PrimeGrid Isaac Newton's Birthday Challenge (part 1 of 9 of PrimeGrid Challenge Series)

Gilthanis

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,860
http://www.primegrid.com/forum_thread.php?id=7183#102795

January 3rd to January 13th @ 18:00 UTC time

This is part 1 of 9 in the Challenge Series. The challenge is for the GCW Sieve. This is a CPU based work unit.

Tips:
You must log into your account at PrimeGrid to change your project preferences so that you are only crunching the correct sub project. In this case GCW Sieve.
Intel boxes tend to perform better. The HT probably needs turned on since these are sieve work units. Do your testing in advance.
Intel chips tend to out perform AMD chips
PrimeGrid work will stress hardware harder and so is more sensitive to overclocking. Therefore make sure you have tested their work units on your rigs before the challenge to make sure they are stable.
These challenges only count work units downloaded during and completed before end of the challenge. So, no bunkering.
 
Less than 2 days to go. Get your boxen ready! Challenge starts on Jan 3rd, 1:00pm EST/10:00am PST.
 
Only a few hours away. My rigs will only be partially in it as I have goals elsewhere. But I should have a few results to turn in during this time frame.
 
OK, I've added seven more servers to the two that were already on the project. Might have a couple more; but I'm down a lot compared to a few months ago.

(Whoops. Make that 8. Turned one on, but forgot to reconfigure it. :p )
 
Last edited:
(y) That's still an awesome show of crunching power... Since this is a 10 days long challenge, perseverance is the key. I just started another rig an hour ago. Not much to show.
 
Early placings look like we are sitting at 7th place as a team and individual rankings below. We have 9 days to go.

5 - fastgeek
86 - pututu
102 - Coleslaw
329 - HighTech67
 
Calling RFGuy_KCCO. Where are you?
 
*edit 2 - you can skip this if you'd like... go down a few posts :) /edit *

So am sure this has been mentioned already, know it has been mentioned in other forums, but enabling HT on Haswell (and am guessing Broadwell) is a BAD idea. Sandy is fine and about to verify that Ivy is fine too.

For instance, on one of my 2P servers equipped with 8c/16t Haswell E5-2640 processors. No HT = 75M / wu. Enabled 32 "cpus" via cc_config and the time left counters shot skywards to absurd numbers. Did let the system sit like this for a while, since the counters can take a bit to settle, but the numbers were just nuts. Next step was tinkering with the number of "cpus" via cc_config. Noticed something odd; or at least odd to me. I would expect that with each CPU dropped that the CPU usage level would also drop as well; it did not. It wasn't until BOINC was using 24 "cpus" that the number finally drop below 100% and the time left counters slowly started dropping as well. All of this resulted in 115M / wu... or over 1.5x longer vs. 16 cores.

Disclosure. I'm mentally wiped and focusing on a lot of things at once. Maybe I screwed up somewhere... but AFAIK my info is good. So if you're running HW/BW, stick with the real core values and leave HT to the SB/IB chips. :)

*edit* BTW, remotely enabled HT on three 4P systems and one 2P system. Have two more 4P systems that I'll need to deal with in person in ~8 hours or so. This ought to help a bit. ;)
 
Last edited:
So you are trying to tell me that running 8 work units at a time at 75 minutes each is better than running 16 work units at 115 minutes each? Or are you saying that just because you turned HT on that it increased the run times regardless of how many threads you had going? I'm a bit confused on what you are trying to say. Also, keep in mind that BOINC may try to re-run benchmarks from time to time. Especially if you fiddle with number of CPU's that BOINC sees. (thus the cc_config) In that case the benchmark may ramp your CPU usage up for a small period of time. How long did you wait for CPU usage to come down from 100% before changing it again? More details needed please as many people have traditionally tested their rigs as each generation comes up and most consensus is that sieve work units produce more results over a given time with HT turned on. It would be good to have each generation documented if we can confirm this.

EDIT: We used to try and keep a list on the OLD All Inclusive DC List thread here: https://hardforum.com/threads/all-inclusive-dc-project-list.1801908/ posts 9 and 10.
 
Last edited:
Basically what I'm getting at if I'm reading your first issue correctly. If you run your rig for 2 hours with HT off, you will essentially complete only 16 total work units and have 16 more that have not completed as it would take 2.5 hours to complete 2 full sets on each core. However, if you are running with HT turned on, over the course of 2 hours, you should complete 32 work units and have 32 more partially completed because 1 full set would take 1 hour 55 minutes. So, I'm not seeing how it is better to run with HT off? Could you please explain?
 
OK, I've setup that E5-2640v3 system to run with HT turned on and all 32 "cpus" running. As such, let's forget what I wrote last night; which BTW was about using 24 "cpus" (16c + 8ht).

