Factorial - TOP GUN Challenge - 13-27 May

On March 3, 1969 the United States Navy established an elite school for the top one percent of its pilots. Its purpose was to teach the lost art of aerial combat and to insure that the handful of men who graduated were the best fighter pilots in the world. They succeeded. Today, the Navy calls it Fighter Weapons School. The flyers call it: TOP GUN.

Top Gun Day is May 13th:
http://www.topgunday.com/2009/02/top-gun-day-is-may-13th/
Generate your Call Sign and prepare for the Top Gun Challenge.

Challenge at the FPS (Factorial prime search project) from the 13th till 27th of May. Rollcall is at 02:00 UTC and graduation from Top Gun flying school will be at 02:00 UTC fourteen days later, if you make it that far.

In addition to the usual stats I'll be running 1 day training sorties.
Teams will be divided into Student Aviators against TOPGUN Instructors.
Winning side gets split into Students and Instructors for the next training sortie with handycaps based on previous sortie until one Team is victorious at the end.
eg. Team rank 1,3,5,7,... TOPGUN Instructors vs Team rank 2,4,6,8,... Student Aviators, based on http://u-g-f.de/PRPNet/allteam_stats.php?proj=FPS
Say Instructors win, then next sortie is 1,5,.. vs 3,7,... with handycap (sum side this sortie/sum side teams scores from previous sortie).
Maverick:

This is what I call a target rich environment.

I feel the need...

...the need for speed!

Does pfgw take advantage of the avx/fma extensions?

"Sentry" here and ready to fly, sir!

I've got 16 AVX cores ready for afterburners just 2.5 hours after challenge go time (PRPNet is a hard one to start remotely from work), and if KWSN needs it, I've got 4 more on a stinky gifted laptop and 4 Nehalem cores.
____________
Eating more cheese on Thursdays.

Is there some indication of the approximate length of time that a task will take? I fear that my poor i5 650 may not be up to the task....

So, in simple terms: since I am not flying a jet, but a kite, lol. How do I participate in this contest? Do I have to empty my entire que? What do I download to participate for best results?

Thank you

Just testing the waters with this... a single unit running on a stock i7-2600 looks like it will take around 4h40m. It will likely slow down if I run 4 units at a time...

WUs for the Challenge get accepted if returned during the challenge, don't have to be started during the challenge. So you can start now!

Running 4 cores now. Might set up another 4 cores shortly.

I note there is a newer version of pfgw than that included in latest prpnet, and the changelog for it states it uses a newer version of gwnum. Anyone who knows what's what with the versions say if there's any benefit to factorial primes from that?

First returned WU's are on the board:
http://u-g-f.de/PRPNet/challenge_stats.php

Viper:

In case some of you are wondering who the best is they are up here on this plaque.
[turns to Maverick]
Do you think your name will be on that plaque?

Yes sir.

That’s pretty arrogant, considering the company you’re in.

Yes sir.

I like that in a pilot.

Anyone who knows what's what with the versions say if there's any benefit to factorial primes from that?

Looks like even the latest pfgw doesn't (yet) use the latest gwnum with AVX fix.

In other news, units are rolling slowly. Running 4 instances seems to have more than doubled the crunching time compared to running a single unit. I'm now looking at an estimated 11+ hours a unit compared to the 4h40m estimated when running a single unit on the i7-2600. I have to wonder if running some other number of units would give maximum throughput...

In other news, units are rolling slowly. Running 4 instances seems to have more than doubled the crunching time compared to running a single unit. I'm now looking at an estimated 11+ hours a unit compared to the 4h40m estimated when running a single unit on the i7-2600. I have to wonder if running some other number of units would give maximum throughput...

The ram should be in dual channel mode, but I can't double check until I'm home later.

http://www.primegrid.com/forum_thread.php?id=6135&nowrap=true#83276

If above link works, this is some previous testing I did on a Haswell box, so might not be exactly the same for Sandy Bridge but likely close enough. Certainly on there, for some of the bigger prime tests, the relative performance drops significantly when running 4 tests at the same time compared to running only one. Of the projects I tested then, ESP was amongst the worst for this effect, and the relative throughput of 4 instances was only 2.6x that of running 1 instance. There was still a gain over running 2 or 3 instances overall.

