PrimeGrid Challenges 2021

[emoga]

I went ahead and rented a 3090 just to see. Looks like 38 seconds in Linux without any CPU units running.
For reference, my 2080 in Linux averaged 54 seconds over the last 24 hours and a 2080Ti is around 43 seconds.

So the one that beat me could definitely improve its timings if they configured it better. It also ran windows too I suppose.

 

[lane42]

Maybe i saw a 3090 under windows vs. a 2080ti under Linux ?
Ill try to find it but don't think i will.
edit, the best i saw was a 2080 super in 54 seconds under windows.

 

[lane42]

Congrats emoga on a GFN 16
Tour de Primes 2021 (primegrid.com)

 

[emoga]

What's everyone's plans for the rest of the month?

Switching over to try for a MEGA now that most of us have a badge? or going for as many T5K's as possible?

It's a long shot, but I think I'll go GFN17 MEGA's on GPU's and MEGA's on CPUs and hope the math gods hear my prayers.

 

[Ken g6]

Congrats to @TennesseeTony on finding a prime!

Edit: That's weird, I don't see the prime any longer. Was I looking at last year's stats?

What's everyone's plans for the rest of the month?

Switching over to try for a MEGA now that most of us have a badge? or going for as many T5K's as possible?

It's a long shot, but I think I'll go GFN17 MEGA's on GPU's and MEGA's on CPUs and hope the math gods hear my prayers.View attachment 39362

I'm doing that too. You can calculate your odds of getting a MEGA, for instance. Michael Goetz said in last year's thread that about one in 30,000 tests yields a prime. But he doesn't spread that around because half the people who'd done 30,000 tasks would ask, "why haven't I found a prime yet?!". If you test TPD tests per day, and there are D days to go, the formula is:

1-(29999/30000)^(TPD*D)

So today there are ~20 days to go, and I'm testing about 715 tests per day. As a result I have a 38% chance of finding a MEGA. And falling! I'd better get on with it!!!

 

[lane42]

Tennessee Tony

TeAm AnandTech

BOINC: GFN-16

10657984465536+1

526,102​

165.19734519427​

Edit: That's weird, I don't see the prime any longer.

Tony's a Good Guy, He can have mine

 

[emoga]

Looks like 2 3 double-checks.
Don't worry @TennesseeTony, you'll get one

 

[TennesseeTony]

My queue is too high to be first, I just need the heat right now. I suppose I could adjust the queue via PG preferences...

 

[Ken g6]

My queue is too high to be first, I just need the heat right now. I suppose I could adjust the queue via PG preferences...

I guess you did. Congrats!

 

[lane42]

Congrats TT Tour de Primes 2021 (primegrid.com)

 

[TennesseeTony]

 

[TennesseeTony]

Just gotta catch @Tejas_2 now....

 

[Markfw]

I should try my new 5950x on the next challenge....

 

[StefanR5R]

The next team challenge at PrimeGrid will be on March 14 – March 24, 12:00 UTC, at the "Seventeen or Bust" subproject.

This is a CPU-only subproject which is checking very large candidates for primality: Currently, its candidates have 10 million digits. Only the PrimeGrid subprojects GFN-21, GFN-22, and GFN-DYFL ( = GFN-22 at a higher search range) look at larger candidates, and these are GPU subprojects. (OK, GFN-21 also has a CPU application, FWIW.)

I looked up the current search range of Seventeen or Bust and checked its FFT sizes on a Haswell CPU. Haswell and all later Intel CPUs which don't implement AVX-512, and several AMD CPU architectures including Zen 2 and certainly Zen 3, will use the same FFT implementation here, which is "all-complex FMA3 FFT".

SoB-LLR candidates	current FMA3 FFT lengths	last-level cache size demand
21181*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
22699*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
24737*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
55459*2^n+1​	3M … 3200K​	24 MB … 25 MB​
67607*2^n+1​	3M … 3200K​	24 MB … 25 MB​

Zen 2 will perform poorly at this, because its level-3 cache is divided into 16 MB segments. (In APUs, it's much less.) Zen 2 will therefore have to perform a lot of main memory accesses and will consequently be unable to keep its FMA units busy.

