Welcome to Summer Paralympics Challenge

To participate in the Challenge, please select only the Extended Sierpinski Problem (LLR) (aka ESP-LLR) project in your PrimeGrid preferences section. The challenge will begin 2 September 2016 18:00 UTC and end 7 September 2016 18:00 UTC. Application builds are available for Linux , Windows and MacIntel 32 bit and 64 bit. Intel CPU's with AVX capabilities will be significantly faster than the ones without (including all AMD CPUs), as this instruction set allows for more computing power.

ATTENTION: The primality program LLR is CPU intensive; so, it is vital to have a stable system with good cooling. It does not tolerate "even the slightest of errors." Please see this post for more details on how you can "stress test" your computer. WU's may take around 6 hours on the fastest/newest computers and more than a day on slower/older computers. If your computer is highly overclocked, please consider "stress testing" it. Sieving is an excellent alternative for computers that are not able to LLR. :)

Highly overclocked Haswell or Skylake (i.e., Intel Core i7, i5, and i3 -4xxx or better) computers running the application will see fastest times. Note that ESP is now running the latest, brand new FMA3 version of LLR which takes full advantage of the new Haswell features. It's faster than the previous LLR app and draws more power and produces more heat. If you have a Haswell or Skylake CPU, especially if it's overclocked or has overclocked memory, and haven't run the new FMA3 LLR before, we strongly suggest running it before the challenge while you are monitoring the temperatures.

Please, please, please make sure your machines are up to the task.

Time zone converter:

The World Clock - Time Zone Converter

NOTE: The countdown clock on the front page uses the host computer time. Therefore, if your computer time is off, so will the countdown clock. For precise timing, use the UTC Time in the data section to the left of the countdown clock.

Scoring Information

Scores will be kept for individuals and teams. Only work units issued AFTER 2 September 2016 18:00 UTC and received BEFORE 7 Septemberl 2016 18:00 UTC will be considered for credit. We will use the same scoring method as for BOINC credit.

Therefore, each completed WU will earn a unique score based on its n value. The higher the n, the higher the score. A quorum of 2 is NOT needed to award Challenge score - i.e. no double checker. Therefore, each returned result will earn a Challenge score. Please note that if the result is eventually declared invalid, the score will be removed.

At the Conclusion of the Challenge

We kindly ask users "moving on" to ABORT their WU's instead of DETACHING, RESETTING, or PAUSING.

ABORTING WU's allows them to be recycled immediately; thus a much faster "clean up" to the end of an LLR Challenge. DETACHING, RESETTING, and PAUSING WU's causes them to remain in limbo until they EXPIRE. Therefore, we must wait until WU's expire to send them out to be completed.

Please consider either completing what's in the queue or ABORTING them. Thank you. :)

About the Extended Sierpinski Project

Waclaw Franciszek Sierpinski (14 March 1882 - 21 October 1969), a Polish mathematician, was known for outstanding contributions to set theory, number theory, theory of functions and topology. It is in number theory where we find the Sierpinski problem.

Basically, the Sierpinski problem is "What is the smallest Sierpinski number" and the prime Sierpinski problem is "What is the smallest 'prime' Sierpinski number?"

First we look at Proth numbers (named after the French mathematician François Proth). A Proth number is a number of the form k*2^n+1 where k is odd, n is a positive integer, and 2^n>k.

A Sierpinski number is an odd k such that the Proth number k*2^n+1 is not prime for all n. For example, 3 is not a Sierpinski number because n=2 produces a prime number (3*2^2+1=13). In 1962, John Selfridge proved that 78,557 is a Sierpinski number...meaning he showed that for all n, 78557*2^n+1 was not prime.

Most number theorists believe that 78,557 is the smallest Sierpinski number, but it hasn't yet been proven. In order to prove it, it has to be shown that every single k less than 78,557 is not a Sierpinski number, and to do that, some n must be found that makes k*2^n+1 prime.

The smallest proven 'prime' Sierpinski number is 271,129. In order to prove it, it has to be shown that every single 'prime' k less than 271,129 is not a Sierpinski number, and to do that, some n must be found that makes k*2^n+1 prime.

