PrimeGrid's 2022 Challenge Series

E

EXT64

[H]ard|DCer of the Year 2020
Joined
Mar 27, 2013
Messages
798
The PrimeGrid Challenge Series is an annual competition based on the results of monthly challenges. The purpose of the Challenge is to spotlight a specific project within PrimeGrid's family of projects as well as to offer a little "healthy" competition for PrimeGrid's participants. It is also an opportunity for people to learn more about each specific project and have fun doing it. Below is a tentative schedule for 2022.

I plan to participate in all of the challenges again this year and it would be great if other team members joined in to help give the [H]orde a strong showing in the final results. Happy crunching!
(Summary stolen from RFGuy - thanks!)

In my opinion, PG is the best run challenge as their servers are typically extremely reliable, there is enough work, consistent/equitable credit, optimized cores/code, and best of all there is NO BUNKERING!
There are however nuances to each of the sub projects to watch out for, which hopefully either I or someone else on the team can highlight prior to the challenge.

I'll try to post some benchmark results for CPUs and GPUs in here too to help you decide if you would like to participate in certain challenges and how best to configure your systems.


2022 Challenge Series
​
DateTime UTCProject(s)ChallengeDurationResults
1​
17-20 January​
17:17:00GFN-17-LOW
GFN-17-MEGA		Heptadecagenarian Challenge3 daysIndividuals | Teams
2​
21-26 March​
03:21:00321-LLRWorld Water Day Challenge5 daysIndividuals | Teams
3​
25-30 May​
18:00:00GFN-19Geek Pride Day Challenge5 daysIndividuals | Teams
4​
17-22 June​
13:00:00SR5-LLRM.C. Escher's Birthday Challenge5 daysIndividuals | Teams
5​
22-25 July​
22:00:00PPS-LLRPi Approximation Day Challenge3 daysIndividuals | Teams
6​
17-20 September​
20:00:00WWRiemann's Birthday Challenge3 daysIndividuals | Teams
7​
4-11 October​
12:00:00TRP-LLRWorld Space Week Challenge7 daysIndividuals | Teams
8​
1-11 November​
05:00:00CUL-LLR
WOO-LLR		Prime Meridian Day Challenge10 daysIndividuals | Teams
9​
7-17 December​
05:00:00GFN-21
GFN-22
DYFL		Great Internet Fermat Prime Search Challenge10 daysIndividuals | Teams

Link to latest Challenge schedule: https://www.primegrid.com/challenge/challenge.php
 
Last edited:
First Challenge - GFN-17-Low and Mega

Comments:
These run on both GPU and CPU, however CPU runs surprisingly well here.
Beware however that these will make your CPU run hot.
They discourage OCing and you should keep an eye on temps. Intel CPUs (not sure about AMDs) will typically run at lower frequencies on these while still pulling very high wattage.
For the competition I recommend running single thread and no hyperthreading (not BIOS disabled, just 50% CPU load) for best throughput.
Also, if you are interested in getting a "first" on a prime speed matters (this is no longer the case for other sub-projects like LLR). This will put the CPUs at a severe disadvantage to GPUs for the purpose of 'first' finding a prime; however, this has no impact on the competition as it is based on points (not finding primes).
Low and Mega seem to be running at nearly the same speed and points for me so far.
Unlike the LLR CPU units, these only seem to leverage AVX2 (not AVX512), so older/AMD CPUs do excellent here too

Here are some initial benchmarks (I will continue to update - feel free to add yours below!):

CPUs:
AMD 2P Epyc 7601 (64c/128t, 2.6GHz AVX2 clocks, Ubuntu 20.04, Zen Cores)
64 single thread units
Mega completes in 7,100 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 1.9 minutes
Overall CPU completes 779 units per day
Overall PPD: 373k PPD

Intel 2P Xeon E5-2698V4 (40c/80t, 1.9GHz AVX2 clocks, Ubuntu 20.04, Broadwell Cores)
40 single thread units
Mega completes in 7,200 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 3.0 minutes
Overall CPU completes 480 units per day
Overall PPD: 230k PPD

AMD Ryzen 3950x (16c/32t, 3.7GHz AVX2 clocks, Windows 10, Zen2 Cores)
16 single thread units
Mega completes in 2,900 seconds and receives 478.87 points
Effectively CPU completes 1 WU every 3.0 minutes
Overall CPU completes 477 units per day
Overall PPD: 228k PPD

