PrimeGrid Races 2017

[TennesseeTony]

Having met my personal goal and seeing that the TeAm is secure in 7th place, with no chance to move up or down, I have finished up my current tasks and aborted the rest.

I admit I don't get the fascination with number crunching, but this race was a bit more interesting, seeing the updates from the staff of PG, and making such substantial progress on this sub-project during the race.

I of course enjoyed the thrill of the race, and the impressive comeback from the TeAm after participating in the Formula BOINC race at the same time.

Thank you Ken, for leading the TeAm on these events, and the score-keeping!

Thank you Stefan, for the detailed information regarding the app_config, for making more than half our points, and your charts are awesome!

Upcoming races:
Formula BOINC 04/27/2017 21:00 (UTC) - 04/30/2017 20:59 (UTC)
8th BOINC Pentathlon 05/05/2017, 0:00 UTC to 05/19/2017, 0:00 UTC
More Formula BOINC Sprints, and then
The next PrimeGrid race: 12-13 June, 0:00 UTC, just a 24 hour event.

 

[Ken g6]

Preliminary final stats:

Rank___Credits____Username
7______8187330____xii5ku
40_____2386008___10esseeTony
148____580881_____Ken_g6
200____373233_____markfw
213____349557_____crashtech
229____309125_____Orange Kid
270____225664_____zzuupp
288____208926_____GLeeM
310____176162_____SlangNRox

Rank__Credits____Team
4_____39064260___Sicituradastra.
5_____18519617___Crunching@EVGA
6_____14094007___BOINC@Poland
7_____12796891___TeAm AnandTech
8_____9753016____Team 2ch
9_____9324885____BOINC@MIXI
10____8944822____Rechenkraft.net

They're preliminary because if any of your WUs get invalidated, they're deducted. But I don't expect we'll fall any as a TeAm.

Great job, everyone!

 

[StefanR5R]

Once again, a prime rank for the TeAm!

(And on the individuals leaderboard, merely two other teams show up ahead of TeAm AnandTech's uppermost rank. )

Some serious horse power has been put on the road, especially throughout the last third of the race:

Spoiler: graphs

That was quite a lot of fun for a 15-day math challenge with ~2-day-WUs... Interesting tinkering with the multithreading option, a thrilling break for the Formula BOINC sprint, and a superb debut of the new AMD CPUs at PrimeGrid!

 

[Orange Kid]

Thanks for the stats
Great race all

 

[crashtech]

Nice stats, I wish I'd realized sooner that the TeAm could use some more CPU cycles. Pretty decent showing though, I think.

 

[Ken g6]

Time to bump this for the race next week!

 

[StefanR5R]

Alas, it's a 24 hour race whose first 4 hours overlap with the upcoming Formula BOINC sprint.

 

[Orange Kid]

A bump and less than 3 hrs to the start of
PrimeGrid's 2017 Challenge Series
PrimeGrid's Birthday Challenge

 

[Ken g6]

The race is on! I've switched all my machines over.

 

[StefanR5R]

Let's pluck some more top-5000 primes.

 

[StefanR5R]

One of my boxes gets a fraction of computation errors:
"finish file present too long"
http://www.primegrid.com/result.php?resultid=822315998

Spoiler: Stderr output

<core_client_version>7.6.33</core_client_version>
<![CDATA[
<message>
finish file present too long
</message>
<stderr_txt>
BOINC llr wrapper (version 8.00)
Using Jean Penne's llr (64 bit)
LLR Program - Version 3.8.20, using Gwnum Library Version 28.13

LLR command line: primegrid_llr -d -oDiskWriteTime=100 ../../projects/www.primegrid.com/llrSGS_287872872
Using zero-padded FMA3 FFT length 128K, Pass1=512, Pass2=256
02:18:40 (2580): called boinc_finish(0)

</stderr_txt>
]]>

What does it mean?

 

[VirtualLarry]

I've had to crank my overclock on my Ryzen 5 1600 down, from 3.800Ghz @ 1.3250V to 3.700Ghz @ 1.2875V.

It seems that the "invisible" LLC, cranks up the voltage beyond what I've set it at, anyways. :|

My MasterLiquid Lite 120 got a bit overwhelmed. With 11 PrimeGrid SGS tasks, and one GPU task, my Package Power was running at 115W, and my Package Temp., was hitting above 80C, after which I walked away and it seemed to promptly black-screen.

So I cranked it back down, to what I was previously running on the Wraith Spire air cooler included with the CPU.

But it's still hitting 108W Package Power with these settings, and those SGS tasks. Ouch.

I guess this 120mm liquid cooler is a bit... underpowered.

I'll look forward to a better case and a bigger AIO CLC in the future. Until then, I'll crank it down until temps are in check.

Currently at 75C Package Power.

Not to mention, the air going through the Rad is going to be warmer, now that I upgraded from a GT630 passive video card, to my RX 470 4GB GDDR5 video card, and started running AP27 tasks on it.