Can anyone work out roughly how big are the factorial numbers currently being tested? At least compared to the likes of ESP, TRP or even SoB. It might not be an exact comparison but it would give a rough indication of the size of the work.

Edit: thinking more, in other projects isn't there a speed difference between +1 and -1 forms? Could that be a contributing factor here also? I haven't checked what form the primes were in when I was looking at their speeds previously.

Size of the current leading edge:

pfgw64 -od -q"175954!+1" | wc -m

Approximately 846586 digits (there's some boilerplate text in there).



compute strlen(gmp_strval(gmp_fact(175954)))
(simple program I wrote that passes the expression to php)

846537 digits exactly

Some timings from a Ivy Bridge i5 with DDR3 1600 RAM.
1 core: 5.75 hrs / test
2 core: 6.4 hrs / test = 1.8x thruput
3 core: 8.0 hrs / test = 2.2x thruput
4 core: 9.75 hrs / test = 2.4x thruput

The 3rd and 4th cores are giving only modest increases. I think I figured out the reason. The FFT size being used is 288K which equals 2.25MB. Thus two such FFTs (plus overhead) should fit neatly within the 6MB cache of my i5. But 3rd and 4th oveflows that and memory comes into play much more heavily.

I believe i7-2600 has 8MB cache, and hence 3 such test should fit in. However the 4th would cause much less increase in thruput. But 4h40m to 11+h is a bit much. Probably because of slower 1333 RAM or maybe because of single channel mode. Of course, turbo boost might also be in play with the 1-core result. EDIT:- You should try with just 3 instances. That might be the best configuration for you.

Some example times from different machine families:

Xeon E-1650 (Haswell), 32GB DDR4, 15MB cache, 6 cores -- ~7 hours

i7-2670QM, 8GB DDR3, 6MB cache, 4 cores -- ~12 hours

i7-860, 12GB DDR3, 8MB cache, 4 cores -- ~17 hours

AMD 1055T, 4GB DDR2, 6MB cache, 6 cores -- ~ 39 hours

QX6700 (stock speed), 8GB DDR2, 8MB cache, 4 cores -- ~ 43 hours


Obviously, the latter two machines are getting beyond cache and into much slower DDR2 RAM. I am going to try 3 cores on the QX6700 to see if that speed it up much and probably 3 cores on the 1055T to see that effect.



EDIT: fixed a couple of errors in cpu cache listing.

My time estimates were based on observing the time taken to do 2500 steps as reported. I repeated the single instance test and got about 15 seconds each time. I only did the 4 instance once, and that was 30-something seconds, I can't remember exactly and probably should repeat it.

Good point on the turbo. I disabled it for the Haswell testing, but my normal desktop has it enabled. So that would be a single core at 3.8 GHz vs. 4 cores at 3.5 GHz. On clock alone that is some 8%. My SB system isn't overclocked, so in probability I am running 1333 ram.

Just did some more testing - again based on extrapolating from time per 2500 steps measured manually to within a second or so.

Results are given in the order:
Instances, seconds per 2500 steps, estimated unit hours, system units per 24h, relative speed compared to 1 instance running. Note this includes the effects of any built in turbo clock function going on when running less than full cores.

Sandy Bridge i7-2600k (8MB cache, dual channel DDR3-1333)
1, 15s, 4.7h, 5.1, 1.00x
2, 18s, 5.6h, 8.5, 1.67x
3, 26s, 8.1h, 8.9, 1.73x
4, 35s, 10.9h, 8.8, 1.71x

When running 4 instances, the units were finishing just over 11h so the about prediction isn't too bad.

Wow... it does look like running 3 units is faster. I'm going to let the current 4 continue for now, but have set the 4th instance to stop. From then on, I'll also edit a batch file set for maintaining 3 instances.

Shocked by the above results, I had to do the same on my Haswell box.
Haswell i5-4570S (6MB cache, dual channel DDR3-1600)
1, 11s, 3.4h, 7.0, 1.00x
2, 16s, 5.0h, 9.6, 1.38x
3, 21s, 6.6h, 11.0, 1.57x
4, 29s, 9.0h, 10.6, 1.52x

Again, it looks like 3 instances is the sweet spot. Note in both cases, it is a small gain from running 3 instances compared to 4.

