PrimeGrid Challenges 2021

Page 19 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,218
3,796
75
Day 5.3-ish stats:

Rank___Credits____Username
1______28645332___Skillz
4______13647844___xii5ku
6______8232449____crashtech
9______7415573____Icecold
11_____5976705____parsnip soup in a clay
26_____2896199____markfw
32_____2405237____emoga
33_____2289809____biodoc
45_____1572053____Orange Kid
94_____699710_____Skivelitis2
114____499418_____Ken_g6
126____413819_____Lane42
165____251485_____HutchinsonJC
213____129555_____Fardringle
233____84887______Howdy
235____84317______Kiska
250____79952______SlangNRox
298____5105_______thigobr
320____2275_______cellarnoise2

Rank__Credits____Team
1_____75331734___TeAm AnandTech
2_____57487090___Antarctic Crunchers
3_____41876045___SETI.Germany
4_____32741502___Czech National Team

We have a decent lead at the halfway point. And nineteen participants! :)
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,218
3,796
75
Day 6 stats:

Rank___Credits____Username
1______34903036___Skillz
4______15775883___xii5ku
7______9337612____crashtech
10_____8403836____Icecold
11_____7752569____parsnip soup in a clay
25_____3831825____markfw
29_____2851323____emoga
34_____2668479____biodoc
45_____1872200____Orange Kid
92_____784868_____Skivelitis2
116____542159_____Ken_g6
123____499912_____Lane42
136____424533_____HutchinsonJC
195____212180_____Fardringle
238____102550_____thigobr
245____84887______Howdy
250____84317______Kiska
263____81170______SlangNRox
267____47685______cellarnoise2

Rank__Credits____Team
1_____90261035___TeAm AnandTech
2_____64516149___Antarctic Crunchers
3_____51534641___SETI.Germany
4_____37466812___Czech National Team

We're building quite a lead now!
 
  • Like
Reactions: Markfw

cellarnoise

Senior member
Mar 22, 2017
709
394
136
Note: Running 2 tasks at 8 CPU cores is showing at least twice as fast at 2 tasks at 16 cores on my 5950x.

I started out at 2 tasks at 16 CPUs and have now switched!
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
Note: Running 2 tasks at 8 CPU cores is showing at least twice as fast at 2 tasks at 16 cores on my 5950x.

I started out at 2 tasks at 16 CPUs and have now switched!
@Skillz , if I change this, then all 10 of my hosts will need to be edited, and setup for my "Home" location will need to be changed to 8. Is it worth it for me ? I am now running over one million ppd.
 

cellarnoise

Senior member
Mar 22, 2017
709
394
136
@Skillz , if I change this, then all 10 of my hosts will need to be edited, and setup for my "Home" location will need to be changed to 8. Is it worth it for me ? I am now running over one million ppd.

Mark, note I changed my settings on the PrimeGrid preference page while I still had two 2 task / 16 core tasks running. When one of the 16 core tasks completed, it downloaded an 8 core task and as of now I have one 16 core task and one 8 core tasks. So there is no need to abort any 16 core tasks that are currently running.

It looks like both the 16 core and the 8 core tasks are running faster than two 16 core tasks. The 16 core task has sped up.
 
Last edited:

Skillz

Senior member
Feb 14, 2014
910
927
136
@Skillz , if I change this, then all 10 of my hosts will need to be edited, and setup for my "Home" location will need to be changed to 8. Is it worth it for me ? I am now running over one million ppd.

Realistically, no. I think we might be pretty safe from falling from 1st for now. Though, I am not 100% confident on that. AC could easily have 1000s of tasks waiting to drop at the end. No way of really knowing for sure. I intend to run till the very last day and will begin shutting things down on that day as tasks complete.

With that being said; switching them over to 2 tasks, 8 threads(cores) per task wouldn't be all that difficult. You can do it all entirely on the PG.com web site. As long as you had your "extra cache" settings setup for .01 so each host only downloaded the tasks they can run immediately and nothing is waiting in queue.
 

cellarnoise

Senior member
Mar 22, 2017
709
394
136
I think I am beginning to figure out how to optimize for this challenge!

