PrimeGrid Challenges 2020

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
@VirtualLarry, since PrimeGrid is among your projects, keep the ~ doubled FP and FMA throughput of Zen2 over Zen/ Zen+ in mind when you calculate bang-for-the-buck. Doubled level 3 cache is good to have for PG too.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
14,968
1,935
55
Although I find it hard to argue for doing it in this time of international crisis, there is a PrimeGrid challenge coming up in a couple of days.
 
  • Like
Reactions: TennesseeTony

VirtualLarry

Lifer
Aug 25, 2001
48,052
4,984
126
Already switched my 6C/12T CPUs over, WUs appear to be fairly short (I'm doing 3x 4T), on the order of 5-8min, so I'm not worried about starting the race right at the "gun".
 

zzuupp

Lifer
Jul 6, 2008
14,777
2,305
126
WTF! We're on teeam "PC-BOINC!? w/ "founder" wuj.

Team Anandtech is [obsolete] w/ founder "
Dwight Ringdahl

I hope this just an April Fool's Day joke.
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
14,968
1,935
55
I think we might have gotten hijacked. That can happen when the founder doesn't check their emails. :(

Edit:
 
Last edited:

biodoc

Diamond Member
Dec 29, 2005
5,596
874
126
I think we might have gotten hijacked. That can happen when the founder doesn't check their emails.
That's a bummer. So is the plan to create another TeAm Anandtech and we all quit the other team and join the new TeAm Anandtech or is there another option?
 

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
Already switched my 6C/12T CPUs over, WUs appear to be fairly short (I'm doing 3x 4T), on the order of 5-8min, so I'm not worried about starting the race right at the "gun".
The Ryzen 6-core processors have core complexes with 3 cores each. With three 4-threaded processes, at least one process will cross CCXs which gives poor performance if there is relevant synchronization between the threads of the process going on.

But what's more, the SGS tasks are so small that best throughput is reached with low threadcount per task but high task count per computer. Specifically, last time I checked, Intel CPUs gave highest throughput with single-threaded tasks and Hyperthreading off (or Hyperthreading unused). I have not yet researched what the optimum setting for Zen or Zen2 is. Their SMT is more powerful than Intel's, on the other hand their FP/FMA resources differ from Intel's, but also from each other --- significantly.
 

zzuupp

Lifer
Jul 6, 2008
14,777
2,305
126
I think we might have gotten hijacked. That can happen when the founder doesn't check their emails. :(

Edit:
aw $%^&!!

Um, now what? I have heard of this happening, but never paid enough attention to remember the ultimate solution.
 

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
I think we might have gotten hijacked. That can happen when the founder doesn't check their emails.
This is a good reminder to us users, to set "Is it OK for <project> and your team (if any) to email you?" to No at all projects, if you don't want your boinc e-mail address become known to random people.

Edit, as a team, should we systematically work towards getting team foundership transferred to TAC at all projects?
 

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
Here is something which I should certainly look up in the PG forum, but haven't yet:

Circa since they implemented the "mt" plan class, the PG server sends only one task per request for work. Is there any way to change that?

Because good hosts are now easily finishing tasks as fast as they can request tasks. (A workaround on hosts which are too fast would be to run more than one client instance at once.)

PS,
my best host (when using a single client) is indeed sometimes almost exactly as fast finishing work as it can get new work, othertimes it cannot fetch fast enough.

Edit,
another workaround is to force a project update every n seconds in order to eliminate some of the request latency within the client. The update period should be slightly lower than the period between task completions. On my best host that's less than 12 seconds.

Edit 2,
I went to the PG forum and asked now.
 
Last edited:

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
14,968
1,935
55
Day 1 stats:

Rank___Credits____Username
12_____673261_____xii5ku
100____90084______emoga
118____75835______Orange Kid
132____62823______VirtualLarry
133____61746______Icecold-Team Anandtech
149____48814______SlangNRox
190____29216______Ken_g6
196____27779______zzuupp
216____23469______waffleironhead
412____2793_______biodoc

Rank__Credits____Team
7_____2094463____Ultimate Chaos
8_____1378414____Crunching@EVGA
9_____1228100____Sicituradastra.
10____1095068____TeAm AnandTech
11____877220_____BOINC@MIXI
12____767537_____The Knights Who Say Ni!
13____722914_____The Scottish Boinc Team

Not bad considering how much else is going on lately.
 

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
Circa since they implemented the "mt" plan class, the PG server sends only one task per request for work. Is there any way to change that?
[...]
I went to the PG forum and asked now.
Update:
There is no hard limit of 1 new task per request.
However, to receive more, the work buffer needs to be set strangely large.

I'll stick with a reasonable buffer depth and unreasonable request rate until I learn more (if I'll do).
 

Ken g6

Programming Moderator, Elite Member
Moderator
Dec 11, 1999
14,968
1,935
55
Day 2 stats:

Rank___Credits____Username
8______1580895____xii5ku
75_____267500_____biodoc
85_____232576_____crashtech
116____161170_____Orange Kid
123____153068_____emoga
126____144327_____Icecold-Team Anandtech
139____123492_____VirtualLarry
161____96790______SlangNRox
179____73919______Ken_g6
194____60109______waffleironhead
240____42986______zzuupp
582____558________amd.borg

Rank__Credits____Team
5_____6706913____SETI.Germany
6_____5067424____[H]ard|OCP
7_____4475386____Ultimate Chaos
8_____2935641____TeAm AnandTech
9_____2895249____Crunching@EVGA
10____2398125____Sicituradastra.
11____1733325____BOINC@MIXI

@waffleironhead shamed me into using more cores for this. And welcome back @amdxborg!
 

VirtualLarry

Lifer
Aug 25, 2001
48,052
4,984
126
Did we run PrimeGrid dry? I don't have any tasks showing for my R5 3600. ?

Edit: Must have been a temporary blip, I seem to have tasks on two out of my three PCs now.
 
Last edited:

StefanR5R

Diamond Member
Dec 10, 2016
3,433
3,674
106
Yes, we did. Or maybe rather, the database has been brought to its knees.

I was sad that the Formula Boinc sprint was cancelled. But with this extraordinary PrimeGrid challenge as a replacement, I feel right at home.

----
Update: Now I get tasks again.
 
Last edited:

ASK THE COMMUNITY