Again, it looks like 3 instances is the sweet spot. Note in both cases, it is a small gain from running 3 instances compared to 4.

I screwed up my user name on one of my clients. So now there is a brinktastee and a brinktasteee. Any way to merge the two? I would be tied with 1st (at the time of this post) if it was not for that stupid mistake.

"You never, never leave your wingman."

I screwed up my user name on one of my clients. So now there is a brinktastee and a brinktasteee. Any way to merge the two? I would be tied with 1st (at the time of this post) if it was not for that stupid mistake.

Announced. Tomorrow is first day training sortie! 15th May 2015
See first post for details.

First Training Sortie 15/05/2015
TOPGUN Instructors
Aggie_The_Pew
Russia
SETI.Germany
L'Alliance_Francophone
The_Knights_Who_Say_Ni!
JISATORA@W-ASCII
Free-DC
MobileUsers
BOINC.SK
My_Way
UK_BOINC_Team
None
Team_2ch
cgkiii
4244
AMD_Users
IBM
L_Alliance_Francophone
US_Navy
Minnesota_Crunchers
PrimeSerarchTeam
BOINC@Heidelberg
Ars_Technica
Rechenkraft.net
RaidersOfTheLostPrimes
PolishNationalTeam
Blasfemias
BOINC.Italy
Portugal@Home
Heinlein_Fans
Sleeping_Dogs
Gentoo_Linux_Users_Everywhere
59
BOINC_Confederation
Siciturdastra.
BOINC@Finland

Student Aviators
PrimeSearchTeam
Team_JPN
USA
Sicituradastra.
Duke_University
Solar_Extreme
CzechNationalTeam
37
Team_Ninja
TeAm_AnandTech
Ukraine
Overclockers_Australia
[AF>Le_Pommier]
BOINC@Poland
Siciturastra.
Team_Germany
BOINC.BE
Crunchers@Freiburg
Dutch_Power_Cows
SETI_USA
539
Ultimate_Chaos
5.0.5
France
BOINCstats
x-beyond_the_frontier
OS/2_Warp
SwissTeam.NET
the-three-stigmata-of-palmer-eldritch
Arc_Technica
Astronomy.Ru_Forum
SETI.hr
Team_Anandtech
l_alliance_francophone
Crunching@EVGA

Any team not returning a WU today will not be considered for the next training sortie. Who will win? TOPGUN Instructors or Student Aviators?

[Song]

Highway To The Danger zone...

After 9 hours flying in the first Sortie we have:
- TOPGUN Instructors 160 WUs
- Student Aviators 60 WUs

Come on PrimeSearchTeam and Sicituradastra.
Time to pull out the big guns to keep the Student Aviators in it (including me).

First Training Sortie complete:
- TOPGUN Instructors Total 488 WUs
- Student Aviators Total 175 WUs

Congratulations to the following teams who have made it onto the next round:
Aggie_The_Pew 252 WUs
Russia 150 WUs
The_Knights_Who_Say_Ni! 62 WUs
Ars_Technica 12 WUs
FREE-DC 4 WUs
UK_BOINC_Team 3 WUs
SETI.Germany 3 WUs
Team_2ch 2 WUs

We have 8 teams so there will be 3 more Training Sorties.

Goose:

No. No, Mav, this is not a good idea.

Sorry Goose, but it’s time to buzz a tower.

Announced. Tomorrow will be the second training sortie 18th May 2015.

TOPGUN Instructors
Aggie_The_Pew
The_Knights_Who_Say_Ni!
FREE-DC
SETI.Germany
handicap: 321 WUs

Student Aviators
Russia
Ars_Technica
UK_BOINC_Team
Team_2ch
handicap: 167 WUs

Handicaps are based on the previous sortie. Winning side this time will be determined on how much more firepower they can bring to bear.

Right now:
TOPGUN Instructors are 146 WUs
Student Aviators are 101 WUs

Taking hadicaps into account:
TOPGUN Instructors are (146*167)/(321*101) = 0.752
Student Aviators are (101*321)/(167*146) = 1.330

Merlin:

What are you doing? You’re slowing down, you’re slowing down!

