for you. I've been watching ESPN and have discovered I am actually a psychotic. I even went so far as to go buy a lottery ticket.

Big primes do seem to be FEW and FAR between these days, don't they? Lot's of AP's left to find though.

******************

Anyone know which PG projects can benefit from AVX-512?

I was under the impression that every PrimeGrid sub-project with a CPU application benefited from AVX-512. 😎 If it's a GPU project, then the CPU version of it would also run better under AVX-512. Except maybe 'Do You Feel Lucky?' since there's no CPU version.

My 7950X was performing around 70% better than my 5950X on every project. The new AMD CPUs that are rumored to be released are supposed to have even better AVX-512 capabilities as well.

Might have to upgrade again here pretty soon. 💰💲

 

[TennesseeTony]

WTH?

We just started!? Ah, Main Task(s) coming in. I win!!!

 

[StefanR5R]

Back when AVX-512 support was enabled in PrimeGrid's LLR2 binary, there was a regression vs. AVX2 (FMA3) on certain lower-tier Intel Skylake-SP CPUs and derivatives which have only one out of the two FMA units in the cores enabled. I am not sure whether this regression is still there with the current application binary; perhaps it is. But I don't think anybody of us has got one of these CPUs. On all other Intel and AMD CPUs which do have AVX-512, PrimeGrid's applications benefit from it.

 

[Ken g6]

Day 1 stats:

Rank___Credits____Username
1______3634969____w a h
6______1523511____ChelseaOilman
8______1245383____cellarnoise2
24_____400839___10esseeTony
31_____323803_____Orange Kid
33_____303455_____biodoc
39_____259960_____waffleironhead
43_____251466_____Fardringle
48_____221073_____crashtech
114____72866______Letin Noxe
119____72476______Icecold
122____72166______Ken_g6
181____831________SlangNRox

Rank__Credits____Team
1_____8382804____TeAm AnandTech
2_____4808069____Czech National Team
3_____4039939____SETI.Germany
4_____3719842____Antarctic Crunchers

Wow, @emoga 's on fire! Close to the top 3 other teams, so far. Remember, a lot of WUs may come in slowly.

 

[StefanR5R]

On March 11, results of the Tour de Primes became final.

• All of the data: http://www.primegrid.com/challenge/tdp_2024.php
• Top three in the four contests for jerseys: https://www.primegrid.com/forum_thread.php?id=10427&nowrap=true#170080
• Some more reporting from pschoefer: in German/ auto translated

w a h finished in 3rd for the green jersey, i.e. cumulative prime score.
TeAm AnandTech finished in 8th for prime count and in 4th for cumulative prime score (or 7th and 3rd respectively if we disregard the combined rank of participants without team affiliation).

Edit,
results filtered for TeAm AnandTech only:

• earliest find by Skivelitis2 on 2/1 02:34 UTC,
  latest find by waffleironhead on 2/29 21:35 UTC

• Red Jersey — discoverer of largest prime:
  9,507.9 ...... 10esseeTony
  9,462.1 ...... w a h
  1,541.9 ...... biodoc
  1,539.0 ...... ChelseaOilman
  1,537.6 ...... cellarnoise2
  221.8 ......... [TA]Skillz
  211.5 ......... Orange Kid
  210.6 ......... Letin Noxe
  206.3 ......... waffleironhead
  200.7 ......... kiska
  199.8 ......... Ken_g6
  199.1 ......... johnnevermind
  197.0 ......... Skivelitis2

• Yellow Jersey — prime count leader:
  10 .............. w a h
  8 ................ cellarnoise2
  6 ................ [TA]Skillz
  5 ................ waffleironhead
  2 ................ 10esseeTony
  2 ................ biodoc
  2 ................ ChelseaOilman
  2 ................ Orange Kid
  2 ................ kiska
  2 ................ johnnevermind
  1 ................ Letin Noxe
  1 ................ Ken_g6
  1 ................ Skivelitis2
  (tie breaker is cumulative prime score)

• Green Jersey — points (prime score) leader:
  16,640.4 .... w a h
  9,712.1 ...... 10esseeTony
  5,610.4 ...... cellarnoise2
  3,079.6 ...... biodoc
  1,737.1 ...... ChelseaOilman
  1,309.1 ...... [TA]Skillz
  1,006.8 ...... waffleironhead
  423.0 ......... Orange Kid
  399.3 ......... kiska
  397.6 ......... johnnevermind
  210.6 ......... Letin Noxe
  199.8 ......... Ken_g6
  197.0 ......... Skivelitis2

• Polka Dot Jersey a.k.a. mountain stage, i.e. prime count on 2/29:
  5 ................ wah
  2 ................ [TA]Skillz
  2 ................ waffleironhead
  (tie breaker is cumulative prime score on 2/29)

 

[Skillz]

I didn't do too bad for just 1.5 EPYCs on it most of the month.

