• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

PrimeGrid Winter Solstice Challenge December 18-21, with GPUs!

Ken g6

Programming Moderator, Elite Member
Moderator
The year's almost over, so it must be time for another PrimeGrid Winter Solstice Challenge. 🙂 Once again we're trying to light up (or at least heat up) the darkness with another GPU sieve challenge. This challenge also happens to use a sieve client I wrote, so I have a few more suggestions than usual. :sneaky: Also note the unusual times below!

What you need to participate:
  • One or more fast x86 processors. (Even slow ones will do!)
  • Windows, Linux, or MacOS 10.4+.
  • BOINC, attached to PrimeGrid (http://www.primegrid.com/).
  • Your PrimeGrid Preferences with only "Proth Prime Search (Sieve)" selected in the Projects section. (It's the one marked as this month's Challenge project.)

What would help a whole lot:
  • An nVIDIA GPU, of the 8000 series or later (so it's CUDA capable). A mid-range GTX 460 does about 10 times as much work as a mid-range quad-core CPU! 😱
  • An AMD GPU, of the 4000 series or later (so it's OpenCL capable), with the 10.10 or later Catalyst drivers, of the APP variety. A mid-range Radeon 5770 does 2-3 times as much work as a mid-range quad-core CPU.
  • "Use ATI GPU" and "Use NVIDIA GPU" set to "yes" in your PrimeGrid preferences.
  • Windows users should make sure they have msvcrt90.dll. You can get it by installing the .NET 3.5 runtime. (Not sure if this is still necessary, but it's listed just in case.)
  • Users with multiple GPUs should add a cc_config.xml file.

What would help a little, but isn't absolutely necessary:
  • A 64-bit OS on a 64-bit processor will run CPU work 30% faster than a 32-bit OS. It matters very little for GPU work. If you can't get a GPU working on a 64-bit OS, but it does work on a 32-bit OS, go back to the 32-bit OS.
  • Overclocking a little. This project produces less heat than the LLR challenge from last month, and it's more tolerant of errors - except on GPUs, so watch out!
  • An app_info.xml file may be useful, especially with a non-standard addition:
    • <cmdline>-m64</cmdline> may improve performance a few percent on Fermi-based nVIDIA cards. Other values may improve performance a little on other cards -- try them and see!
    • <count>0.5</count> would let two computations run at once on Fermi-based nVIDIA cards. This generally doesn't seem helpful, but you're welcome to try it anyway.

What won't help (but won't hurt either):
  • A large amount of RAM.
  • Any older video cards or Intel video cards - though you might want to use them for your display while newer cards crunch.

This challenge starts on December 18, at 17:11 UTC and runs until the Winter Solstice three days later. (Note the unusual start and end times!) Only WUs downloaded after the start of the challenge and completed before the end of the challenge (and before the end times) will be counted. Should you not be able to be in front of one or more computers at the start time, either set their queue level to zero (0) days or try one of the scheduled start methods from the strategy thread (for Linux or Windows). Network connection time options may also be useful - though they depend on your computer's clock being correct, so set the start time a few minutes late.

Welcome and good luck to all! :thumbsup:
 
What would help a whole lot:
  • An nVIDIA GPU, of the 8000 series or later (so it's CUDA capable). A mid-range GTX 460 does about 10 times as much work as a mid-range quad-core CPU! 😱
  • An AMD GPU, of the 4000 series or later (so it's OpenCL capable), with the 10.10 or later Catalyst drivers, of the APP variety. A mid-range Radeon 5770 does 2-3 times as much work as a mid-range quad-core CPU.

Between two 7790 1GB cards, or a GTX460 2GB card, which GPU config would be better?
 
I'm in, and getting everything set up.

I've just got to remember tomorrow that 17:11 UTC is a bit earlier than usual
 
Yeah, I had forgotten about the odd start time. Fortunately, my script didn't.
 
Well, I missed your first post until last night. :$ So I guess we're even.

Even if I had noticed it, I wouldn't know how to answer it. I don't know how 7790s perform. I know my OCed GTX 460 is doing one WU every ~32 minutes, so if your 7790s can each do one in an hour or less, they're clearly superior.

I meant to do stats last night, but I forgot that too. :$ So here are the current stats:

Rank___Credits____Username
35_____1402336____Lane42 😱
128____468569_____GLeeM
136____448343_____zzuupp
188____313503_____biodoc
285____155066_____Icecold
303____134840_____Ken_g6
326____111243_____SlangNRox
425____47194______Rudy Toody
469____30339______wayliff
500____23597______geecee

Rank__Credits____Team
16____3647422____US Navy
17____3526066____Ukraine
18____3148514____Duke University
19____3135030____TeAm AnandTech
20____2993448____Universe Examiners
21____2791188____Nemesis Germany
22____2568702____Team China

Looks like Lane42's GTX780 is working out very nicely. :awe:
 
Between two 7790 1GB cards, or a GTX460 2GB card, which GPU config would be better?