I’m bringing him in closer, Merlin.

You’re gonna do WHAT?

Second Training Sortie complete 18th May 2015:
- TOPGUN Instructors Total 327 WUs
- Student Aviators Total 198 WUs

Compared to First Training Sortie:
- TOPGUN Instructors Total 327/321 = 1.019
- Student Aviators Total 198/167 = 1.186

Congratulations to the following teams who have made it through to the next round:
Russia 150+173 WUs
Ars_Technica 12+8 WUs
UK_BOINC_Team 3+5 WUs
Team_2ch 2+12 WUs

So we have 4 teams left and 2 more Training Sorties.

Charlie:

I’ll have what he’s having. Hemlock is it?

Ice water.

Almost 6.5 days into the 14 day TOP GUN challenge.
So far we've done 4,233 WUs for 31,151,653 points.
That's 5.35% of the WUs ever crunched in the FPS sub-project, 9.94% of the points!
Leading edge is now at n=181,500. That's an advance of 4.3% from when the challenge started.

Goose:

Great balls of fire!

Announced. Tomorrow will be the third training sortie, 23rd May 2015.

TOPGUN Instructors
Russia
Team_2ch
handicap: 337 WUs

Student Aviators
Ars_Technica
UK_BOINC_Team
handicap: 28 WUs

Handicaps are based on the sum of previous sorties. Winning side will be determined by how much more firepower they can bring to bear and get a chance to compete in the final round, shooting it out to be crowned TOP GUN.

After 8 hours into the third sortie we have:
TOPGUN Instructors
Russia: 94
Team_2ch: 7
TOPGUN Instructors are at 101 WUs = 30% of their handycap (101/337)

Student Aviators
Ars_Technica: 10
UK_BOINC: 0
Student Aviators are at 10 WUs = 36% of their handycap (10/28)

Can Russia and Team_2ch hit their after burner in time?

Third Training Sortie complete 23rd May 2015:
- TOPGUN Instructors Total 187 WUs
- Student Aviators Total 23 WUs

Compared to their Previous Training Sorties:
- TOPGUN Instructors Total 187/(150+173+2+12) = 0.555
- Student Aviators Total 23/(12+8+3+5) = 0.821

Congratulations to Ars_Technica and UK_BOINC_Team who made it to the final TOPGUN round:
Ars_Technica 12+8+16
UK_BOINC_Team 3+5+7

Viper:

Damn, this kid is good!

...Congratulations to Ars_Technica and UK_BOINC_Team who made it to the final TOPGUN round:
Ars_Technica 12+8+16
UK_BOINC_Team 3+5+7

I've been riding on the coattails of others to get here but on behalf of my team it's time to concede victory and pull out of the final. My trigger finger is in need of some r & r and my only F-14A is going to need some time in the hanger.

11.5 days into the 14 day TOP GUN challenge.
We've now done 7,679 WUs for 58,212,405 points.
That's 9.3% of the WUs ever crunched in the FPS sub-project, 17.0% of the points!
Leading edge is now at n=186,460. That's an advance of 7.1% from when the challenge started.

Final training sortie 26th May 2015.

TOPGUN Instructor
Ars_Technica
handicap: 36 WUs

Student Aviator
UK_BOINC_Team
handicap: 15 WUs

Handicap is the sum of teams WUs during previous the sorties. Winning teams are distrubuted 1,3,5,.. 2,4,6,.. between the two sides. Losing teams scores don't contribute to the next round handicaps.

Which team will be crowned TOPGUN?

Final training sortie is complete.

Congratulations to Ars_Technica 2015 TOPGUN!

Stinger:

They gave you your choice of duty son, anything, anywhere. Do you believe that? Where do you think you wanna go?

I saved the stats here and made it available in the formerly list

note: brinktasteee was merged into brinktastee, but there are only effects in the ranking at positions 31, 32 and 33 (each one better because brintasteee was eliminated)
____________
Sysadm@Nbg
my current lucky number: 113856050^65536 + 1
PSA-PRPNet-Stats-URL: http://u-g-f.de/PRPNet/