Zen 3 has got 32 MB level-3 cache segments, one per each compute die. (Zen 3 based APUs will have less.) The Zen 3 CPUs should therefore perform very well in the SoB-LLR subproject, provided that there is exactly one LLR program instance running per each compute die. (That is, one SoB task at a time on 5600X and 5800X, and two tasks at a time on 5900X and 5950X. Thread count per task should be 6 or maybe 12 on 5600X and 5900X, and 8 or maybe 16 on 5900X and 5950X.)

My old 22-core Broadwell-EP CPUs have 55 MB unified L3-cache and will therefore perform OK with 2 simultaneous SoB tasks per socket.

 

[lane42]

10esseeTony

TeAm AnandTech

BOINC: GFN-16

11321987665536+1

527,822​

166.86193287378​

Saturday 13th of February 2021 03:06:38 UTC

Tony's 2nd, congrats

 

[Ken g6]

Congrats, @emoga, you got a TDP mountain stage prime!

 

[emoga]

Thanks, I almost fell out of my chair when I saw that

 

[lane42]

emoga

TeAm AnandTech

BOINC: PPS-MEGA

1275·23356045+1

1,010,274​

1224.2881811697​

Tuesday 23rd of February 2021 18:19:41 UTC

#3 for emoga, Congrats

 

[lane42]

10esseeTony

TeAm AnandTech

BOINC: GFN-16

12114045865536+1

529,747​

168.73767500626​

Thursday 25th of February 2021 17:17:42 UTC

#3 for Tony, Congrats

 

[Ken g6]

​

10esseeTony

TeAm AnandTech

BOINC: GFN-16

12114045865536+1

529,747​

168.73767500626​

Thursday 25th of February 2021 17:17:42 UTC

#3 for Tony, Congrats

And two more!

121857822^65536+1
121974060^65536+1

 

[lane42]

​

User	Count of Primes	Score

​

10esseeTony

5​

840.16696339676​

​

emoga

3​

1523.20276944192​

​

Tejas

1​

813.13781107928​

​

Ken_g6

1​

130.34367454003​

​

Lane42

1​

165.19734519427​

 

[StefanR5R]

pschoefer made a graph of which types of primes were found when:
post in the SETI.Germany forum

 

[StefanR5R]

I looked up the current search range of Seventeen or Bust and checked its FFT sizes on a Haswell CPU. Haswell and all later Intel CPUs which don't implement AVX-512, and several AMD CPU architectures including Zen 2 and certainly Zen 3, will use the same FFT implementation here, which is "all-complex FMA3 FFT".

SoB-LLR candidates	current FMA3 FFT lengths	last-level cache size demand
21181*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
22699*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
24737*2^n+1​	2880K … 3M​	22.5 MB … 24 MB​
55459*2^n+1​	3M … 3200K​	24 MB … 25 MB​
67607*2^n+1​	3M … 3200K​	24 MB … 25 MB​

I checked again now, and the FFT lengths are still the same, as to be expected.

What's weird is that my computers used sllr2_1.1.0_linux64_201114 during the January challenge at PPS-DIV, but one test computer which fetched two SoB-LLR tasks today to test the waters downloaded and uses sllr2_1.0.0_linux64_200814 instead for SoB.

 

[crashtech]

I plan to participate in the Serpinski's Birthday Challenge from the 14th until the 19th, when the FB Sprint starts, I'll need to come up with some optimal configs by then.
I guess SMT off, all physical cores on one task for most CPUs.

 

[StefanR5R]

The two test tasks were completed by a single-socket E5-2696 v4 @ 2.3 GHz. (AVX turbo was supposed to be 2.6 GHz. I'll need to check the BIOS settings. They always get lost during a longer downtime, probably because the battery cell is depleted. But the cell sits hidden beneath 3 watercooled GPUs. Anyway, even with the 0.3 GHz deficit, the CPU still drew a lot of power.)

Run time ~1 day,
CPU time 247 h for a task with 3M FFT length and 74k credit,
CPU time 265 h for a task with 3200K FFT length and 79k credit.

The "edit PrimeGrid project preferences" page says that recent average CPU time at SoB is 382 h. — That is, these olden but golden Broadwell-EPs still fare very well at it in comparison to the average PrimeGrid contributors. The large cache of the high-core-count models certainly helps.

Next step is to think up some tests with Zen 2 CPUs.