GPU to 72 status... - Page 395

chalsall · 2019-10-29, 00:04

Quote:

Originally Posted by storm5510

A reservation limit can be gotten around with a recursive batch file. They run a bunch of loops, and there you are. Thousands of assignments in just a few minutes.

No disrespect intended, but I both write spiders and write spider defenses...

If done properly, there should be no "got-arounds" (at least, not without a LOT more work than most people are willing to do).

Frankly, my understanding is very often these large stagnating batch reservations are usually because of SPEs during spider creation.

chalsall · 2019-10-29, 00:13

So, I spent a bit of time bringing in the candidates to TF appropriately, now that we're back to a somewhat sane state wrt the reservations.

I realized that we are actually under a bit of crunch. But, somewhat ironically, in P-1'ing.

Primenet is currently assigning work in 94.5M for "Cat 4" assignments. We're good there for the moment -- a couple of thousand or so candidates already optionally TF'ed and P-1'ed.

However, this will move into 95M within a couple of weeks or so. We don't have as much of a buffer there -- currently only about 800 candidates.

So, my question... Would the GPU72 P-1'ers mind being given work in 95M to do, instead of 92M?

It will take longer per run, but each assignment will be appropriately TF'ed to 77 before assignment for P-1'ing. And once completed they can immediately be released back to Priment to give to "unproven" GIMPS participants (read: the Cat 4 "Churners").

Thoughts?

LaurV · 2019-10-29, 06:30

Quote:

Originally Posted by chalsall

Not just Mr. Delo... There's also Ryan ...

Then why are you handling out 34M to 73 for my collab work?

Start immediately to handle out 97M to 77 bits or so (I have seen you handle those out if I type a wrong key, so the anonymous users get 97M already - I did some of them by mistake and reported the results by hand, as the spider didn't get the wrong key right (pun intended)).

Also, should I move my ~10THzD/D worth of cards to LL front requests?

chalsall · 2019-10-29, 12:38

Quote:

Originally Posted by LaurV

Then why are you handling out 34M to 73 for my collab work?

Because that's what you asked for when you set up the AKey...

I had hoped to get it finished yesterday, but you will soon be able to change the WorkType for each AKEY.

I've got meetings with representitives from Estonia all day today, but will try to get this finished tonight.

Quote:

Originally Posted by LaurV

Also, should I move my ~10THzD/D worth of cards to LL front requests?

Yes please!!! LG72D would be ideal! Thanks!

James Heinrich · 2019-10-29, 13:32

Quote:

Originally Posted by chalsall

Would the GPU72 P-1'ers mind being given work in 95M to do, instead of 92M?

I can't speak for anyone else, but I'm happy to work on whatever GPU72 deems most useful.

I'd been working on "other" P-1 work (re-doing badly done P-1) but I've thrown that back and my two P-1 systems will chew on whatever GPU72 sends my way (at least for the next few months).

Uncwilly · 2019-10-29, 13:35

Quote:

Originally Posted by chalsall

I've got meetings with representitives from ~~Estonia~~ Elbonia all day today, but will try to get this finished tonight.

Fixed that for you.

lycorn · 2019-10-29, 13:49

@chalsall:

I´ve been consistently getting DC TF work (using The GPU72 Trial Factoring Colaboratory Notebook), although the Akey I´m using is supposed to correspond to LL TF work type.

Also, just so you know, there are two assignments I finished 3 weeks ago and reported to the Primenet server, that still appear (100% done and with no "days to go") in my assignment list. You may wish to have a look at it.

LaurV · 2019-10-29, 16:59

Quote:

Originally Posted by chalsall

I've got meetings with representitives (sic) from Estonia all day today

Wow, Estonia? Renju? Have you meet Ando?

(one of my wildest dreams is to play him face to face and beat him, the first will probably never happen, and for the second there is no hope, I played him on his renjuclass many years ago and he beat me so hard, like 12 to 0 or so, haha).

chalsall · 2019-10-29, 20:20

Quote:

Originally Posted by Uncwilly

Fixed that for you.

