On to 999M -- to 70 Bits

[petrw1]

Basically I'm looking at the effort remaining to factor all exponents to 999M to 70 bits
I defined the lower bound as 30M but since all exponents from 30M to 100M are already at 70 bits I am only looking at the 100-999M ranges.

My spreadsheet takes the numbers from: http://www.mersenne.ca/status/tf/0/0/2/0
I've been ciphering and have a few months of stats now....since the above report has been accurate and complete:
1. I take the number of exponents currently at each bit level below 70;
2. I calculate the GhzDays to take those to the next bit level. To make it manageable I:
-- use the GhzDays to TF the 'mean' exponent up 1 bit level
-- multiple the count by that number
-- and 'ass-u-me' that range will average out
3. Calculate the expected number to be factored and carry the remaining to the next bit level column
4. Continue up to to 70 bits.

And here is a 3-month summary:

	66-Exp	        66-Days	67-Exp	        67-Days	        68-Exp	        68-Days	        69-Exp	        69-Days	        Total-Days
Aug-24	8,762,562	351,170	14,371,316	1,135,950	16,599,700	2,491,221	16,636,680	5,019,375	8,997,716
Sep-30	7,931,520	312,019	13,685,155	1,103,546	16,485,174	2,473,561	16,527,300	4,987,211	8,876,337
Oct-30	6,996,815	275,606	13,337,018	1,086,349	16,373,884	2,457,974	16,405,450	4,946,490	8,766,419

The 66-Exp is the current number of exponents at the 66 bit-level.
The 67-Exp is the current number at the 67 bit-level plus the expected exponents to move up as un-factored from the 66 bit-level.
At the current linear rate of 100,000 GhzDays per Month we have over 7 years to go....Moore's Law to the rescue.

[chalsall]

Originally Posted by petrw1

At the current linear rate of 100,000 GhzDays per Month we have over 7 years to go....Moore's Law to the rescue.

A complete and utter waste of time of course.

But people should do whatever rocks their boat....

[petrw1]

Originally Posted by chalsall

A complete and utter waste of time of course.

But people should do whatever rocks their boat....

Completely agree .... and I have no intention of focusing any efforts there.
Though several are.

I just find it an "interesting" milestone to watch.

[TObject]

For 9E8 range (and maybe some lower ones too), I would recommend GPU trial-factoring bit levels 65 to 72 in one swoop. Otherwise more time is wasted loading and unloading, while taking the exponents one bit level at a time.

[LaurV]

Originally Posted by TObject

For 9E8 range (and maybe some lower ones too), I would recommend GPU trial-factoring bit levels 65 to 72 in one swoop. Otherwise more time is wasted loading and unloading, while taking the exponents one bit level at a time.

Actually not, I didn't try mfakto on that, but mfaktc goes better if only to 70. Of course, the "less classes" version is a must here (or chose 420 classes if you use the version that lets you chose), to get a serious speedup. The one with 4620 classes seems indeed losing time with overhead when the time per class is very small.

OTOH, I share the opinion that is not worth to concentrate your resources there. If you "really" and "badly" need and want to do some LMH, better take all 332M to 80 bits or so. More useful for some big guns doing LL in that area.

But the best use of TF resources right now, yes, is the LL front, like 75M to 75 bits (or 74 bits only, if you have a card that is very good at LL/P-1/FFT, don't do the 75th bit, and better do some P-1 for the same expo instead).

[petrw1]

I compute approximately 6.4M Ghz Days of work left.

YYYY-MM	   67-Exp 67-Days     68-Exp   68-Days     69-Exp   69-Days Total-Days	
2016-11	4,389,521 303,507 13,240,976 1,989,772 13,493,464 4,117,564 6,410,843

[petrw1]

Here is the forum for finishing to 75Bits up to 100M

http://www.mersenneforum.org/showthr...178#post446178

[petrw1]

YYYY-MM	 68-Exp	    68-Days	     69-Exp	    69-Days	   Total-Days
2017-04	9844184	1541819.288	11028563.67	3526501.028	5,068,320.316

[Gordon]

Originally Posted by petrw1

I compute approximately 6.4M Ghz Days of work left.

YYYY-MM	   67-Exp 67-Days     68-Exp   68-Days     69-Exp   69-Days Total-Days	
2016-11	4,389,521 303,507 13,240,976 1,989,772 13,493,464 4,117,564 6,410,843

That's only 33 years (calendar) on a SINGLE gtx-970. Doesn't seem that daunting a task...

[petrw1]

Only took 4 years and a bit.

Note my initial estimate was 7 years so Moore did help.

[ATH]

Current ranges for minimum factored bit depths:

2^64:   0 M	- 1000 M	Link
2^65:   0.044 M	- 1000 M	Link
2^66:   0.380 M	- 1000 M	Link
2^67:   0.795 M	- 1000 M	Link
2^68:   1.5 M	- 1000 M	Link
2^69:   9.934 M	- 1000 M	Link
2^70:  19.965 M	- 1000 M	Link

2^71:  29.68 M	-  200.1 M	Link
2^72:  39.56 M	-  180.3 M	Link
2^73:  44.9 M	-  100 M	Link
2^74:  56 M	-  94 M		Link
2^75:  70 M	-  93.52 M	Link
2^76:  82.91 M	-  93.04 M	Link