Info PrimeGrid Challenges 2024, sieve-free edition

[Ken g6]

Oh, right, I don't need to wait for this stats run.

More-or-less final stats:

Rank___Credits____Username
1______31340806___w a h
8______19346258___crashtech
11_____15008760___cellarnoise2
16_____13014411___ChelseaOilman
30_____4739571____[TA]Skillz
37_____3955065____markfw
39_____3887667____mmonnin
46_____3257185____Orange Kid
55_____2888664____waffleironhead
63_____2489396____Fardringle
107____1333386____Letin Noxe
124____1090012____Pokey
138____939350_____kiska
151____854251_____Ken_g6
168____633214_____Icecold
258____270321_____johnnevermind

Rank__Credits____Team
1_____106720558___TeAm AnandTech
2_____75614104___Czech National Team
3_____72306810___Antarctic Crunchers
4_____55679571___SETI.Germany

 

[TennesseeTony]

(upcoming) GFN-19 app_config.xml

1 task, 1 sample, 4090 - 1318 sec, 430 sec CPU
2 tasks, 2 samples, 4090 - 2020 sec, 536 sec CPU
3 tasks, 3 samples, 4090 - 2870 sec, 760 sec CPU, but wide variation, perhaps 2 tasks is best? Both 2 and 3 concurrent tasks yielded about 68% GPU usage.

<app_config>
<app>
<name>genefer19</name>
<gpu_versions>
<gpu_usage>.5</gpu_usage>
<cpu_usage>.5</cpu_usage>
</gpu_versions>
</app>
 </app_config>

 

[TennesseeTony]

What is this Primorial project coming up in August?

 

[Ken g6]

What is this Primorial project coming up in August?

Primorial Sieve on GFN Server

It's searching for primes near factorials or primorials (plus or minus one). A factorial is a number multiplied by every whole number below it. (3 factorial, written 3!, is 3*2*1 = 6.) Primorial is like that but multiplying by only primes.

Sieving for the project is going on now on the Private GFN Server.

 

[StefanR5R]

Note that the (as yet still tentative) August challenge project won't be the sieve project, but is a planned prime search project which is a continuation of two discontinued PRPNET projects: Factorial Prime Search (FPS), Primorial Prime Search (PRS).

PrimeGrid won't just put existing PRPNET's FPS/PRS applications into a BOINC wrapper; they want to modernize it all around. Last news about the start date of this new PrimeGrid project, in the message board at least, is from April 2023, saying that it will start when ready.

AFAIU it will be a CPU-only project.

 

[TennesseeTony]

Ok, thanks, the Primorial sub-project does not yet exist to the public.

That said, we have the GFN-19/PG's 19th birthday starting very soon, less than 3 hours.

 

[Markfw]

Ok, thanks, the Primorial sub-project does not yet exist to the public.

That said, we have the GFN-19/PG's 19th birthday starting very soon, less than 3 hours.

Have you edited your prefs to run gfn-19 ? You had t be running something else all week with those high scores.

I just finished my configs. Some running, since they only run one at a time. And the video units finish in 15 minutes, no worth pausing and restarting.

 

[Ken g6]

And we're off!

 

[Markfw]

delete

 

[Ken g6]

Day 1 stats:

Rank___Credits____Username
3______8424826____crashtech
4______8351907____[TA]Skillz
7______7234700____markfw
12_____4432520____w a h
23_____2328886___10esseeTony
25_____2233564____cellarnoise2
36_____1539682____mmonnin
39_____1424594____Orange Kid
42_____1348815____Icecold
48_____1247251____ChelseaOilman
65_____865167_____biodoc
88_____613679_____waffleironhead
101____531459_____Fardringle
148____284636_____Letin Noxe
150____275474_____johnnevermind
189____138930_____Ken_g6

Rank__Credits____Team
1_____41276096___TeAm AnandTech
2_____33570455___Antarctic Crunchers
3_____23742441___Czech National Team
4_____22865665___SETI.Germany

 

[crashtech]

We remain the team to beat, it seems!

 

[StefanR5R]

GPUs and CPUs can be used, the former tending to be more power-efficient at it.