Intel 2P Xeon E5-2695V3 (28c/56t, 2.7GHz AVX2 clocks, Ubuntu 20.04, Haswell Cores)
28 single thread units
Mega completes in 5,200 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 3.1 minutes
Overall CPU completes 465 units per day
Overall PPD: 223k PPD

Intel i9 10980XE (18c/36t, 3.3GHz AVX2 clocks, Windows 10, Cascade Lake Cores)
18 single thread units
Mega completes in 4,100 seconds and receives 478.85 points
Effectively CPU completes 1 WU every 3.8 minutes
Overall CPU completes 379 units per day
Overall PPD: 182k PPD

Intel Xeon Phi 7250 (68c/272t, AVX2 clocks, Ubuntu 20.04, Airmont Knights Landing Cores)
68 single thread units
Mega completes in 15,500 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 3.8 minutes
Overall CPU completes 379 units per day
Overall PPD: 181k PPD

AMD Ryzen 3700x (8c/16t, 3.8GHz AVX2 clocks, Ubuntu 20.04, Zen2 Cores)
8 single thread units
Mega completes in 2,800 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 5.8 minutes
Overall CPU completes 247 units per day
Overall PPD: 118k PPD

Intel Xeon E5-2683V4 (16c/32t, 1.9GHz AVX2 clocks, Ubuntu 20.04, Broadwell Cores)
16 single thread units
Mega completes in 7,100 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 7.4 minutes
Overall CPU completes 195 units per day
Overall PPD: 93k PPD

Intel Xeon E5-2697V2 (12c/24t, 3.0GHz AVX clocks, Windows 10, Ivy Bridge Cores)
12 single thread units
Mega completes in 6,000 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 8.3 minutes
Overall CPU completes 173 units per day
Overall PPD: 83k PPD

AMD Ryzen 2700x (8c/16t, 3.9GHz AVX2 clocks, Windows 10, Zen+ Cores)
8 single thread units
Mega completes in 6,000 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 12.5 minutes
Overall CPU completes 115 units per day
Overall PPD: 55k PPD

Intel i5 3450 (4c/4t, 3.5GHz AVX clocks, Windows 10, Ivy Bridge Cores)
4 single thread units
Mega completes in 5,000 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 20.8 minutes
Overall CPU completes 69 units per day
Overall PPD: 33k PPD

Intel Atom C3758 (8c/8t, 2.2GHz clocks, Ubuntu 20.04, Goldmont Cores)
8 single thread units
Mega completes in 25,000 seconds and receives 478.86 points
Effectively CPU completes 1 WU every 52.1 minutes
Overall CPU completes 28 units per day
Overall PPD: 13k PPD


GPUs:
Nvidia GTX 3070 (70% Power Limit, Windows 10, Ampere GA104 Cores)
1 GPU unit
Mega completes in 180 seconds and receives 478.86 points
Effectively GPU completes 1 WU every 3.0 minutes
Overall GPU completes 480 units per day
Overall PPD: 230k PPD

Nvidia Titan Xp (70% Power Limit, Ubuntu 20.04, Pascal GP102 Cores)
1 GPU unit
Mega completes in 330 seconds and receives 478.86 points
Effectively GPU completes 1 WU every 5.5 minutes
Overall GPU completes 262 units per day
Overall PPD: 125k PPD

Nvidia GTX 1070 (70% Power Limit, Ubuntu 20.04, Pascal GP104 Cores)
1 GPU unit
Mega completes in 560 seconds and receives 478.86 points
Effectively GPU completes 1 WU every 9.3 minutes
Overall GPU completes 154 units per day
Overall PPD: 74k PPD
 
Last edited:
Haha, yeah. I'll see how competitive I am at that point, and that will probably drive my motivation.
 
I need to do a back to back comparison of Windows vs. Linux for PG GFN - I'm nervous that there may be a handicap there.

Edit: Happy to say it looks like I am wrong - 3950x (Windows 10) and 3700x (Ubuntu 20.04) run at about the same speed.
 