Here are two WUs I'm tracking that started from 0% with the new 16c + 16ht configuration:
http://www.primegrid.com/workunit.php?wuid=508199038
http://www.primegrid.com/workunit.php?wuid=508199036

Going to wait for these WUs to complete and then going to disable HT in the BIOS vs just cutting 50% of the available cpus from the cc_config file. (Of course I'll edit cc_config to reflect 16 cpus after turning off HT)

Now I have noticed something else.

On a different 2P E5-2640v3 box HT is already turned off. This box results in very consistent spread of 4,350 - 4,750s /wu. Call it an average of 4,500/s or 75m /wu.

On yet another box with the same CPUs, I have HT turned on, but cut by 50% via cc_config. Haven't messed with this box at all since challenge start. Times all over the place ranging from 4,100 to 7,300s /wu. We'll call that an average of 95m /wu.​

I understand that doing the cc_config cut isn't 100% the same as disabling it in the BIOS; but that's just nuts.
 
well...even an average of 95m / wu puts out more PPD than running half as many averaging 75m... Unless your deadline to end of the challenge is in 76m...lol
 
20% done with the challenge and the current rankings are:

2 - fastgeek
74 - pututu
89 - Coleslaw
384 - HighTech67

and the team is in 7th place over all.
 
Those two WUs that were linked in my last thread completed in ~8,300 seconds, or roughly 138 minutes / 2h 18m. So... despite what I thought, and others too on other forms, there doesn't seem to be any disadvantage to running HT on these E5-2640 v3 chips.

Still doesn't explain what the heck is going on with using cc_conf to restrict BOINC to 50% of the CPU; aka the lazy way of "disabling" hyper-threading.

Note that SB/IB boxes are making much better use of HT in this challenge vs. HW. Here are a couple examples

Sandy Bridge 4P E5-4640 (8/16 @ 2.40GHz)
  • With HT turned off in the BIOS, WU's were coming in around 6,600-7,150 seconds.
  • With HT turned on, and BOINC set to use all threads, the range is 10,100-11,700.

Ivy Bridge 4P E5-4640 v2 (10/20 @ 2.20GHz)
  • With HT turned off in the BIOS, WU's were coming in around 7,000-7,450 seconds.
  • With HT turned on, and BOINC set to use all threads, the range is 10,500-12,700.
That works out to roughly 60% longer per WU, but with twice the output for SB/IB; while HW is barely coming in under 2x the time for 2x the output.
 
Some projects do gain from turning HT off. It just depends on the the calculations used, I think it was CSG subset sum benefited with HT off on my 2p 2670s v1. WUProp data has HT on/off in the data if you want to look for more examples.
 
Realized I forgot to finish configuring a 4P SB; so have that up with HT on.

Found another 2P HW I had over looked; it's up without HT as, for whatever reason, it refuses to go over ~58% with it on. Even did a clean install of BOINC, checked all settings, etc. Oh well, small loss.

Am spinning up a 4P Broadwell for long as I can; might get a couple 2P BW's online for a little bit too.
 
fastgeek said:
Those two WUs that were linked in my last thread completed in ~8,300 seconds, or roughly 138 minutes / 2h 18m. So... despite what I thought, and others too on other forms, there doesn't seem to be any disadvantage to running HT on these E5-2640 v3 chips.

Still doesn't explain what the heck is going on with using cc_conf to restrict BOINC to 50% of the CPU; aka the lazy way of "disabling" hyper-threading.

Note that SB/IB boxes are making much better use of HT in this challenge vs. HW. Here are a couple examples

Sandy Bridge 4P E5-4640 (8/16 @ 2.40GHz)
  • With HT turned off in the BIOS, WU's were coming in around 6,600-7,150 seconds.
  • With HT turned on, and BOINC set to use all threads, the range is 10,100-11,700.

Ivy Bridge 4P E5-4640 v2 (10/20 @ 2.20GHz)
  • With HT turned off in the BIOS, WU's were coming in around 7,000-7,450 seconds.
  • With HT turned on, and BOINC set to use all threads, the range is 10,500-12,700.
That works out to roughly 60% longer per WU, but with twice the output for SB/IB; while HW is barely coming in under 2x the time for 2x the output.
Click to expand...
"Luckily" I got one SB and one IB rig for this sieve :pbut these older XEONs probably lost out (a bit) in other PG sub-projects or in WCG for example.:(
 
fastgeek said:
Realized I forgot to finish configuring a 4P SB; so have that up with HT on.

Found another 2P HW I had over looked; it's up without HT as, for whatever reason, it refuses to go over ~58% with it on. Even did a clean install of BOINC, checked all settings, etc. Oh well, small loss.