Should both of these problems be solved, k = 78557 will be established as the smallest Sierpinski number, and k = 271129 will be established as the smallest prime Sierpinski number. However, this would not prove that k = 271129 is the second provable Sierpinski number. Since the prime Sierpinski problem is testing all prime k's for 78557 < k < 271129, all that's needed is to test the composite k's for 78557 < k < 271129. Thus, the Extended Sierpinski Problem is established.

The following k's remain for each project:

• Waclaw Franciszek Sierpinski (WIKI)
  
• Sierpinski number / The Sierpinski problem (WIKI)
  
• The Sierpinski problem (Prothsearch)
  

What is LLR?

The Lucas-Lehmer-Riesel (LLR) test is a primality test for numbers of the form N = k*2^n − 1, with 2^n > k. Also, LLR is a program developed by Jean Penne that can run the LLR-tests. It includes the Proth test to perform +1 tests and PRP to test non base 2 numbers. See also:

• Lucas-Lehmer-Riesel test (WIKI)
  
• Download LLR by Jean Penné
  

(Edouard Lucas: 1842-1891, Derrick H. Lehmer: 1905-1991, Hans Riesel: 1929-2014).
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Two days until the challenge starts! Some thoughts about ESP:

Looks like my Haswell can do the ESP tasks in roughly 15 hours.

PSP has always had very large tasks, and SoB's tasks are positively huge. But ESP has always been the "small" Sierpinski tasks. Probably shouldn't think about it that way any more. The tests are closing in on 3 million digits and aren't that much smaller than the 321 tasks. ESP overtook TRP a while ago (TRP has a LOT more Ks and therefore progresses much more slowly.)

Unlike all those other projects, ESP is relatively new. It only started about 4 or 5 years ago.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Looks like my Haswell can do the ESP tasks in roughly 15 hours.

Would you like benchmarks posted once we start completing tasks? I expect my Skylakes to be around 9-10 hrs per unit.

Looks like my Haswell can do the ESP tasks in roughly 15 hours.

Would you like benchmarks posted once we start completing tasks? I expect my Skylakes to be around 9-10 hrs per unit.

For Haswell and later quad core systems, ram speed is as important as CPU clock speed. My faster Skylake quad does 960k FFT units in around 9 hours. This particular system has dual rank dual channel 3200 ram. Another system that has a similar ram configuration but only 3000 speed is closer to 10 hours at that FFT size. Expect times to go up a lot if you're running slower ram.

For Haswell and later quad core systems, ram speed is as important as CPU clock speed. My faster Skylake quad does 960k FFT units in around 9 hours. This particular system has dual rank dual channel 3200 ram. Another system that has a similar ram configuration but only 3000 speed is closer to 10 hours at that FFT size. Expect times to go up a lot if you're running slower ram.

I can't claim to have tested every possible scenario, but in general you can look at the combined CPU demand compared against the available ram bandwidth. More CPU power will lead to more throughput, up to such a point where ram becomes the ultimate bottleneck. In such a scenario, going 3 cores instead of 4 might not cost you throughput, but you have an improvement in unit time.

As a very rough guide, for a quad core processor (Haswell+), having dual channel ram at a nominal marketing speed equal to the CPU clock would be practically ram unlimited, or processor limited. In my case I have CPU at 4200 and ram at 3200, and that is still good for a high level of efficiency. Certainly there is a major performance hit of tens of % if I run the same ram at standard 2133.

For Haswell and later quad core systems, ram speed is as important as CPU clock speed. My faster Skylake quad does 960k FFT units in around 9 hours. This particular system has dual rank dual channel 3200 ram. Another system that has a similar ram configuration but only 3000 speed is closer to 10 hours at that FFT size. Expect times to go up a lot if you're running slower ram.

Any opinion on 3 vs 4 units at a time and throughput? I would expect faster ram to favor 4 units, but enough to make a difference?

It won't give you a time estimate, but at least it's very precise to measure throughput.

If you were running forever, yes. As a challenge has a fixed finishing time, you might need to micro-manage workunit lengths (buy running fewer cores) near the end.