Edit: Bit curious, does SGS CPU tasks use AVX or AVX2 or FMA? They do seem to be running hotter / more power than SoB CPU tasks did.

 

[StefanR5R]

One of my boxes gets a fraction of computation errors:
"finish file present too long"
http://www.primegrid.com/result.php?resultid=822315998

Spoiler: Stderr output

<core_client_version>7.6.33</core_client_version>
<![CDATA[
<message>
finish file present too long
</message>
<stderr_txt>
BOINC llr wrapper (version 8.00)
Using Jean Penne's llr (64 bit)
LLR Program - Version 3.8.20, using Gwnum Library Version 28.13

LLR command line: primegrid_llr -d -oDiskWriteTime=100 ../../projects/www.primegrid.com/llrSGS_287872872
Using zero-padded FMA3 FFT length 128K, Pass1=512, Pass2=256
02:18:40 (2580): called boinc_finish(0)

</stderr_txt>
]]>

What does it mean?

It's an application bug, isn't it? (Race condition between disk I/O and boinc_finish() call.)

A nasty side effect of these errors, besides the lost work, is that the boinc client defers next communication with the server for several hours after such an error occurred. This is bad in a race which lasts only a day.

 

[VirtualLarry]

I would check your BOINC prefs, and make sure that you have enough disk space allocated for BOINC to utilize.

 

[StefanR5R]

I would check your BOINC prefs, and make sure that you have enough disk space allocated for BOINC to utilize.

Indeed -- but it is enough. Configured: "Use no more than ... GB" disabled; enabled "Leave at least 0.1 GB free" and "Use no more than 90 % total", which in this case results in "free, available to BOINC: 360 GB".

the boinc client defers next communication with the server for several hours after such an error occurred.

To break such communications pauses if they happen again, I am now running a shell script which calls "boinccmd --host abc --passwd xyz --project 'http://www.primegrid.com/' update" for each host every ten minutes.

 

[StefanR5R]

does SGS CPU tasks use AVX or AVX2 or FMA? They do seem to be running hotter / more power than SoB CPU tasks did.

The announcement mentions FMA3 at least.

 

[StefanR5R]

One of my boxes gets a fraction of computation errors:
"finish file present too long"
http://www.primegrid.com/result.php?resultid=822315998

Spoiler: Stderr output

<core_client_version>7.6.33</core_client_version>
<![CDATA[
<message>
finish file present too long
</message>
<stderr_txt>
BOINC llr wrapper (version 8.00)
Using Jean Penne's llr (64 bit)
LLR Program - Version 3.8.20, using Gwnum Library Version 28.13

LLR command line: primegrid_llr -d -oDiskWriteTime=100 ../../projects/www.primegrid.com/llrSGS_287872872
Using zero-padded FMA3 FFT length 128K, Pass1=512, Pass2=256
02:18:40 (2580): called boinc_finish(0)

</stderr_txt>
]]>

What does it mean?

Whoops --- primegrid.com promoted this task, and all the other errored ones, from "error" to "Completed, waiting for validation". Nice.

 

[TennesseeTony]

Just a reminder to X99 users, AVX (if employed by these tasks) somehow makes overclocked systems draw max power from the board. My poor i7 was once force-fed 300watts (max my MB can provide) for many many many hours before I noticed the temp was extreme. As more projects bring AVX to their tasks, I just run stock speed now on Mjolnir.

 

[StefanR5R]

Haha, this is one of those moments where one better takes a screenshot quickly ---

 

[StefanR5R]

Just a reminder to X99 users, AVX (if employed by these tasks) somehow makes overclocked systems draw max power from the board. My poor i7 was once force-fed 300watts (max my MB can provide) for many many many hours before I noticed the temp was extreme.

On X99, only Haswell-E is affected by this problem. Haswell-EP and Broadwell-EP automatically clock down while AVX or FMA ops are encountered, and Broadwell-E can optionally be configured to a clock offset for AVX/FMA.

 

[TennesseeTony]

They can't be overclocked though, so they are naturally excluded from this bug.

 

[crashtech]

Well, my new 2P mobo did not make it for this contest, but I will add every little thing I can, down to the last dual core...

 

[Ken g6]

The announcement mentions FMA3 at least.

No, LLR isn't using FMA yet. It does use AVX for sure though. The potential for FMA use should first be announced on mersenneforum.org.

 

[Ken g6]

3-hour stats (This is gonna be a quick race!):

Rank___Credits____Username
2______83379______xii5ku
15_____27899___10esseeTony
95_____3233_______Ken_g6
128____2275_______crashtech
137____2035_______VirtualLarry
143____1915_______Orange Kid
184____957________salvorhardin

Rank__Credits____Team
2_____266941_____SETI.Germany
3_____201962_____Czech National Team
4_____171069_____Sicituradastra.
5_____121696_____TeAm AnandTech
6_____62464______Crunching@EVGA
7_____61945______Christians
8_____51049______BOINC@Poland

Wow! If it holds, fifth would be really impressive!

 

[crashtech]

I started late, so expect more!