Give me another day or two and I will have this one cranking! :(

I really messed up two days ago while trying to undervolt for the same watts and lost 10 hours or so of work... twice..... It is an addiction that I need to work on... Starving CPU's that is...

Running better and better now without playing with voltages or the MegaHurtz...

Oh, (edit), it is finally :) cooling off here and not so smoky the past day. I am ready for some rain and the fall season!
 
  • Like
Reactions: Icecold and Markfw

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,716
136
Note: Running 2 tasks at 8 CPU cores is showing at least twice as fast at 2 tasks at 16 cores on my 5950x.

I started out at 2 tasks at 16 CPUs and have now switched!
Do you have a Windows host or a Linux host?

From what I remember, measurements with a fixed PSP-LLR workunit on a Linux 5950X computer showed a certain advantage for 2x 8-threaded tasks over 2x 16-threaded tasks, but the advantage was not as big as you report. (Tests with a fixed workunit are worthwhile because they are repeatable and eliminate the influence of size differences (and credit differences) between random workunits.)

@Skillz , if I change this, then all 10 of my hosts will need to be edited, and setup for my "Home" location will need to be changed to 8. Is it worth it for me ? I am now running over one million ppd.
@cellarnoise and @Skillz already responded, but for completeness I add that it is even possible to change the thread count of an already running task: Suspend the task (to disk, not to RAM¹), let the boinc client read the new setting (if using the web preferences, not a local app_config.xml: trigger a project update in the client), resume the task. After that the client will likely still display the old thread count, but using a process viewer like the top console command — or viewing the respective "slots/*/stderr.txt" file — will show that the task is now running with the new setting.

¹) This is controlled by the computing preferences: "Leave tasks in memory while suspended?" = "Yes" lets suspended tasks stick around in RAM with their complete process state. "No" will cause tasks, when they are requested to suspend, to save their state to disk and then to exit (but not finish). When such tasks are resumed, they not only re-read their former state from disk, they also re-read settings from the boinc client which may have been modified while the task was suspended.
 
Last edited:

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,218
3,796
75
Day 7 stats:

Rank___Credits____Username
2______39958516___Skillz
4______19928897___xii5ku
9______11018639___crashtech
11_____9891638____parsnip soup in a clay
12_____9712409____Icecold
22_____5268713____markfw
29_____3463138____emoga
33_____3280484____biodoc
48_____2165914____Orange Kid
93_____954230_____Skivelitis2
118____667914_____Ken_g6
127____605979_____HutchinsonJC
147____499912_____Lane42
174____379192_____Fardringle
192____298088_____thigobr
221____185802_____cellarnoise2
238____130746_____SlangNRox
264____84887______Howdy
268____84317______Kiska

Rank__Credits____Team
1_____108579423___TeAm AnandTech
2_____78416213___Antarctic Crunchers
3_____66919073___SETI.Germany
4_____47006665___AMD Users

Fun to note: Even without @Skillz we'd still be ranked #2. This really is a TeAm effort! :)
 
  • Like
Reactions: lane42 and Markfw

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
We now have a 30.3 million point lead. Looking good !

They seem to be matching our production, only HOLDING the 30.3 million lead, not gaining more at the moment.

Edit: gaining slowly again, 30.49 million
 
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,716
136
At some of the past PG races, I entered the race late so that I could overtake a lot of people.
Somebody stole this sophisticated tactic and is turning it against myself now. >:-(

;-)
 
  • Haha
Reactions: Kiska and Icecold

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
16,218
3,796
75
Day 8 stats:

Rank___Credits____Username
2______46290758___Skillz
5______23003865___xii5ku
8______13061789___parsnip soup in a clay
11_____12590455___crashtech
12_____11150271___Icecold
22_____6510878____markfw
30_____3924563____emoga
32_____3840374____biodoc
48_____2492613____Orange Kid
93_____1122544____Skivelitis2
118____800375_____Ken_g6
120____790979_____HutchinsonJC
167____499912_____Lane42
175____459239_____Fardringle
186____384682_____thigobr
204____309279_____cellarnoise2
233____172783_____SlangNRox
263____124477_____Kiska
277____84887______Howdy

Rank__Credits____Team
1_____127614730___TeAm AnandTech
2_____89775563___Antarctic Crunchers
3_____78677763___SETI.Germany
4_____57105409___AMD Users

Our lead is pretty steady. :)
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
37.8 million point lead. Yup, consistant.