Ah... Thanks for that... I hope you don't mind that I didn't share the joke with them...

Seriously efficient people!!!

Much to learn from them...

chalsall · 2019-10-29, 20:28

Quote:

Originally Posted by lycorn

I´ve been consistently getting DC TF work (using The GPU72 Trial Factoring Colaboratory Notebook), although the Akey I´m using is supposed to correspond to LL TF work type.

OK... An important thing to clarify about the G72_TF assignment "policy" design...

1. Any assignments which are given to one of your "Instances" is yours until completion.

2. Thus, for example, if you run an AKey asking for DCTF, but then later switch to one asking for LLTF, the outstanding DCTF work is still assigned out first.

This /might/ explain what you're reporting. But... perhaps not...

Could you please PM me the "Name" of the AKey in question?

Quote:

Originally Posted by lycorn

Also, just so you know, there are two assignments I finished 3 weeks ago and reported to the Primenet server, that still appear (100% done and with no "days to go") in my assignment list. You may wish to have a look at it.

Thank's for the observations. I do appreciate them -- will help me a lot when I *finally* get back to some quality "Console Time" to deal with this.

This last week has been a bit, shall we say, scattered...

storm5510 · 2019-10-29, 22:50

Quote:

Originally Posted by chalsall

No disrespect intended, but I both write spiders and write spider defenses...

If done properly, there should be no "got-arounds" (at least, not without a LOT more work than most people are willing to do).

Frankly, my understanding is very often these large stagnating batch reservations are usually because of SPEs during spider creation.

None taken.

A spider. That's as good a name as any. As long as they can be stopped...

A note about P-1's. I did a test earlier today. I ran a single P-1, from 92M, with Prime95 to get a sense of the time. 6.5 hours, it took using two cores as workers. I am repeating now with CUDAPm1. I found an updated binary on James' site. The older one would not run, period. It estimated 2 hours for stage 1. Stage 2, I don't know yet. Still, it should finish in less time than Prime95. Anyone grabbing, or trying to grab, a lot of these, should keep the time required to run them in mind.

2019-10-29, 00:13	#4336
chalsall If I May "Chris Halsall" Sep 2002 Barbados 2·5·7·139 Posts	A bit of a "squeeze" in P-1... So, I spent a bit of time bringing in the candidates to TF appropriately, now that we're back to a somewhat sane state wrt the reservations. I realized that we are actually under a bit of crunch. But, somewhat ironically, in P-1'ing. Primenet is currently assigning work in 94.5M for "Cat 4" assignments. We're good there for the moment -- a couple of thousand or so candidates already optionally TF'ed and P-1'ed. However, this will move into 95M within a couple of weeks or so. We don't have as much of a buffer there -- currently only about 800 candidates. So, my question... Would the GPU72 P-1'ers mind being given work in 95M to do, instead of 92M? It will take longer per run, but each assignment will be appropriately TF'ed to 77 before assignment for P-1'ing. And once completed they can immediately be released back to Priment to give to "unproven" GIMPS participants (read: the Cat 4 "Churners"). Thoughts?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Status	Primeinator	Operation Billion Digits	5	2011-12-06 02:35
62 bit status	1997rj7	Lone Mersenne Hunters	27	2008-09-29 13:52
OBD Status	Uncwilly	Operation Billion Digits	22	2005-10-25 14:05
1-2M LLR status	paulunderwood	3*2^n-1 Search	2	2005-03-13 17:03
Status of 26.0M - 26.5M	1997rj7	Lone Mersenne Hunters	25	2004-06-18 16:46

2019-10-29, 13:49	#4341
lycorn Sep 2002 Oeiras, Portugal 2⁶·23 Posts	@chalsall: I´ve been consistently getting DC TF work (using The GPU72 Trial Factoring Colaboratory Notebook), although the Akey I´m using is supposed to correspond to LL TF work type. Also, just so you know, there are two assignments I finished 3 weeks ago and reported to the Primenet server, that still appear (100% done and with no "days to go") in my assignment list. You may wish to have a look at it.