I just did a Google search and turned up this page I didn't know about before. It looks like the single 460 is better in this PPS Sieve race. But keep those 7790s around for the GFN race in January.
 
I know my OCed GTX 460 is doing one WU every ~32 minutes
I have a 6870 doing WUs in ~45 minutes.
The 7950 OCed to 1165 is doing them in ~30 minutes.
And my GTX560 Ti core 448 OCed to 825 is taking 15.5 minutes to finish a WU.
 
Thanks for the stats Ken g6 🙂


9.28 for a 780. I would of thought a much bigger gap for biodoc's TI.

I still want your TI biodoc 😀
 
Thanks for the stats Ken g6 🙂


9.28 for a 780. I would of thought a much bigger gap for biodoc's TI.

I still want your TI biodoc 😀

:biggrin: It's only marginally better than the 780 but it's a rush to fire it up!
 
Lazy stats before bedtime!
Movin' on up


Last update: 2013-12-20 03:45:07
Rank Name Score
32 Lane42 .... 2636122.00
113 biodoc .... 967477.00
121 zzuupp .... 910170.00
143 GLeeM ..... 785443.00
164 Icecold ... 680942.00
330 Ken_g6 .... 239341.00
371 SlangNRox ..185405.00
429 Rudy Toody .131469.00
439 geecee .... 117985.00
543 wayliff .... 53936.00
790 Ryan ........ 6742.00

Rank Team Score
11 Aggie The Pew 9546672.00
12 BOINC@MIXI 8676954.00
13 PrimeSearchTeam 8208385.00
14 The Knights Who Say Ni! 7443168.00
15 Keep The Fire Alive! 6772339.00
16 TeAm AnandTech 6715032.00
17 Ukraine 6536369.00
18 US Navy 6458836.00
19 Team China 5670022.00
20 Duke University 5086839.00
21 Nemesis Germany 4871095.00
 
Would you believe it's the halfway point of the race already? Thanks for the stats, zzuupp! I just wanted to add one more late-breaking stat:

15____6994825____TeAm AnandTech
16____6988083____Keep The Fire Alive!

We're on fire! 😉
 
I couldn't figure out why my score is so low. It turns out that I downloaded 2 seconds too early. 20 WUs sent at 17:10:58 UTC!

I need to get back to my Mathematica models, which means I must suspend BOINC for those sessions. I will try to get my remaining 35 WUs reported before the deadline.
 
Last edited:
Day 2 stats:
Rank___Credits____Username
31_____3869908____Lane42
92_____1638306____biodoc
116____1371997____zzuupp
134____1254012____Icecold
148____1098946____GLeeM
347____343842_____Ken_g6
384____276422_____SlangNRox
432____212373_____geecee
439____202260_____Rudy Toody
578____80904______wayliff
881____13484______Ryan

Rank__Credits____Team
12____12543491___BOINC@MIXI
13____11983905___PrimeSearchTeam
14____11144526___The Knights Who Say Ni!
15____10362454___TeAm AnandTech
16____9826465____Ukraine
17____9688254____Keep The Fire Alive!
18____8966860____Team China

Wow, we're so close to our traditional rivals, KWSN. Anybody got an extra GPU to throw at this?

Fred, do your Mathematica models use CUDA or OpenCL? If not, can you keep running PG on your video cards? (Just uncheck the CPU box on the PrimeGrid settings page. You likely have enough CPU work on the rest of your machines to finish out the race.)

And speaking of reporting WUs, sieves now have to be double-checked just like LLR races. So please try to either return or abort all your work, whether you do it before the end of the race or not.
 
Here are the stats with 12 hours to go:
Rank___Credits____Username
30_____4978967____Lane42
86_____2231602____biodoc
118____1779888____zzuupp
133____1645048____Icecold
154____1382110____GLeeM
364____424746_____Ken_g6
398____340471_____SlangNRox
411____316874_____geecee
443____269680_____Rudy Toody
625____101130_____wayliff
655____87646______Ryan

Rank__Credits____Team
12____16292043___BOINC@MIXI
13____15560536___PrimeSearchTeam
14____14188539___The Knights Who Say Ni!
15____13558162___TeAm AnandTech
16____12590685___Ukraine
17____12530007___Keep The Fire Alive!
18____12314263___Team China

Well, I don't think we're going to make 14th, but at least we're closer to 14th than 16th. Also, there is a chance of losing some credit to bad WUs, now that sieves are double-checked. But the chance of that is slim for CPU WUs, and almost nonexistent for GPU WUs, except for the cheaters this change was designed to stop.
 
Back
Top