7 pm PST 40 million plus lead with 2 days to go !
 
Last edited:

Fardringle

Diamond Member
Oct 23, 2000
9,184
753
126
46.5 million lead now.

I'm glad you talked me into joining. Without my overwhelming power, the lead would only be 46.0 million, and that's just unacceptable.. :D
 

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
They appear to be stuck at about 48.5 million points below us, so either we are shutting things off (not me) or they are ramping up. But for 14 hours to go, thats a lot of points to make up.
 

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,716
136
Number of participating members in the top twenty teams:
........ 19 ... TeAm AnandTech
........ 20 ... Antarctic Crunchers
........ 40 ... SETI.Germany
.......... 7 ... AMD Users
........ 50 ... Czech National Team
.......... 1 ... Ural Federal University
........ 11 ... Aggie The Pew
.......... 2 ... BOINC@MIXI
.......... 4 ... Storm
........ 10 ... BOINC@AUSTRALIA
.......... 2 ... Universe Examiners
.......... 8 ... SETI.USA
.......... 4 ... The Knights Who Say Ni!
.......... 6 ... Sicituradastra.
.......... 7 ... Crunching@EVGA
........ 17 ... Team 2ch
........ 10 ... L'Alliance Francophone
........ 10 ... BOINC@Poland
.......... 9 ... Rechenkraft.net
.......... 6 ... BOINCstats

In the previous PrimeGrid challenges of the 2021 season, this is how many TeAm mates went in, and which rank the TeAm achieved:
... 17 members ... 5th place ... January: Good Riddance 2020! Challenge, PPS-DIV (CPUs)
... 12 members ... 7th place ... March: Sier"pi"nski's Birthday Challenge, SOB-LLR (CPUs)
... 11 members ... 9th place ... April: Yuri's Night Challenge, WW (GPUs and CPUs)
... 11 members ... 1st place ... June: PrimeGrid's 16th Birthday Challenge, ESP-LLR (CPUs)
... 13 members ... 1st place ... July: World Emoji Day Challenge, GFN-17-low (CPUs and GPUs)

Remaining PrimeGrid challenges in the 2021 season:
... October: Martin Gardner's Birthday Challenge, GCW-LLR (CPUs)
... November: Euler's Constant Challenge, AP27 (GPUs, and perhaps many-core CPUs)
... December: Geminids Shower Challenge, GFN-21/-22/DYFL (GPUs, and perhaps many-core CPUs)
 
  • Like
Reactions: Ken g6

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,716
136
it is even possible to change the thread count of an already running task: Suspend the task (to disk, not to RAM¹), let the boinc client read the new setting (if using the web preferences, not a local app_config.xml: trigger a project update in the client), resume the task.
Looks like while this works if you have an app_config.xml locally, it doesn't if the thread count was configured via the web preferences instead. In this case, shut down the client, edit the <cmdline> tag within the respective <app_version> block of the local client_state.xml file, restart the client.
 

Skillz

Senior member
Feb 14, 2014
910
927
136
RankTeamCredit
1TeAm AnandTech
168,187,145.77​
2Antarctic Crunchers
111,136,492.32​
3SETI.Germany
98,454,372.62​
4AMD Users
74,282,070.62​
5Czech National Team
70,509,782.32​
6Ural Federal University
38,615,552.63​
7Aggie The Pew
30,764,979.69​
8BOINC@MIXI
14,794,120.26​
9Storm
14,522,595.17​
10BOINC@AUSTRALIA
13,610,331.62​


Congrats everyone involved. We took 1st place.

This is our third, 1st place finish in a row this year. Let's keep up the progress!