For the numbers given previously by Rafael, that works out to be 7.5, 7.5, 8.1, 10.2 hours for running 1, 2, 3, 4 units respectively. Does that correlate?

I can pretty much guarantee that you will be seeing tasks with different FFT sizes and possibly significantly different run times during this challenge. Planning your task strategy down to the minute probably won't work very well since later tasks may take longer to run than you anticipate.

Since ESP doesn't have many Ks to search, n, and therefore size, increases fairly quickly.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Planning to the hour might be good enough. I've had past challenges where I knew I would be late by some hours for a last set if I ran all cores blazing, but reducing the number of active cores near the end would reduce it enough for a fighting chance.

Also... it remains to be seen just how much the leading edge can be pushed. Lower work (resends or whatever) isn't critical, since it will be shorter. Time taken should scale with the n value, at least until we hit the next FFT size. Maybe tasks at the end will be some low % longer than now.

ESP-LLR work units:

If I remember it correctly one has to turn off the HT and select only two cores to to have good results working this work-unit type.

What percent of the CPU should be set?

So 75% of the computer cores would be a good starting point?

Assuming it has the same IPC as first gen Core i3/5/7 CPUs which also didn't AVX, ball park estimate would be 60 hours a unit.

I would suggest trying it regardless, and after letting it run an hour or so see what the estimated time says.

I am tempted to throw into the mix my two Dell laptops with a duo core 2.67GHz P9600 chip .... they did just fine with the ESP and TRP Sieves and I would not mind seeing one or two units complete from each with ESP LLR.

How to join? Or do i join automatically? Automatically i mean that all computed wu's after challenge started are part of Summer Paralympics challenge?

If anyone along the Eastern seaboard of the United States sees what appears to be a huge dust storm heading their way, that is NOT Hurricane Ermine. I was blowing the dust out of my computer. Carry on...
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Let's go folks, time to knock a few K out of this conjecture.

and here we go ...
____________
Sysadm@Nbg
my current lucky number: 113856050^65536 + 1
PSA-PRPNet-Stats-URL: http://u-g-f.de/PRPNet/

Go Go Go! I'm still finishing off a build to add but the running systems are away.

1708 tasks sent out in the first 5 minutes.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

About 7300 tasks sent out in the first hour.

About 16-18 hours running 4 cores on a stock speed Haswell i5-4670K.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

The i7 6700k (4.0 GHz) is showing about a 20-hr runtime per unit per core with all four physical cores running.

No HT with CPU usage set at 50% for all of the Skylake systems.

Does not compute (excuse the pun) with the performance of the i7 6700k.

Also, can the variable FFT account for a factor of 2 in the runtimes?

I am referring to the BOINC Manager and where I have set CPU usage to 50% so as to effectively turn off HT. The times you list are what I would have expected (somewhere around 10-ish).

I wonder of a who delivery first unit completed and validated. If the rules are working so it must be in the current fastest cpu, computer or some system..most suitable for calculating primegrid.. or no..?:-)

http://www.primegrid.com/challenge/2016_6/top_users.html
I wonder of a who delivery first unit completed and validated. If the rules are working so it must be in the current fastest cpu, computer or some system..most suitable for calculating primegrid.. or no..?:-)

The fastest completed ESP task in the database was done by Rafael's Skylake i5-6600K in just under 8 hours.

There's also a number of tasks completed in less than 9 hours. Of the top 25 fastest completed ESP tasks, all were done by Skylake i5-6600K and i7-6700K CPUs.

The fastest completed ESP task in the database was done by Rafael's Skylake i5-6600K in just under 8 hours.

There's also a number of tasks completed in less than 9 hours. Of the top 25 fastest completed ESP tasks, all were done by Skylake i5-6600K and i7-6700K CPUs.

O.O

I feel so important right now....

Still, that's a bit surprising. While 4263mhz core/cache with 3146mhz dual channel dual rank RAM is blazing fast, I would expect other systems (say, 3200mhz 4.3 owners with 8mb of cache) to be even faster, even if marginally. I guess running 3 WU at a time rather than 4 really speeds things up....

