mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet > GPU to 72

Reply
 
Thread Tools
Old 2013-12-02, 05:42   #1
davieddy
 
davieddy's Avatar
 
"Lucan"
Dec 2006
England

6,451 Posts
Default The "(W)TF Depth Debate" state of the bunion

Quote:
Originally Posted by petrw1 View Post
There were over 2,000 in July....as of today under 1,700.
Is George no longer giving LL testers 25% DC?
Or is DC so like last year :)
Interesting, However the DC completion rate has remained around 115/day during this period. Countdown watchers may disagree, but this is a reasonableTM rate, providing a useful source of work for CPUs like Brian-E's, (especially since TF by CPU is obsolete), and the chance of finding a prime per year only about 5x smaller than for first time LLs.

A drop in the number of assignments currently out is not all bad news:
since the completion rate hasn't dropped, assigments must be getting done faster which is a Good ThingTM .

Quote:
Originally Posted by Prime95 View Post
Correct. Perhaps, I should again give "Do what makes most sense" users a few DCs?
No George.
See above and below.
And any such move would be an admission that we can't sustain TF to 74 for all assigments above 63M, i.e. keep those pigs airborne without cheating.
If that were to happen, and it is already, you, Garo and Chris will owe me a massive grovelling apologyTM

Quote:
Originally Posted by chalsall View Post
Wouldn't hurt...

We "survived" the incident earlier this week where Primenet didn't recycle expiring exponents for a day; Primenet still didn't issuing anything for LLing not TFed to at least 74 and P-1ed. And the TFing is continuing to slowly pull ahead.

But perhaps a bit more of a lead would be advisable... Very easily accomplished in the short term by sending even only something like 10% of Primenet.WMS to DCing for a short while.
That swinging ban has given me time to carefully interpret your most valuable TF-level data. Some of this I will post elsewhere, or perhaps privately might lead to less rowdiness from the likes of Foetus Boy
It would be helpful if we could have the number of current LL assignments, like what Petrw quoted.

As you have conceded, every day at midnight it is touch and go whether or not we will run out of available exponents, and "a bit more of a lead would be advisable". Funny, I think I might have said that before.
Crucially, it would enable us to find out whether thís brinkmanship is holding back the assignment rate as I strongly suspect.

Anyway guys, it's good to be back, although the delay to my posts is tiresome in the extreme.

Bedtime!

D
davieddy is offline   Reply With Quote
Old 2013-12-02, 17:01   #2
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

221068 Posts
Default The "TF Depth Debate" reunion.

Hey All...

This thread is the re-opening of the open debate between those who think we're TFing too far, and those who think we're doing OK.

For the record, I would very much appreciate it if discussion was civil, sincere and meaningful.

No YouTube links will be tolerated.
chalsall is online now   Reply With Quote
Old 2013-12-02, 17:05   #3
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

221068 Posts
Default

Quote:
Originally Posted by davieddy View Post
As you have conceded, every day at midnight it is touch and go whether or not we will run out of available exponents, and "a bit more of a lead would be advisable". Funny, I think I might have said that before.
Yes. This is the nature of Primenet. It "inhales" deeply shortly after midnight UTC, and then slowly exhales throughout the rest of the day.

Consider it something like automated yoga breaths....
chalsall is online now   Reply With Quote
Old 2013-12-02, 22:09   #4
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23·3·17 Posts
Default

I want to note that for the last several months my TF work has been exclusively in the LMH range, and will soon (~Dec 16) swing back to regular LL work. I know the work of one individual doesn't matter a whole lot, but in the past this debate has come down to some very narrow margins, and an additional 800 GHZD/Day is a sginificant increase in LLTF - enough to swing very narrow margins.

I occasionally feel compelled to apologize for moving all of my work into the LMH range just as we started going to 74 bits, but I will not because I still believe that the staggering increase in p-1 work was worth it.
Aramis Wyler is offline   Reply With Quote
Old 2013-12-02, 22:47   #5
kracker
ἀβουλία
 
kracker's Avatar
 
"Mr. Meeseeks"
Jan 2012
California, USA

23×271 Posts
Default

Quote:
Originally Posted by Aramis Wyler View Post
I occasionally feel compelled to apologize for moving all of my work into the LMH range just as we started going to 74 bits, but I will not because I still believe that the staggering increase in p-1 work was worth it.
"staggering increase in p-1 work" Where? Haven't noticed on the charts lately.
kracker is offline   Reply With Quote
Old 2013-12-02, 22:53   #6
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

2×4,643 Posts
Default

Quote:
Originally Posted by kracker View Post
"staggering increase in p-1 work" Where? Haven't noticed on the charts lately.
It's there.

A certain person brought forward some software which ran on GPUs involving P-1 work.

There is a private joke associated with this.
chalsall is online now   Reply With Quote
Old 2013-12-02, 23:07   #7
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23·3·17 Posts
Default

Maybe 'staggering' is too dramatic for you, but it seems to me UncWilly by himself did about 5% of all the p-1 work gpu72's been turning out this last 6 months. Far more than I expected any one person to be able to do. I think that, considering we were getting behind in p-1 so badly, that my sacrificing ~5% of our LLTF throughput for a similar ~5% increase in p-1 throughput was worthwhile.

The cudap-1 program alters the balance a bit though. I'm going to focus on LLTF again, but if we got behind in P-1 I could just switch to that for a bit, or put one card on each duty if Chris thought that was better.