# RankUserTeamCredit
1 2SkillzTeAm AnandTech
55,684,440.72​
2 5xii5kuTeAm AnandTech
37,554,370.08​
3 7parsnip soup in a clay bowlTeAm AnandTech
18,103,901.68​
4 11crashtechTeAm AnandTech
15,881,337.25​
5 13IcecoldTeAm AnandTech
12,510,385.03​
6 20markfwTeAm AnandTech
9,093,438.15​
7 32emogaTeAm AnandTech
5,133,318.71​
8 34biodocTeAm AnandTech
4,913,476.87​
9 49Orange KidTeAm AnandTech
3,073,123.58​
10 97Skivelitis2TeAm AnandTech
1,348,149.27​
11 111HutchinsonJCTeAm AnandTech
1,171,093.90​
12 116Ken_g6TeAm AnandTech
1,072,463.98​
13 152FardringleTeAm AnandTech
711,971.62​
14 178cellarnoise2TeAm AnandTech
541,459.08​
15 184thigobrTeAm AnandTech
512,155.35​
16 188Lane42TeAm AnandTech
499,912.37​
17 253SlangNRoxTeAm AnandTech
172,783.05​
18 280KiskaTeAm AnandTech
124,477.28​
19 296HowdyTeAm AnandTech
84,887.81​
 
Last edited:

StefanR5R

Elite Member
Dec 10, 2016
5,459
7,716
136
Hmm, https://www.primegrid.com/apps.php is showing this:

Generalized Cullen/Woodall (LLR) -- "llrGCW"
PlatformVersionInstallation time
Microsoft Windows (Vista or later) running on an Intel x86-compatible CPU9.01 (mt)7 Dec 2020 | 20:22:53 UTC
Microsoft Windows (Vista or later) running on an AMD x86_64 or Intel EM64T CPU9.02 (mt)9 Sep 2021 | 9:18:29 UTC
Linux running on an Intel x86-compatible CPU9.01 (mt)7 Dec 2020 | 20:22:55 UTC
Linux running on an AMD x86_64 or Intel EM64T CPU9.02 (mt)9 Sep 2021 | 9:18:32 UTC
Mac OS 10.14+ running on an Intel 64-bit CPU9.01 (mt)7 Dec 2020 | 20:22:51 UTC

I don't see a mention of the September 9 update in the GCW prime search forum. — Oh wait, it is discussed in the "Number crunching" forum, "LLR2 installed on all big LLR projects" thread:
On September 9 stream said:
An update for GCW and SR5 projects.

New LLR2 (application version 9.02) is installed on these projects (major OS only - Linux/Windows 64-bit). It's built with more recent GWNUM version (30.x). For PrimeGrid users, the only noticeable difference is a new fast checkpoint loading code.

In old version, file loading is just awfully slow for tests where b<>2. It affects GCW and SR5 subprojects. A compression of checkpoints could lasts for many minutes, where compression itself takes just a few seconds, and rest was spent converting data from on-disk to in-memory formats. We worked around this problem as much as possible caching checkpoint data in memory, but it worked only if a whole test was finished uninterrupted. In new GWNUM, this problem was fixed and loading of files is now fast enough for any base. A compression of checkpoints at the end of test should be significantly faster even if test progress was interrupted.

This version already received sufficient testing on GFN server (it was used on GFN-MEGA and "LLR2 testing" for many months). Anyway, please report if you've noticed something unexpected or suspicious.


Furthermore, the "Edit PrimeGrid preferences" page claims:
"This app uses a lot of memory. It requires about 1.5 GB of swap space / virtual memory, and 750 MB of physical memory. If you are using Digital Ocean or TheScience.cloud 3-core/1GB droplets, you can only run one task at a time."
I haven't checked yet whether this is true with the up-to-date application version.
Also,
"Recent average CPU time: 154:13:19
This project has a 10% long job credit bonus."

Edit,
I tried a Linux x86-64 client just now and it downloaded this binary:
XML:
<name>sllr2_1.1.1_linux64_210204</name>
<download_url>http://www.primegrid.com/download/sllr2_1.1.1_linux64_210204.gz</download_url>

Edit,
a random task with zero-padded FMA3 FFT length of 2304K is taking ~650 MB virtual and ~540 MB resident memory at 1 % completion. I expect the footprint to grow further during the run time.

Estimated time to complete this task with 4 threads on a 4-core Haswell with dual-channel DDR3-1600 is 1d9h. (Some desktop stuff is running in parallel.)
 
Last edited:
  • Like
Reactions: crashtech