Factoring humongous Cunningham numbers

[bsquared]

For grins I dusted off this code to see if it still worked and as a comparison (hence all the questions above).

Stage1 and Stage2 both running on a Xeon Phi 31S1P:

./phiecm 135208777652707551627613394323172349287658234884006214941830419038629271216360971380190458698241219665584896965795899 3584 11000000 224        
starting 3584 curves on 135208777652707551627613394323172349287658234884006214941830419038629271216360971380190458698241219665584896965795899 with B1 = 11000000 (726517 primes) and 224 threads (16 curves/thread)
initializing montgomery representation
Building curves took 0.0874 seconds.

10991909, 24
Stage 1 took 3537.8537 seconds.
writing savefiles
starting stage 2

starting at q = 11000220
found factor 1259183377997512211834207021579593490689 in stage 1 in thread 184, vec position 7, with sigma = 3285301936
skipped 10579375 paired primes out of 54935953

Stage 2 took 2200.9643 seconds.
found factor 1259183377997512211834207021579593490689 in stage 2 in thread 217, vec position 14, with sigma = 3815041854
found factor 1259183377997512211834207021579593490689 in stage 2 in thread 103, vec position 10, with sigma = 1355344152

...about 0.98 seconds/curve in stage 1 and 0.61 seconds/curve in stage 2 at B1 = 11M

[xilman]

Originally Posted by Batalov

The site still appears down (was then and still is)

I've contacted the hosting company and asked for an investigation. Best guess is that it will be working hours UK time before anything significant happens.

The ftp server which I use to upload the pages is working just fine.

Sorry about this, but it's beyond my direct control

Paul

[RichD]

My ecmclient is getting GCW work now.

[jyb]

Originally Posted by xilman

I've contacted the hosting company and asked for an investigation. Best guess is that it will be working hours UK time before anything significant happens.

The ftp server which I use to upload the pages is working just fine.

Sorry about this, but it's beyond my direct control

Paul

It seems to be back now. And that previous link had the wrong domain. See this.

[Batalov]

Yep, it is up!

... But no, it is not working: the UPDATE.txt file is empty!

[jyb]

Originally Posted by Batalov

Yep, it is up!

... But no, it is not working: the UPDATE.txt file is empty!

A feature, not a bug. You may notice this line in the explanatory text:

"The file UPDATE contains a list of changes to the files made since 1 April 2016 and here are those reported to me before that date."

[Batalov]

Exactly!
I would expect at least 25 rows in it already. :-)
(and that's just from me, other people must have also contributed)

[Batalov]

# 6^383+5^383, difficulty: 300.99
ECc17-Input number is 36287917053866719785650526120998695279651734974791134151823208042000072571786261956542711381724414523218450146275040333518910642897546255192019209886576681961865035885850245555280408341112885216841333441815115622137706368514173061738967426558336547379986780599522623648651383887996939 (284 digits)
ECc17-Using B1=22000000, B2=192375122956, polynomial Dickson(12), sigma=396537383
ECc17-Step 1 took 215000ms
ECc17-Step 2 took 87996ms
ECc17:********** Factor found in step 2: 125828096035824850471308490303855554743239305664987
ECc17-Found probable prime factor of 51 digits: 125828096035824850471308490303855554743239305664987
ECc17-Probable prime cofactor 288392800949122609426168026708966026618949815661743800160109743798134370484828724587949951879344491901299395725790134532294729101609430625166584918724535106488321167716287488173170790096175943819110712585470352719911739311150650423697 has 234 digits

[WraithX]

Originally Posted by bsquared

Nice!
Out of curiosity, can you say what NVidia card was used to do this? Also, I'm ignorant of how the program works... does the 70-some seconds of CPU usage occur concurrent to the GPU time? i.e., does the GPU operation consume some CPU cycles as well? Or it is up-front CPU initialization or something? I assume the program stops in step 2 after a factor is found, so that the timing there is for 809 curves instead of 832?

The 70 seconds on the CPU is the computation of the "batch product": the product of all primes, and prime powers, up to B1. This happens before any work is done on the GPU. Once the batch product, called s in the code, is computed, the code loops over each bit in s and computes x curves (in this case = 32*26) in parallel on the GPU.

The way gpu-ecm is currently configured, it will try to find all factors that it can. ie, it actually does search all 832 curves for factors. I'll talk with PaulZ about possibly recognizing the -one option for gpu-ecm so that the program will stop as soon as it knows it has found a factor.

Also, frmky, based on "Grid: 26x1x1", are you running these "half-sized" curves on a GTX 970?

[Batalov]

Originally Posted by jyb

It seems to be back now. And that previous link had the wrong domain. See this.

It appears down again.

It's not just you! http://www.leyland.vispa.com looks down from here.

[frmky]

Originally Posted by WraithX

Also, frmky, based on "Grid: 26x1x1", are you running these "half-sized" curves on a GTX 970?

Tesla K20c.
Edit: And don't use that run for timing. That wasn't the only program running on the GPU. For timing purposes, stage 1 on 832 curves on a <= 506-bit number at B1=11e6 take 1640 seconds, or 1.97 s/curve.