Primegrid Challenge series started yesterday

We are over half way through with this short challenge. We haven't broken the top twenty yet, but are close.

1706106094911.png


1706106020304.png
 
[Ion] said:
Count me in -- somehow I never picked a team. Oops.
Click to expand...
You can register at boincgames.com, too, if you like. The BOINC projects are tracked there quarterly and annually as "Marathons", so once registered you contribute to the team and your individual stats, but can drop in and out whenever you like, switch projects, etc. Very casual but helpful since these events coincide with their "Marathon" tracking. A lot of us in the northern hemisphere stop or cut back in the summer months.

AMD users have been giving us a run, recently. With a few more signups and some project direction/focus we'd be able to make a run at TAAT.

1706113337212.png

1706113466441.png
 
1706121095949.png

Is there something obvious that y'all can think about explaining the mismatch between score & tasks? I'm at #2 on tasks it looks like, but a ways down in the points list.


Also top 20, let's gooooo:
1706121161177.png
 
[Ion] said:
View attachment 629846
Is there something obvious that y'all can think about explaining the mismatch between score & tasks? I'm at #2 on tasks it looks like, but a ways down in the points list.


Also top 20, let's gooooo:
View attachment 629847
Click to expand...
There are proof and main tasks for each prime search. The main proof task is the one that does the prime search so it takes longer but with higher credit score. The proof main task is for validation and runs very short with low credit score. I think the stat above shows the aggregate of both type of tasks. Everyone doing the prime search in this subproject is "first", so no need to have the fastest cpu.

1706122084112.png
 
Last edited:
pututu said:
There are proof and main tasks for each prime search. The main task is the one that does the prime search so it takes longer but with higher credit score. The proof task is for validation and runs very short with low credit score. I think the stat above shows the aggregate of both type of tasks. Everyone doing the prime search in this subproject is "first", so no need to have the fastest cpu.

View attachment 629848
Click to expand...
Oh right...should have remembered that, since I've seen the check tasks blasting through BOINC in a fraction of the time of the others. Thanks!
 
Thanks to Ion jumping in like a rocket and RFGuy_KCCO performing consistently we managed 15th.

Sign up at Boincgames.com.

Now, back to F@H.

1706231339293.png


1706231564167.png
 
  • Like
Reactions: [Ion]
like this
It's nice and cold here in Maine...lots of opportunities to use the computers to heat the house :)
 
[Ion] said:
It's nice and cold here in Maine...lots of opportunities to use the computers to heat the house :)
Click to expand...
Thanks for giving us a hand. I've to divert all my electricity towards folding ;)
You are welcome to join in the fun and for a good cause too. That will easily heat up your house, lol.
The team is currently in last position.
 
PrimeGrid down for anyone else? Perhaps we killed it with the challenge....I can't access the website, nor can my devices upload work :O

EDIT: And now the website is back up, with a maintenance message. That's better, at least.
 
Last edited:
Looking forward to it. Started experimenting on some of my Xeons exactly the right mix in terms of # of CPUs/task.
 
14th place so far 😑

I wrote a Python script last night to handle pinning individual WUs to CCDs in my EPYC chips, and the results are _drastically_ better than letting the OS do it itself. With that in mind, I've got a couple more of my systems now running TRP as well, and hopefully that'll get us at least into the top 10.
 
[Ion] said:
14th place so far 😑

I wrote a Python script last night to handle pinning individual WUs to CCDs in my EPYC chips, and the results are _drastically_ better than letting the OS do it itself. With that in mind, I've got a couple more of my systems now running TRP as well, and hopefully that'll get us at least into the top 10.
Click to expand...
Nice. I saw that in the discord. Thinking if I should switch one rig to PG and test out your py script. 🤔

BTW, 2024 boinc pentathlon is coming soon. The first marathon project will be announced 17 days from now.
 
  • Like
Reactions: [Ion]
like this
pututu said:
Nice. I saw that in the discord. Thinking if I should switch one rig to PG and test out your py script. 🤔

BTW, 2024 boinc pentathlon is coming soon. The first marathon project will be announced 17 days from now.
Click to expand...
Seems to be making a huge difference (tasks run in < half the time) on my EPYC system. Can't argue with that.

So overall the script is workable, I think. It could use a bit more polish to make it easier to use. So far it lets you skip the first couple CPU cores if necessary to keep things lined up, but if f.ex you have 2x16c CPUs & want to run 3x5c WUs per CPU, it isn't going to be terribly easy to do that. You'd have to know to align the first WU to start on core 1 of the first CPU, and then it'll fill the last 15 cores of the first CPU, and the first 15 cores of the second CPU. So I'll do some poking at it and see if I can find a way to easily & precisely specify what CPUs to use. I'll poke at it when I have some time, but for now it works OK enough to use.
 
[Ion] said:
Seems to be making a huge difference (tasks run in < half the time) on my EPYC system. Can't argue with that.

So overall the script is workable, I think. It could use a bit more polish to make it easier to use. So far it lets you skip the first couple CPU cores if necessary to keep things lined up, but if f.ex you have 2x16c CPUs & want to run 3x5c WUs per CPU, it isn't going to be terribly easy to do that. You'd have to know to align the first WU to start on core 1 of the first CPU, and then it'll fill the last 15 cores of the first CPU, and the first 15 cores of the second CPU. So I'll do some poking at it and see if I can find a way to easily & precisely specify what CPUs to use. I'll poke at it when I have some time, but for now it works OK enough to use.
Click to expand...
Not sure if it is working but here are the screenshots with 32-core epyc (SMT disabled) running 4 tasks with each task having 4 cpus. Somehow the htop shows no pinning at all but the run time looks ok? clock freq at 3.1GHz
1713227868773.png
1713227937287.png


1713227973527.png



Here is the epyc ccx topology on that rig.
1713228153386.png
 
I didn't know about this change, lol. Now htop is showing the correct cpu core affinity. Thanks!
 