Whole machine throughputs and power consumptions "at the wall":

• Intel Kabylake w/o turbo boost + Nvidia RTX 4090 at 300 W board power limit, running only GPU tasks on this host:
  950,000 PPD / 350 W = 2,700 PPD/W
• AMD Epyc 9554P at 400 W PPT limit:
  1,200,000 PPD / 500 W = 2,400 PPD/W

Closer than I thought.

 

[Markfw]

Whole machine throughputs and power consumptions "at the wall":

• Intel Kabylake w/o turbo boost + Nvidia RTX 4090 at 300 W board power limit, running only GPU tasks on this host:
  950,000 PPD / 350 W = 2,700 PPD/W
• AMD Epyc 9554P at 400 W PPT limit:
  1,200,000 PPD / 500 W = 2,400 PPD/W

Closer than I thought.

I don't think that my experience. The 9554's do 8 units at a time in 1.5 hours. The 4090 on the box that I am typing on uses 414 watts JUST for the 4090 at 13 min per units.

The 9554's are 360 watt, and 8k per unit and 64k in 1.5 hours. The 4090 does a little over 6 units in that time (same 8k credit) 48k in 1.5 hours. So the 9554 wins. Is my math good ?

 

[StefanR5R]

Differences of my setups to yours:

• I powerlimited the 4090 to 300 W. Tasks take less than 13 minutes nevertheless according to the average over the last 24 hours which is listed on the PrimeGrid account web page. I haven't tested how fast it would be without powerlimit.
• The KBL+4090 box runs Linux Mint, and there is nothing running besides BOINC and the idle Cinnamon desktop.

• I increased the powerlimit of the 9554P from default 360 W to 400 W. If I reduced the power limit instead, I could improve power efficiency, and then power efficiency of this EPYC box would perhaps match that of the KBL+4090 box.
• I run 16 tasks at once on the EPYC. That's because the 9554 has got 8 CCXs with 32 MB L3 cache each, and GFN-19 wants just a little bit more than 10 MB cache per task. Of course I ensure cache-aligned task scheduling with a script which sets CPU affinities.
• I use 8 threads per task on the EPYC (thus, 128 threads total). Could as well use 4 threads per task instead though (i.e. 64 threads total) because that would be almost as fast. I haven't checked power consumption when I tested this; maybe I should. Could be another possibility to increase power efficiency of the 9554P. I'll check this another time.

 

[crashtech]

I'd caution against using SMT for those also running GPU tasks on the same computer, doing so absolutely decimates the PPD of the CPU tasks in my Zen4 testing.

 

[StefanR5R]

Every time when I read forum posts — here, or more so elsewhere — about performance optimization issues with concurrent CPU and GPU loads, I silently congratulate myself for my old decision to keep these two workload types on separate hosts, which are specialized on one or the other workload type. Costs me a bit of shelf space, but spares me of all the associated worries. (And hosts which are good at one thing are easier to build than ones which are good at two.)

PS,
there have been moments when I thought of throwing out the Z270 mobos of my GPU hosts and put AM5 boards in. This is a good reminder to refrain from this.

 

[Ken g6]

Day 2.2 stats:

Rank___Credits____Username
4______20143308___crashtech
6______19743376___[TA]Skillz
7______16849762___markfw
17_____7404745____cellarnoise2
20_____6660194___10esseeTony
24_____5716989____w a h
37_____3552286____Icecold
38_____3417880____mmonnin
39_____3210012____ChelseaOilman
44_____2959250____Orange Kid
69_____1906145____biodoc
97_____1309349____Fardringle
100____1287722____waffleironhead
132____786983_____johnnevermind
161____537576_____Letin Noxe
197____303552_____Ken_g6

Rank__Credits____Team
1_____95789136___TeAm AnandTech
2_____68037179___Antarctic Crunchers
3_____56261883___Czech National Team
4_____52424322___SETI.Germany

I meant to post this before going to the dentist, but I forgot, and that's the tooth: 🦷

 

[Skillz]

Dang it, @crashtech passed me. Guess I need to put all my CPUs in now.

