PrimeGrid Races 2017

[VirtualLarry]

Nice to see you, Larry.

Sorry I'm so late to the party. I've got one i3-8100 crunching on this, in an ITX mobo / case. (No overclocking.)

 

[StefanR5R]

I just "caught up" with you all, even with @crashtech...

Made it into the top twenty. But will I stay there until the end? Only my laptop and the old 4960X are rieseling forth until the very end of the challenge.

Edit:
And the TeAm is in the top ten now. Will we cling on?

 

[crashtech]

I just "caught up" with you all, even with @crashtech...

Made it into the top twenty. But will I stay there until the end? Only my laptop and the old 4960X are rieseling forth until the very end of the challenge.

Edit:
And the TeAm is in the top ten now. Will we cling on?

I think the term is more like "blew past us like we were standing still!"

 

[StefanR5R]

@Markfw, here is a guy who uses hardware which definitely doesn't fit in his home, and probably isn't even located on the same continent as himself.

TRP-LLR: Diwali/Deepavali Challenge (2017-10-18 00:00:00 to 2017-10-23 00:00:00)
Last update: 2017-10-22 20:15:07
Rank ..... Name ......... Team ............................. Score
4 ............ mexxus ..... BOINC@AUSTRALIA ..... 4406645.45

http://www.primegrid.com/hosts_user.php?userid=14407

(edit) Apparently, those are gratis trial cloud instances.

 

[Ken g6]

Day 5 (preliminary final) stats:

Rank___Credits____Username
18_____1816430____xii5ku
45_____796026_____crashtech
90_____333656_____Ken_g6
142____191769___10esseeTony
148____187046_____zzuupp
192____118663_____Orange Kid
263____60038______VirtualLarry

Rank__Credits____Team
5_____5621287____BOINC@Poland
6_____5247194____Crunching@EVGA
7_____5236586____BOINC@AUSTRALIA
8_____3503631____TeAm AnandTech
9_____3374311____BOINC@MIXI
10____2902007____Meisterkuehler.de Team
11____2670785____US Navy

Great race everyone! And, @StefanR5R, that was a nice surprise.

Now, everybody jump on TN-Grid for the next 5 hours.

 

[crashtech]

@Markfw, here is a guy who uses hardware which definitely doesn't fit in his home, and probably isn't even located on the same continent as himself.

TRP-LLR: Diwali/Deepavali Challenge (2017-10-18 00:00:00 to 2017-10-23 00:00:00)
Last update: 2017-10-22 20:15:07
Rank ..... Name ......... Team ............................. Score
4 ............ mexxus ..... BOINC@AUSTRALIA ..... 4406645.45

http://www.primegrid.com/hosts_user.php?userid=14407

(edit) Apparently, those are gratis trial cloud instances.

My curiosity is piqued by the thought of running DC on a cloud hosting account. I wonder how much you get in the way of resources, or if they would frown on such usage.

 

[Orange Kid]

Thanks for the stats
That xii5ku guy has a way of entering the room
Good run all !!!

 

[Ken g6]

My curiosity is piqued by the thought of running DC on a cloud hosting account. I wonder how much you get in the way of resources, or if they would frown on such usage.

Some do. You generally need to look for hosts that explicitly list computing power available, rather than just bandwidth or storage.

 

[StefanR5R]

I have seen arguments that there is no sportsmanship in using cloud computers during DC competitions. It was pointed out in such arguments that these people did at least pay the rent. But in this particular instance, it is possible that the person didn't pay anything, merely used promotional trial accounts.

 

[VirtualLarry]

Is the race over? Sorry peeps, I tried, with a single i3-8100. Now we know what it can do (although, I started a day or so late).

 

[Ken g6]

Is the race over? Sorry peeps, I tried, with a single i3-8100. Now we know what it can do (although, I started a day or so late).

Yes, the race is over. We got 8th, and we couldn't have done better without a rack's worth more computers. So good job!

 

[VirtualLarry]

Wow, I saw "PrimeGrid Race" on the top thread on the forum, and I opened it, and this race wasn't in it. Then I realized, that's the 2016 thread, and this is the 2017 thread. I though the forum puked, or I went back in time, or something. Whew.

 

[StefanR5R]

Confusingly though, the real current PrimeGrid race thread is the Formula Boinc thread.

 

[TennesseeTony]

Anyone know how much performance is lost in the upcoming race projects,due to PG only having 32bit tasks for Windows, versus 64bit tasks for Linux?

The GTX1070/1080/1080Ti really seem to fall behind until you get to the GFN17-short tasks, according to this. But how do the listed values translate into points differences? Or, more importantly, is there a workaround, to bring the points up to the 64bit level?

 

[Ken g6]

Anyone know how much performance is lost in the upcoming race projects,due to PG only having 32bit tasks for Windows, versus 64bit tasks for Linux?

I don't know for sure, but I don't think it's significant. Primality tests like this are all FPU work anyway.

The GTX1070/1080/1080Ti really seem to fall behind until you get to the GFN17-short tasks, according to this.

Yep. You should really try to work on tasks as big as you can. Here, that's GFN17 Low. If one of them doesn't fully occupy your GPU, trying running two or more at once.

 

[zzuupp]

From Micheal Goetz's post on the primegrid race thread


Both CPUs and GPUs may be used, but GPUs are MUCH faster.

• It pays to leave one CPU core idle to service the GPU(s). This will allow the GPU to run faster. On a 4 core CPU, setting the BOINC "Use at most ###% of the CPUs" setting to 75% will leave one core free; with 8 cores (or 4 cores with hyperthreading) set it to 87.5%.

• On GPUs, you'll get better credit per hour (and hence challenge score) running GFN-17-Low. Today's large GPUs can't be fully utilized on the smaller GFN-15 and GFN-16 tasks.