But yeah, I think Uncwilly's borg army has produced a really impressive amount of P-1 work this year, and I think it was worth it even if we lost a step on lltf for a time.
Aramis Wyler is offline   Reply With Quote
Old 2013-12-02, 23:19   #8
kracker
ἀβουλία
 
kracker's Avatar
 
"Mr. Meeseeks"
Jan 2012
California, USA

23·271 Posts
Default

You can check monthly etc status on P-1 on gpu72.
https://www.gpu72.com/reports/workers/p-1/month/
kracker is offline   Reply With Quote
Old 2013-12-02, 23:21   #9
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

221068 Posts
Default

Quote:
Originally Posted by Aramis Wyler View Post
But yeah, I think Uncwilly's borg army has produced a really impressive amount of P-1 work this year, and I think it was worth it even if we lost a step on lltf for a time.
I was talking about Carl....
chalsall is online now   Reply With Quote
Old 2013-12-02, 23:53   #10
davieddy
 
davieddy's Avatar
 
"Lucan"
Dec 2006
England

6,451 Posts
Default

Quote:
Originally Posted by Aramis Wyler View Post
I want to note that for the last several months my TF work has been exclusively in the LMH range, and will soon (~Dec 16) swing back to regular LL work. I know the work of one individual doesn't matter a whole lot, but in the past this debate has come down to some very narrow margins, and an additional 800 GHZD/Day is a sginificant increase in LLTF - enough to swing very narrow margins.

I occasionally feel compelled to apologize for moving all of my work into the LMH range just as we started going to 74 bits, but I will not because I still believe that the staggering increase in p-1 work was worth it.
Welcome back to the Prime Search from your flight of fancy!

800 GHzDays amounts to about 20 65M exponents taken from 73 to 74 bits does it not?
The daily TF work for the past months has been equivalent to 500 of same, so you will be contributing a valuable 4^ boost in firepower.

Now considering that the mvast majority of exponents below 74 bits have been allowed to languish on 71 bits, the current fire power is sufficient to take 300 exponents a day to 74.
Now since the rate of LL completion is a very steady 300/day, it would appear that we can neatly maintain the number of LLs in progress.

However the picture is not that rosy.
We have been taking about 340 to 74 a day, and TFassignments eat them all up. This seems all to the good, since we would like assignment to exceed completion by a bit.

But this is my argument: how come we are TFing 340 a day to 74, when the firepower can only manage 300, even if concentrated exclusively on the "virgin" exponents?
The answer is that we have beenhoming in on the exponents already TFed to 73, and these are running out fast.
TFing a recycled expo below 63M to 74 does not create a new assignment - it is squandering much needed firepower.

If you have half followed this, you will see that another to or three contributions like your 800 GHzD/D could be enough to get Chris out of his scrape,, at the cost of conceding the argument to me.

D
davieddy is offline   Reply With Quote
Old 2013-12-03, 02:20   #11
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23·3·17 Posts
Default

Quote:
Originally Posted by davieddy View Post
Welcome back to the Prime Search from your flight of fancy!
Thank you! Welcome back from your plight of exile.
Quote:
Originally Posted by davieddy View Post
800 GHzDays amounts to about 20 65M exponents taken from 73 to 74 bits does it not?
Yup, sounds about right.

Quote:
Originally Posted by davieddy View Post
Now considering that the mvast majority of exponents below 74 bits have been allowed to languish on 71 bits, the current fire power is sufficient to take 300 exponents a day to 74.
Now since the rate of LL completion is a very steady 300/day, it would appear that we can neatly maintain the number of LLs in progress.
Sounds good!

Quote:
Originally Posted by davieddy View Post
However the picture is not that rosy.
We have been taking about 340 to 74 a day, and TFassignments eat them all up. This seems all to the good, since we would like assignment to exceed completion by a bit.
Miracles!

Quote:
Originally Posted by davieddy View Post
But this is my argument: how come we are TFing 340 a day to 74, when the firepower can only manage 300, even if concentrated exclusively on the "virgin" exponents?
The answer is that we have beenhoming in on the exponents already TFed to 73, and these are running out fast.
OMG you're right! We've only got about 2000 available currently! We could be out in a week!

Quote:
Originally Posted by davieddy View Post
TFing a recycled expo below 63M to 74 does not create a new assignment - it is squandering much needed firepower.
Our ratio of Truth to Crap just took a major nosedive there, but that statement does not invalidate the truthiness of previous statements so lets just carry on like it didn't happen.

Quote:
Originally Posted by davieddy View Post
If you have half followed this, you will see that another to or three contributions like your 800 GHzD/D could be enough to get Chris out of his scrape,, at the cost of conceding the argument to me.
Ok so hrm, we have enough firepower to do 300/day at 71-74, but we're currently getting 340/day done because we're chewing up the low hanging fruits that we're almost out of. Hmm, yes. So we're currently running up a lead (340/day surpassing the 300/day needed) but soon we'll only be able to tread water at 300/day because there's no way we can keep up this 340/day stuff. Hmm. So... it seems like then we're in good shape, because we're going to run up a good lead until the point were we can only keep up with assignments and hold on to (but not improve) our significant lead.

Sounds like we've exactly nailed the amount of TF we should be doing. An impressive feat!
Aramis Wyler is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Stockfish game: "Move 8 poll", not "move 3.14159 discussion" MooMoo2 Other Chess Games 5 2016-10-22 01:55
Aouessare-El Haddouchi-Essaaidi "test": "if Mp has no factor, it is prime!" wildrabbitt Miscellaneous Math 11 2015-03-06 08:17
Specifing TF factor depth in "Manual Assignments"? kracker PrimeNet 2 2012-07-22 17:49
Would Minimizing "iterations between results file" may reveal "is not prime" earlier? nitai1999 Software 7 2004-08-26 18:12

All times are UTC. The time now is 15:14.

Tue Oct 27 15:14:51 UTC 2020 up 47 days, 12:25, 1 user, load averages: 2.87, 3.17, 3.13

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.