Newer milestone thread

[S485122]

Originally Posted by kriesel

The lone last DC assignment before the 60M DC milestone is reached was reassigned to an account that has made zero progress so far, with a projection of expiration before completion (8 d vs 10). https://www.mersenne.org/report_expo...9917901&full=1
https://www.mersenne.org/assignments...rst=1&excert=1
A RadeonVII could complete it in several hours.

Let the user finish another exponent first.

[slandrum]

Originally Posted by kriesel

The lone last DC assignment before the 60M DC milestone is reached was reassigned to an account that has made zero progress so far, with a projection of expiration before completion (8 d vs 10). https://www.mersenne.org/report_expo...9917901&full=1
https://www.mersenne.org/assignments...rst=1&excert=1
A RadeonVII could complete it in several hours.

I would let it expire out and get reassigned. 60M is a "big" milestone, but not that big or urgent. There's also been pointless poaching going on in DC that's just ending up in wasted resources.

[kriesel]

Originally Posted by S485122

Let the user finish another exponent first.

Right. IF it landed on an mprime/prime95 instance with the default 3 day work queue setting still in place, it should begin making progress in another day or so. Once it subsequently reports some progress in, within a day after that, it will get an extension of expiration, from 10 days since assignment, to 30 days, as a Cat 0 DC assignment. Perhaps this fourth LLDC attempt will finish before assignment expiration. (At ~17 months since first DC was assigned. Then we'll see whether the res64s match or ~4% probability an LLTC is needed, with more possible delay.)
I've set most of my DC engines to 1 day queue and DC with PRP/proof instead of LLDC, to reduce those effects. In prime.txt:

DaysOfWork=1
WorkPreference=155

PRP/proof is quicker and more efficient and more reliable, than LL DC with the combined error rate of ~2% first LL and ~2% LLDC resulting in ~4% probability of LLTC possible tiebreaker and occasional LLQC etc. if staying with the LL route. Effort to verify a composite by LL is ~1.041 tests, vs 1.005 with PRP/proof8. Assuming the current LLDC is with Jacobi check, its likely error rate halves, but the LL first test was done before Jacobi check was added, so that remains.

[congsz]

All tests below 60M verified. (but in UTC 08:39 Apr. 5 2022 the mersenne.org milestone page has not been updated yet)

[Uncwilly]

Originally Posted by congsz

All tests below 60M verified. (but in UTC 08:39 Apr. 5 2022 the mersenne.org milestone page has not been updated yet)

That has some human in the loop involvement. It will be updated soon enough.

[tuckerkao]

Originally Posted by Zhangrc

Seems that they are getting PRP without any P-1.

v308	108356371	F-PM1	2022-03-23 02:25	8.5	Factor: 996452764760981189944109089 / (P-1, B1=471000, B2=41197002)	18.0118
v308	108327223	F-PM1	2022-03-21 03:26	6.5	Factor: 2101743747139768387760441 / (P-1, B1=439000)	2.6107
v308	108320761	F-PM1	2022-03-17 09:51	2.8	Factor: 60009555748293413447823073 / (P-1, B1=463000, B2=40465194)	17.6942

3 factors were found in about 50 trials.
And the factor of 108327223 is smooth enough that could be found by bounds as low as B1=50000, B2=100000.
The only possible explanation is that they had no prior P-1 done.

M108356371 was a strong example why P-1 should be finished before additional trial factoring efforts were added on. I only ran TF from 2⁷⁶ to 2⁷⁷ because it was your assignment that you wanted me to finish during Dec 2021.

M108355781, M108355913, M108356797, M108357871 also.

I always do this way in the free upper areas -> NF-PM1 on M120202123 <- P-1 before the final recommended bit of GPU72, or F-PM1 on M115173323.

[Uncwilly]

This graph shows the gap (in days) between the current DC milestone and when the FTC milestone was the same or lower. I don't know when the last time the minimum GHz-days output for a machine to get FTC's was moved up. It might be time.

[firejuggler]

Seem accurate.
I recently had 3 DC of M64 xxx xxx which were first tested in 2014.
And a PRP that will end on the first of june. maybe the PRP will reduce the time gap in what? 4 or 5 year?

[slandrum]

Originally Posted by Uncwilly

This graph shows the gap (in days) between the current DC milestone and when the FTC milestone was the same or lower. I don't know when the last time the minimum GHz-days output for a machine to get FTC's was moved up. It might be time.

I believe it was moved up within the last year. I know that the requirements listed on https://www.mersenne.org/thresholds/ changed within the last year.

[slandrum]

I've noticed that my resources that are set on "Do what makes the most sense" recently have been getting more FTC assignments and fewer DC assignments. This may be the server responding to the massive drop in the amount of FTC work that's being done over what was going on a few months ago.

[kriesel]

ETA 5 days on M108938483 (James Carlson)