 

[Ken g6]

Oh, right, since we're in a challenge, I should post stats.

Day 2 stats:

Rank___Credits____Username
3______6038990____w a h
8______3540113____ChelseaOilman
9______2963140____cellarnoise2
20_____1263024___10esseeTony
31_____934245_____crashtech
39_____735366_____Orange Kid
45_____659885_____Fardringle
51_____595481_____biodoc
52_____591326_____waffleironhead
90_____329456_____mmonnin
114____224660_____Letin Noxe
140____149511_____Ken_g6
146____144971_____Icecold
259____3788_______[TA]Skillz
274____831________SlangNRox

Rank__Credits____Team
1_____18174795___TeAm AnandTech
2_____14463740___Czech National Team
3_____11689464___Antarctic Crunchers
4_____11468467___AMD Users

 

[StefanR5R]

Last 12 hours production, if I copied the figures correctly:
5.13 51.3 M . . . . . . AMD Users
4.69 46.9 M . . . . . . Czech National Team
4.40 44.0 M . . . . . . TeAm AnandTech
3.42 34.2 M . . . . . . Antarctic Crunchers

Edited, see below.

 

[Skillz]

Last 12 hours production, if I copied the figures correctly:
51.3 M . . . . . . AMD Users
46.9 M . . . . . . Czech National Team
44.0 M . . . . . . TeAm AnandTech
34.2 M . . . . . . Antarctic Crunchers

Not sure I follow what you mean by the last 12 hours with those numbers. Since we're in first place with 20.8M.

 

[StefanR5R]

These are the deltas from 3/21 22:00 UTC to 3/22 10:00 UTC. If I calculated them correctly. I took data from https://statsgr.id/top_team_graph.php?challenge_id=2024_2.

Edit: Argh... Divide this by 10.

TAAT: 16.03 M --> 20.43 M (delta: 4.40 M)
AMD Users: 9.04 M --> 14.17 M (delta: 5.13 M)

--------
Edit 2:
Normally I look just at this site, which is a whole lot better: https://www.seti-germany.de/pg_challenge/stats/
But it had a ~24 h downtime until shortly ago, therefore wasn't usable to look up last-half-day production.
SG's site shows thousands separators, Tyler's site does not...

 

[Skillz]

So AMD Users and CNT is out pacing us in the last 12 hours?

 

[StefanR5R]

The speeds of all teams fluctuate a lot. Especially that of AMD Users. Here is another take on last-12h output, from 3/22 01:00 UTC to 13:00 UTC:

TeAm AnandTech ............... 17.01 M ---> 21.91 M ........ +4.90 M
Czech National Team ......... 12.98 M ---> 17.10 M ........ +4.12 M
AMD Users ............................ 9.78 M ---> 14.66 M ........ +4.88 M
Antarctic Crunchers ........... 10.94 M ---> 14.63 M ........ +3.69 M

 

[StefanR5R]

The speeds of all teams fluctuate a lot. Especially that of AMD Users.

...and CNT's too.

 

[Ken g6]

It's easier to see on a graph.



Nobody seems to be approaching us very quickly unless somebody has a bunker.

 

[StefanR5R]

SG's site has got "points per update" graphs, IOW graphs of the speed. But for this challenge these graphs are illegible because of the gap in SG's updates for one day.

From 3/22 10:00 UTC to 22:00 UTC:

TeAm AnandTech ............... 20.62 M ---> 25.55 M ........ +4.93 M
Czech National Team ......... 16.20 M ---> 23.39 M ........ +7.19 M (overtake in 11.5 h)
AMD Users ......................... 14.24 M ---> 18.77 M ........ +4.53 M
Antarctic Crunchers ........... 13.53 M ---> 17.60 M ........ +4.07 M

Time remaining: 53 hours.

Well, I trust that you guys will sort this out. I'm tied up elsewhere. :-)

 

[Skillz]

I'm working on it.

 

[Markfw]

SG's site has got "points per update" graphs, IOW graphs of the speed. But for this challenge these graphs are illegible because of the gap in SG's updates for one day.

From 3/22 10:00 UTC to 22:00 UTC:

TeAm AnandTech ............... 20.62 M ---> 25.55 M ........ +4.93 M
Czech National Team ......... 16.20 M ---> 23.39 M ........ +7.19 M (overtake in 11.5 h)
AMD Users ......................... 14.24 M ---> 18.77 M ........ +4.53 M
Antarctic Crunchers ........... 13.53 M ---> 17.60 M ........ +4.07 M

Time remaining: 53 hours.