 

[Ken g6]

Day 3 stats:

Rank___Credits____Username
4______29592870___crashtech
6______27925062___[TA]Skillz
7______24053836___markfw
17_____10515021___10esseeTony
18_____10341040___cellarnoise2
29_____6661052____w a h
37_____5396802____Icecold
38_____4827020____mmonnin
39_____4601007____ChelseaOilman
45_____4214906____Orange Kid
66_____2762405____biodoc
92_____1906792____Fardringle
102____1747813____waffleironhead
133____1167293____johnnevermind
154____824923_____Letin Noxe
207____417779_____Ken_g6

Rank__Credits____Team
1_____136955626___TeAm AnandTech
2_____95245507___Antarctic Crunchers
3_____80360179___Czech National Team
4_____75637164___SETI.Germany

 

[Ken g6]

Day 4 stats:

Rank___Credits____Username
4______39902316___crashtech
6______33646099___[TA]Skillz
7______31854922___markfw
14_____14949979___cellarnoise2
19_____12911474___10esseeTony
34_____7702542____w a h
37_____6958788____Icecold
39_____6368740____mmonnin
41_____6131050____ChelseaOilman
46_____5590527____Orange Kid
63_____4011426____biodoc
94_____2514499____Fardringle
106____2266481____waffleironhead
132____1614681____johnnevermind
158____1053674____Letin Noxe
210____556685_____Ken_g6
311____128532_____kiska

Rank__Credits____Team
1_____178162424___TeAm AnandTech
2_____125521596___Antarctic Crunchers
3_____105985977___Czech National Team
4_____105222645___SETI.Germany

Hey, somebody actually showed up behind me.

 

[Ken g6]

Day 5.2 stats:

Rank___Credits____Username
3______52785390___crashtech
6______41666846___markfw
8______35497372___[TA]Skillz
14_____20278157___cellarnoise2
20_____15467145___10esseeTony
37_____8962171____w a h
38_____8507088____Icecold
39_____8284835____mmonnin
43_____7610633____ChelseaOilman
45_____7267677____Orange Kid
73_____4101663____biodoc
93_____3370131____Fardringle
106____2877320____waffleironhead
130____2166990____johnnevermind
176____1132561____Letin Noxe
214____721126_____Ken_g6
266____414630_____kiska

Rank__Credits____Team
1_____221111742___TeAm AnandTech
2_____162105305___Antarctic Crunchers
3_____140775898___SETI.Germany
4_____137477701___Czech National Team

No excuses today; I just forgot.

 

[crashtech]

Thanks for doing it at all, Ken, we appreciate it!

 

[Ken g6]

Day 6 stats:

Rank___Credits____Username
3______62083369___crashtech
6______48275008___markfw
8______36117783___[TA]Skillz
13_____23609384___cellarnoise2
24_____16380045___10esseeTony
38_____9677284____w a h
39_____9632278____mmonnin
40_____9629687____Icecold
43_____8723252____ChelseaOilman
46_____8574642____Orange Kid
89_____4101663____biodoc
90_____4037106____Fardringle
114____3092253____waffleironhead
132____2465182____johnnevermind
184____1175690____Letin Noxe
219____835863_____Ken_g6
243____663485_____kiska

Rank__Credits____Team
1_____249073983___TeAm AnandTech
2_____188144153___Antarctic Crunchers
3_____166417818___SETI.Germany
4_____160513745___Czech National Team

 

[StefanR5R]

PDW joined the Birthday party, late but effective:
https://statsgr.id/top_team_graph.php?challenge_id=2024_4
:-D

 

[Ken g6]

More or less final stats:

Rank___Credits____Username
4______70539612___crashtech
6______55662160___markfw
10_____36901076___[TA]Skillz
13_____27234891___cellarnoise2
26_____18427574___10esseeTony
39_____10783646___Icecold
41_____10543347___mmonnin
43_____9944771____Orange Kid
45_____9835636____ChelseaOilman
49_____9677284____w a h
94_____4578906____Fardringle
103____4101663____biodoc
120____3340669____waffleironhead
132____2790835____johnnevermind
190____1252380____Letin Noxe
220____993494_____kiska
221____990941_____Ken_g6

Rank__Credits____Team
1_____277598894___TeAm AnandTech
2_____215290781___Antarctic Crunchers
3_____192526089___SETI.Germany
4_____185501425___Czech National Team

@Kiska beat me by half a WU. I discovered with my last WU that I could have run them faster, too.