  • Like
Reactions: [Ion]
like this
Since it looks like there's only 5 people on the team running this I'd say we're doing pretty damn good so far.
 
pututu said:
I didn't know about this change, lol. Now htop is showing the correct cpu core affinity. Thanks!
Click to expand...
Yes, I misunderstood how Primegrid spawns additional worker threads (I think) for each of the tasks, and you have to get all of them bound to the correct CPUs or it won't work. Oops.
 
Got 2 failures. Have to dial back the clock to 2.9GHz with Vcore of 0.97V. Average temp for each of the four CCDs looks ok to me. Kill-a-watt shows 215W.

1713279484535.png
 
pututu said:
Got 2 failures. Have to dial back the clock to 2.9GHz with Vcore of 0.97V. Average temp for each of the four CCDs looks ok to me. Kill-a-watt shows 215W.

View attachment 648069
Click to expand...
Any overclocking going on? My systems are all running stock (+ turbo as temps/power limit allows) and are completing tasks w/o issue.
 
[Ion] said:
Any overclocking going on? My systems are all running stock (+ turbo as temps/power limit allows) and are completing tasks w/o issue.
Click to expand...
I'm undervolting the epyc core voltage and figuring the clock which will operate close to or at peak optimum power-crunching efficiency ;)

I can overclock my epyc but this require higher voltage and power and I rarely does that.

Got too many warnings on the proof tasks but somehow the credits are granted, so I assume it is still ok other than the two failed tasks. The run time seems to vary somewhat from 28000 secs to 42000 secs due to this. The stderr file says something about some errors and restarting from the last save file (checkpoint).

1713290413343.png
 
Last edited:
Looks like our present threat and opportunity come from AMD and Romania team with 4 days and 17 hours left in the race. Where is Holdolin?


1713291476765.png
 
Gilthanis said:
It's probably require more than double...haha.
Click to expand...
I'm givin' her all she's got, captain!

(not quite, but it's finally warming up enough that I can't bear to turn on the last two systems. If the house heats up too much, the dual-CPU EPYC system overheats the VRMs and throttles to 400MHz, do it does more harm than good :( )
pututu said:
I'm undervolting the epyc core voltage and figuring the clock which will operate close to or at peak optimum power-crunching efficiency ;)

I can overclock my epyc but this require higher voltage and power and I rarely does that.

Got too many warnings on the proof tasks but somehow the credits are granted, so I assume it is still ok other than the two failed tasks. The run time seems to vary somewhat from 28000 secs to 42000 secs due to this. The stderr file says something about some errors and restarting from the last save file (checkpoint).

View attachment 648118
Click to expand...
Huh. How are you undervolting them? That could be nice to experiment with after the challenge :)
 
[Ion] said:
I'm givin' her all she's got, captain!

(not quite, but it's finally warming up enough that I can't bear to turn on the last two systems. If the house heats up too much, the dual-CPU EPYC system overheats the VRMs and throttles to 400MHz, do it does more harm than good :( )

Huh. How are you undervolting them? That could be nice to experiment with after the challenge :)
Click to expand...
Unfortunately, you need specific Epyc ROME ES/QS chip with part number ending with suffix -04, iirc. The -05 does not work as I've tried it before. This link here will get you started together with the some discussions and observations: https://forums.servethehome.com/index.php?threads/finally-overclocking-epyc-rome-es.28111/

In a nutshell, taking advantage of a "bug" in the PBO in selected epyc rome engineering/qualification samples. Someone modified the zenstates.py and I use this python script to create a bash file to modify voltage and frequency of my epyc cpu.

Unfortunately it doesn't work with retail chip.
 
[Ion] said:
I'm givin' her all she's got, captain!

(not quite, but it's finally warming up enough that I can't bear to turn on the last two systems. If the house heats up too much, the dual-CPU EPYC system overheats the VRMs and throttles to 400MHz, do it does more harm than good :( )

Huh. How are you undervolting them? That could be nice to experiment with after the challenge :)
Click to expand...
I hear you. I shut down and 8 thread Optiplex and fired up a 32 thread Precision server just recently with a GTX980 in it. My bedroom was already hot but adding that swap to the mix at the same time that the temps went up about 15 degrees outside is like a kick in the nuts...lol
 
Gilthanis said:
I hear you. I shut down and 8 thread Optiplex and fired up a 32 thread Precision server just recently with a GTX980 in it. My bedroom was already hot but adding that swap to the mix at the same time that the temps went up about 15 degrees outside is like a kick in the nuts...lol
Click to expand...
At least here in Maine it's not hot yet outside, so I can open up the windows, stick a box fan in the window, and keep the place perfectly comfortable that way. Definitely going to be running at reduced strength for the summer contests though.
 
2 days 21 hours remaining...

We are neck and neck with the other team.

1713449953388.png
 
  • Like
Reactions: [Ion]
like this
Looks like we just overtook them, so seeing as they've been ahead for most of the challenge, that looks like a good deal for us. Top-10 finish I can take :)
 
You must log in or register to reply here.
Back
Top