Well, I trust that you guys will sort this out. I'm tied up elsewhere. :-)

I added 1040 NON-SMT cores

 

[Ken g6]

Day 3 stats:

Rank___Credits____Username
3______8362189____w a h
9______5508157____ChelseaOilman
11_____4861301____cellarnoise2
25_____1871386___10esseeTony
30_____1696981____crashtech
42_____1156448____Orange Kid
47_____1048879____Fardringle
49_____1032642____markfw
57_____855956_____waffleironhead
59_____826110_____biodoc
83_____596580_____mmonnin
116____371991_____Letin Noxe
140____290710_____Ken_g6
161____218542_____Icecold
226____77282______kiska
253____42989______[TA]Skillz
258____39087______SlangNRox

Rank__Credits____Team
1_____28857240___TeAm AnandTech
2_____25258327___Czech National Team
3_____20914706___Antarctic Crunchers
4_____20847257___AMD Users



I see we're accelerating. 🤞

 

[TennesseeTony]

Oh boy, just looked in the rearview mirror and saw Markfw back there weaving in and out of traffic, moving up fast!

 

[StefanR5R]

From PrimeGrid's challenge thread:

Anyone an idea why an 7900X is slower than most of my 5900X ?

I am seeing the same problem with my 7950X vs 5900X. AVX-512 issue?

This is coming up repeatedly. It's not related to AVX-512 (at least not directly). It is also not a Windows vs. Linux thing, as it has been reported by users of either operating system.

It only happens if the application is run such that it has to access RAM all the time instead of accessing almost only level 3 cache.

In case of Zen CPUs with more than one CCX, such as 5900X, 5950X, 7900X, and 7950X, which all have two CCXs and therefore two level 3 caches (which are not connected with each other, at all), this problem happens if the application threads are freely scheduled by the operating system's scheduler.

(I don't have Zen 3 CPUs, just Zen 2 and Zen 4, therefore I can't reproduce the phenomenon that Zen 4 is hit harder by this problem than Zen 3. Perhaps GWNUM's AVX-512 FFT has worse access patterns to the coefficients of the transform than its FMA3 FFT. This could be tested by booting a Zen 4 CPU with AVX-512 disabled. Or it has something to do with how AMD changed the CCD--IOD link from Zen 3 to Zen 4. Or all the Zen 4 users who perceived a regression vs. Zen 3 use RAM with worse latency than the RAM in the Zen 3 computers to which they compare to, but that doesn't seem plausible to me.)

Solution on Zen server CPUs:

• Declare each last-level cache domain as a NUMA node by means of the respective ACPI option in the BIOS.

Linux's thread scheduler tries to concentrate all threads of a process on the same NUMA node as much as possible. I don't know about Windows' scheduler.

Solutions on any Zen CPUs:

• Launch as many BOINC client instances in parallel as there are CCXs. Bind each BOINC instance to the logical CPUs of one exclusive CCX. The CPU affinities of the BOINC processes will be inherited by all subsequently started science tasks.

• Or leave BOINC alone and only set CPU affinities of the science tasks. This is typically implemented by having a background program (can be a tiny shell script or tiny powershell script, or a more extensive program depending on desired extra features) which periodically looks if there are any science tasks around, and sets their CPU affinities.

 

[Fardringle]

Oh boy, just looked in the rearview mirror and saw Markfw back there weaving in and out of traffic, moving up fast!

I didn't see any weaving. It's closer to something like this:

âï¸ Kiwi Snow Plow

37 seconds · Clipped by Michael Bird · Original video "5 POWERFUL Snow Plow Trains Videos" by Engineering World

youtube.com

 

[Ken g6]

Day 4 stats:

Rank___Credits____Username
2______19426735___w a h
10_____7509746____ChelseaOilman
11_____6508238____cellarnoise2
14_____5107249____markfw
17_____4146363____[TA]Skillz
23_____2884917___10esseeTony
26_____2651593____crashtech
47_____1583243____Orange Kid
52_____1461128____Fardringle
62_____1140247____biodoc
64_____1124031____waffleironhead
98_____677992_____mmonnin
125____489394_____Letin Noxe
148____364515_____Ken_g6
178____256711_____Icecold
242____111686_____kiska
293____39087______SlangNRox

Rank__Credits____Team
1_____55482882___TeAm AnandTech
2_____36910622___Czech National Team
3_____30953406___AMD Users
4_____29015170___Antarctic Crunchers

I'd post a graph but the forums are too slow tonight.