Last edited:
And on second thought I'm not certain lol. Ran the 3950x on Linux and it seems slightly faster. Also, Windows seems to run some units slower sometimes. Not sure why. Ugh. So I guess run Linux if it is convienent, otherwise just run Windows if that's what you have, haha.
 
EXT64 , the 3950x 228PPD is almost the same as RTX 3070 230PPD at the settings that you have. Probably 3950x is more power efficient than RTX 3070 at about the same PPD?
 
  • Like
Reactions: EXT64
like this
3060ti - 160 Watts (Stock except for power at 80%)
Low WU - Averaging 190 seconds per WU for 457 pts. At 448 WU/day is 214k pts.
Mega WU - Averaging 199 seconds per WU for 479 pts. At 434 WU/day is 208k pts.

6900xt - 242 Watts (All stock settings)
Low WU - Averaging 176 seconds per WU for 457 pts. At 472 WU/day is 215k pts.
Mega WU - Averaging 187 seconds per WU for 479 pts. At 462 WU/day is 221k pts.

5900x - 104 Watts w/12 threads active. (All stock settings, built in auto OC)
Low WU - Averaging 2706 seconds per WU for 457 pts. At 375 WU/day is 171k pts.
Mega WU - Averaging 2849 seconds per WU for 479 pts. At 356 WU/day is 170k pts.

3600x - 63 Watts w/6 threads active. (All stock settings, built in auto OC)
Low WU - Averaging 2948 seconds per WU for 457 pts. At 172 WU/day is 79k pts.
Mega WU - Averaging 3095 seconds per WU for 479 pts. At 164 WU/day is 79k pts.
 
That may be true pututu. The 3070 is capped at 150W. I'm not sure what the 3950x is pulling, likely over 100W but less than 150W. Since both the 3950x and 3070 are in the same computer it is one very fast and efficient system for this competition. Skylaar's numbers above are interesting - not a huge performance change for 5000 series it seems.
 
Yeah, very strange. The 5900x at the setting skylaar is running at yields about the same PPD as 3950x when scaled from 12 to 16 cores (33% gain, 171K x 1.33 = 227K PPD).
 
  • Like
Reactions: EXT64
like this
Yes, both are pushing almost exactly 14,250 points per day per core looking at these numbers. I have not played with PBO or performed any overclocking on the 5900x and just left it as is for now.
 
  • Like
Reactions: EXT64
like this
skylaar24 , my 4669v3 with 3060 Ti at 150W power limit in ubuntu 20.04 performance (no cpu tasks running):
  • Low avg run time 175 seconds
  • Mega 183 seconds.
1642535664564.png


Much faster than your 3060 Ti ;)

Also, just 3 seconds slower than EXT64 3070 for Mega.
 
Last edited:
  • Like
Reactions: EXT64
like this
That 3060ti is the very base model from eVGA. It is the short card of the bunch only requiring a single power connection. Solid little card but I am sure it lacks some performance compared to higher end 3060ti. You also have the ubuntu advantage going for you, haha!
 
Last edited:
  • Like
Reactions: EXT64
like this
Great job everyone! I'll let pututu provide a summary of final stats ;). A little while until the next competition, which is CPU only. I'll provide some suggestions and tuning data closer to the competition.
 
Awesome performance by our team. Solid 7th place in the 3-day GFN-17 challenge which happens to actually favor both cpu and gpu almost equally. Gratz to our friend TAAT for taking the top spot. Hmm, can we beat TAAT in next week FAH challenge? We will know soon. Stay tune for the battle of who has the most and latest GPUs. Did I say GPUs are easily available?

1642719803186.png


Individual performance. Yours truly is at the bottom of the list. I guess I've been busy whipping those in front of me. ;)
Coleslaw/Gilthanis used to be at the bottom/middle pack but this time, he turned on his afterburner.
1642719866831.png
 
Thanks - interesting detail: 30 primes were found during the competition; 9 GFN-17-Low, and 21 GFN-17-MEGA.
 
Looks like we are coming up on the next challenge. Here is a link to the specific challenge thread with lots of great detail: https://www.primegrid.com/forum_thread.php?id=9888#154866

This one can be a challenge to tune as there are a lot of overlapping factors. Hyperthreading seemed to help a little on AMD, but I may just skip it to reduce power (it isn't a huge gain). A bigger factor is likely cache usage which is managed by controlling the number of WUs running at a time

"In General": I found 4 MT threads and 1T/core an excellent compromise across all my systems. On some of my AMD systems 8 MT threads and 2T/core showed a benefit. ymmv....