• On CPUs, you'll get almost identical credit/hour on GFN-15, GFN-16, and GFN-17-low, so it doesn't matter which you run.

• On CPUs, you will get better overall performance with hyperthreading turned ON. The tasks will take longer to run, but you'll be running twice as many simultaneously, and overall you'll complete more tasks with hyperthreading.

• If you're more interested in primes than score, you're most likely to find primes on GFN-15 and least likely on GFN-17-LOW.

• GFN-15 primes are too small to be reported to T5K.

• Only GFN-15, GFN-16, and GFN-17-LOW count for the challenge. All of the larger GFN tasks do not count!

 

[Ken g6]

Thanks for bumping this, @zzuupp! The race starts in the middle of the night for me, so I've left my computers running with zero cache.

 

[StefanR5R]

I have my mix of GTX 1080 and 1080Ti hosted by a 2.8 GHz Xeon, which seems to present a bottleneck in getting GPU utilization up for genefer17low.

WCG load + 1 GFN17low per card: ~87 % GPU core utilization
WCG load + 2 GFN17low per card: ~88 % GPU core utilization
no CPU load + 2 GFN17low per card: ~87 % GPU core utilization​

The latter drop is probably because I have EIST on, and Windows 7 pro, and the GPU feeder task presumably hopping between cores which in turn are being wiggled between 1.2 and 2.8 GHz.

Furthermore, genefer17low ignores the <process_priority_special> tag in cc_config.xml and runs the feeder processes at lowest priority. I am running a powershell script now which switches the feeder processes to normal priority, but this doesn't improve CPU GPU core utilization either.

I have long been thinking I should clean up my DC equipment and divide it cleanly into separate hosts for GPU projects vs. CPU projects. (The former with inefficient high-frequency CPUs, the latter with efficient CPUs.)

PS, my GPUs run capped by reliability voltage. Power consumption is below 80 % on 1080Ti and slightly above 80 % on 1080, while a moderate GPU core overclock is applied (against Roger's recommendation), and 52...57 °C GPU temperature (under water and with moderate fan speeds).

Edit,
I have yet to compare actual throughput of 1/card vs. 2/card. No time for this right now.

 

[crashtech]

With Primegrid, it seems like a different app_config is needed for every project and every different GPU. My 1070 is only getting about 85% utilized atm. Might be something for SSC...

 

[StefanR5R]

There is a GTX 1070 of mine which hasn't been put towards PG yet. I'll see how that one will fare in a few hours hopefully; it has a 4.2 GHz host CPU.

Edit,
It is utilized by 89...90 %, whether I run one task or two tasks at a time.

 

[StefanR5R]

Genefer 17 Low 3.18,

GTX 1080Ti @ 1.97 GHz, host CPU @ 2.8 GHz:

1 task per card: 5m32s mean task duration (10.83 tasks/h, 109.0 kPPD)
2 tasks per card: 11m03s mean task duration (10.86 tasks/h, 109.2 kPPD)
sample size: 26 tasks each​

GTX 1080 @ 1.99 GHz, host CPU @ 2.8 GHz:

1 task per card: 7m00s mean task duration (8.57 tasks/h, 86.2 kPPD)
2 tasks per card: 13m53s mean task duration (8.64 tasks/h, 86.9 kPPD)
sample size: 10 tasks each​

Edit,
GTX 1070 @ 2.01 GHz, host CPU @ 4.2 GHz:

1 task per card: 7m29s mean task duration (8.01 tasks/h, 80.6 kPPD)
2 tasks per card: 14m53s mean task duration (8.06 tasks/h, 81.2 kPPD)
sample size: 5 tasks each​

Windows 7, driver version 384.94.

 

[StefanR5R]

@yodap measured 2906 p/h = 69.7 kPPD on a 980Ti. In proportion to CUDA core count and clock, this is quite a bit more than my Pascals do. (But not in proportion to power draw.)

 

[Ken g6]

Half-day stats:

Rank___Credits____Username
33_____176950___10esseeTony
35_____169748_____xii5ku
48_____115897_____crashtech
93_____57853______Orange Kid
104____51147______IEC
114____45252______Ken_g6
191____18967______zzuupp
226____11130______SlangNRox
365____1008_______Kiska

Rank__Credits____Team
7_____700526_____AMD Users
8_____670905_____US Navy
9_____653674_____Rechenkraft.net
10____647954_____TeAm AnandTech
11____514221_____BOINC@Poland
12____446138_____Storm
13____401816_____Team 2ch

Well at least we're in the top ten. My 1060 does one every 10 minutes, but my 750ti does one every 20. So several slow cards can beat one fast one. This seems like a perfect project for @VirtualLarry.

 

[zzuupp]

Thanks for the stats, Ken.

Something oopsed this afternoon. I got home this evening and 'trip' was frozen solid.

I guess that's one advantage to an ATI hair dryer, you can hear it when it's crunching.

 

[Ken g6]

Roughly halfway stats:

Rank___Credits____Username
36_____567809_____xii5ku
42_____465370_____crashtech
54_____367554_____IEC
61_____325428___10esseeTony
92_____196171_____Orange Kid
126____137508_____Ken_g6
208____57775______zzuupp
232____45121______SlangNRox
258____33563______GLeeM
319____14104______Kiska
417____4029_______catavalon21

Rank__Credits____Team
6_____2424406____AMD Users
7_____2385394____BOINC@MIXI
8_____2328945____Rechenkraft.net
9_____2214437____TeAm AnandTech
10____1985554____US Navy
11____1727750____BOINCstats
12____1720960____BOINC@Poland

Up one place from last time, but a little more power would get us several more places. Also, welcome to catavalon21!