Still, that's a bit surprising. While 4263mhz core/cache with 3146mhz dual channel dual rank RAM is blazing fast, I would expect other systems (say, 3200mhz 4.3 owners with 8mb of cache) to be even faster, even if marginally. I guess running 3 WU at a time rather than 4 really speeds things up....

As the numbers earlier showed, running 4 cores still gives a benefit to throughput, but individual unit times will be longer than 3 or fewer.

Is running fewer tasks significant on far slower machines? I currently have a quad core N3530 running 4 tasks. A little extrapolation shows that they will take about 125 hours to finish. If I reduce this to 3 in order to speed things up, should I abort the 4th task rather than suspend it in order to clear it from memory?

Edit: The machine has 2x8Gb 2400 MHz RAM which is dual rank and dual channel. On other projects (ex AP27 and TRP Sieve), this system consistently outperformed the i3 6100 and i3 6300 (as expected due to the greater stock speed).

Problem solved: according to CPU-Z, the RAM is single channel. Someone is going to get a mouthful (dealer for the RAM).

Do you mean dual rank?

If you have 2 dimm slots populated and it is reporting a single channel, that might be location-related.

Are they in slots 2 and 4?

https://us.msi.com/Motherboard/support/Z170A-SLI-PLUS.html#down-manual

I used CPU-Z on my i3 systems and it does report dual channel. I could exchange sticks so as to get the extra processing power of the i7 6700k but I would rather leave things as they are right now. My mistake for not checking the RAM upon receipt.

It looks like the next time the leaderboards are generated (at 02:30 UTC) there will be at least 3 results that have been returned.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

As of right now, three results have been returned, from two computers. Both are Haswell i7-6700K CPUs.

The first was returned by Steve of team Aggie the Pew, at 02:21:14.

23 seconds later DeleteNull of team SETI.Germany returned the second result, and 112 seconds after that he returned the third result.


Congrats on returning the first results! For what it's worth, my Haswell's tasks are at 50% right now.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

I have two e5-2683 V4 cpus with 16 cores each, 40mb cache, no HT. And only two wu are way ahead of the other wu. Very dissapointed with this performance. Single channel memory, it's a HP DL360 Gen8 server.

16 per socket. It only has one stick of ram in one socket. I think that might be the problem. Each cpu should have memory per socket.

Temperature 0 89 degC (192 degF) (Core #0)
...
Temperature 16 92 degC (197 degF) (Package)

After 1 day:

Challenge: Summer Paralympics
App: 20 (ESP-LLR)
(As of 2016-09-03 18:07:47 UTC)

15990 tasks have been sent out. [CPU/GPU/anonymous_platform: 15990 (100%) / 0 (0%) / 0 (0%)]

Of those tasks that have been sent out:

468 (3%) came back with some kind of an error. [468 (3%) / 0 (0%) / 0 (0%)]
1590 (10%) have returned a successful result. [1590 (10%) / 0 (0%) / 0 (0%)]
13932 (87%) are still in progress. [13932 (87%) / 0 (0%) / 0 (0%)]

Of the tasks that have been returned successfully:

1231 (77%) are pending validation. [1231 (77%) / 0 (0%) / 0 (0%)]
342 (22%) have been successfully validated. [342 (22%) / 0 (0%) / 0 (0%)]
0 (0%) were invalid. [0 (0%) / 0 (0%) / 0 (0%)]
17 (1%) are inconclusive. [17 (1%) / 0 (0%) / 0 (0%)]

The current leading edge (i.e., latest work unit for which work has actually been sent out to a host) is n=9704126. The leading edge was at n=9464166 at the beginning of the challenge. Since the challenge started, the leading edge has advanced 2.54% as much as it had prior to the challenge!

Almost 16 thousand tasks were sent out in the first day. We normally do about 300 per day. We're sending them out at about 200 per hour. That's not the surge at the start of the challenge; that's 200 per hour being sent out every hour right now.

Great work everyone. Let's keep it up and see if we can't knock off one of the remaining Ks!
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Thanks for the heads up. just increased the cooling to max. Let's see what happens temp wise.

update:

got them down to 60 C.

My question is, would it be better off with single rank in quad channel or dual rank in dual channel? Maybe it would be a wash?