Linux was faster than Windows, but it wasn't anything shocking (maybe 10%) and it wasn't a very scientific test.

Here are all the configurations I tested. OS is (W)indows or (L)inux. The best for each system has two stars (**) to mark it, however 4MT threads and 1T/core worked well for all systems.

ComputerClockCoresClock*CoresThreads/CoreMT ThreadsAvgMinMaxAvgMinMax# Runninghr/WUWU/dayPPDPPD/Clock*CoresUtilization
2P Epyc 7601(L)2.764172.8145782857193586212232782213352251221616.0623.912079771203.63.86
Xeon Phi 7250(L)**1.36888.4147008168624714662599782543312654851719.4720.961823402062.73.71
i9 10980XE(W)**2.81850.41417505170641800767181656346909044.8619.741717633408.03.84
2P E5-2695V3(L)2.72875.61432497312373369812497211910012997679.0318.611619152141.73.85
2P E5-2698V4(L)**1.940761447602472064786671828061805251845731013.2218.151579092077.83.84
Ryzen 3950x(L)**3.71659.21428241272312869110746710496010882747.8412.241064661798.43.81Windows:
33673​
Ryzen 3700x(L)3.9831.21422519219752293586372848868787526.267.67667602139.73.84
2P E5-2695V3(L)2.72875.61719214185472002412245011735912683745.3417.991564862069.96.37
2P Epyc 7601(L)2.764172.81830823302203170222608122254523136788.5622.421950961129.07.33
2P E5-2698V4(L)1.9407615386913812439140183413179191186327810.7517.861554222045.04.74
Ryzen 3950x(L)3.71659.21816596158921705812033111679512269224.6110.41905861530.27.25Windows:
17869​
Ryzen 3700x(L)3.9831.2181365913403139171008109959010205813.796.33550321763.87.38
2P E5-2695V3(L)**2.72875.627371263534339012235187227948244769810.3118.621619742142.56.33
2P Epyc 7601(L)**2.764172.8285600055331568984060444040324090421615.5624.692147661242.97.25
2P E5-2698V4(L)1.94076259160786100952144164733992644279391625.4515.091312881727.54.55
Ryzen 3700x(L)**3.9831.22818619182181889112675012569712762625.179.28807432587.96.81

I put this together a little while back, so the WUs have likely gotten slower since this and some of my computers have been reconfigured a bit.
 
Just a reminder, that it is a 5 day challenge and ONLY the 321 Prime Search LLR (321) work units should be selected in your preferences at PG.
 
  • Like
Reactions: EXT64
like this
EXT64, thanks for the update PPD/run time study. I'm debating to either run intel or amd rig for this 5-day challenge.
 
  • Like
Reactions: EXT64
like this
For intel, I have i7-6900K replacing the 4669v3 on the current x99 platform. Don't know if I should keep the 4669v3 that I have for 4+ years or just sell it. This isn't an upgrade (some downgrade though in total output) but more to feed the faster ampere gpus when running WU requiring 1 full core.
 
And the AMDs? That 6900K should do reasonably well, but the newer AMDs (3xxx and newer) are pretty good.
 
And we're off!

On a side note, I've been having some challenges tuning my 3950x (as usual, it has always been a pain). Anyway, I noticed that for 321-LLR it is really sensitivity to memory/infinity fabric speed. Something to keep in mind if you have one of these dual/multi chiplet CPUs.
 
All right. Intel it is. Running 4 cores per task.
 
Looks like we hit the ground running at 10th place. Mine are getting a late start still and am down some threads for a bit. Maybe down for this event but others should be kicking in a little here shortly.
 
Thanks for the chart(s) - that is awesome. I had not seen that before.
 
Day 1.5 result. Only five contributors versus nine in the last challenge.

1647963332379.png
 
Looks like the team is up to 8th right now. Finally got some more of my work uploaded but still down 2 rigs.
 
Day 2.5 result. Another 2.5 days to go. Looks like I'm going to be last again for the second time in a row.

1648050697469.png
 
You must log in or register to reply here.
Back
Top