Am spinning up a 4P Broadwell for long as I can; might get a couple 2P BW's online for a little bit too.
Click to expand...
A new big boy just showed up and took your number #2 spot.:nailbiting:
 
fastgeek, see ATP PG message board here. You might want to caution them on name calling again if you read it that way... or maybe trash talking?
 
Also Scott Brown used to be in Duke University team last year until he switched to ATP team in the last challenge of 2016. Don't know why.
 
Scot Brown said that his old team was kind of dead on his old Duke team, and liked the guys at A.T.P. so he joined them. Pretty much how I read what he said.
That is a lot of crunching CPU power on that team now. We will see how well they do come the next GPU challenge.
 
Last edited:
pututu said:
fastgeek, see ATP PG message board here. You might want to caution them on name calling again if you read it that way... or maybe trash talking?
Click to expand...

Anthony and I have chatted several times via PM; he's a good guy. The whole "wtfgeek" thing was created by someone else after I lost a challenge to Zune-san at the last moment. Someone figured I was wondering WTF happened and just had fun with my name. The next challenge or two I smacked the living daylights out of everyone, so it's all good. :)

As for NM; yeah, he came out of nowhere big time. We'll see if he's in it for the long haul.
 
fastgeek said:
Anthony and I have chatted several times via PM; he's a good guy. The whole "wtfgeek" thing was created by someone else after I lost a challenge to Zune-san at the last moment. Someone figured I was wondering WTF happened and just had fun with my name. The next challenge or two I smacked the living daylights out of everyone, so it's all good. :)

As for NM; yeah, he came out of nowhere big time. We'll see if he's in it for the long haul.
Click to expand...
Good to make some friends at ATP and having some fun during the challenge. ATP managed to lure Scott Brown, so don't think about going there :rolleyes:
 
Maybe you need to lure Scott Brown over to [H].....lol
 
[H]ah [H]ah. We need to make [H] attractive for these folks.... lol
 
So been slowly reeling in Zune-san; however my systems will be down for the weekend. Bad news for me/us... good news for Scott and Zune though. :p
 
No worries mate. We are glad that you are already participating in this challenge. Seeing new comers with large amount of rigs is quite amazing.
 
After 4 full days of [H]ard crunching, we are at 7th place. Pretty much everyone settles on their position ranking barring any significant rig downtime...

Current [H] members position:
Rank Name Score
3 fastgeek 10,860,384
72 pututu 916,970
91 Coleslaw 741,671
382 HighTech67 81,922.9
 
Only 4 days left.

Current [H] members position:
Rank Name Score
4 fastgeek
71 pututu
97 Coleslaw - I started getting work from CAS and RALPH over the weekend... slight delay....
344 HighTech67
 
The servers had a nice nap over the weekend; but they're all being woken up and put back to work. Hope Mr. Brown wasn't planning on keeping that third spot; or that tng* thought they were going to overtake me. ;) :p
 
You can't let the two rats (as they called themselves) be in the top three positions. Well, Mr B actually just moved from Duke U to ATP team recently.
With about 3.5 days left, go fastgeek! Show what you got.
 
Finally spun up three more 2P Broadwells. Have 17 systems running full tilt, one 2P HW that's still offline that I haven't found yet, and maybe another 4P BW that I might add to the pool if warranted. ;)
 
fastgeek said:
Finally spun up three more 2P Broadwells. Have 17 systems running full tilt, one 2P HW that's still offline that I haven't found yet, and maybe another 4P BW that I might add to the pool if warranted. ;)
Click to expand...
You just mowed over Mr B.
 
Well, I did say he shouldn't get too used to the #3 spot; although it's not over until it's over.

That said... just added five more 2P Broadwells (28c/56t each) to the mix. They ought to help. :D
 
With about a day left to the finishing line, we are holding steady at 7th place barring any big WU dump close to the end. Not bad for a team of 4 crunchers :D with fastgeek leading the way and keeping his 3rd spot in this challenge.

Current [H] members position & scoring:
Rank Name Score
3 fastgeek 23,789,511
72 pututu 2,315,080
113 Coleslaw 1,522,622
320 HighTech67 357,172
 
Really wish I didn't have that ~52 hours of downtime... #2 would have easily been mine. (BTW, added another 4P and 2P BW Wednesday evening) As for NeuralMiner though? Damn. Even if I had no downtime + all twenty-four servers from the start, still not sure I would've caught him/her/them.
 
Great push here fastgeek. I know you can do it... Next PG challenge is 15-days CPU only. Let see what you've got. :D
Congrats to ATP for taking the pole position from a strong contender SG. There are still 8 more challenges for the year.
 
You must log in or register to reply here.
Back
Top