mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet > MISFIT

Reply
 
Thread Tools
Old 2013-03-28, 22:41   #67
swl551
 
swl551's Avatar
 
Aug 2012
New Hampshire

14508 Posts
Default

Quote:
Originally Posted by flashjh View Post
Now you can see why I needed it so badly for 7 GPUs

Edit: Any way you can make MISFIT cook breakfast for me? ;)
Flashjh. I failed to get MISFIT to cook breakfast with any consistency, but I did discover that a GPU cooling fan can whisk up a nice vanilla custard sauce as good as any French chef.

Try it!
swl551 is offline   Reply With Quote
Old 2013-03-29, 00:21   #68
flashjh
 
flashjh's Avatar
 
"Jerry"
Nov 2011
Vancouver, WA

1,123 Posts
Default

[OT]
Quote:
Originally Posted by swl551 View Post
Flashjh. I failed to get MISFIT to cook breakfast with any consistency, but I did discover that a GPU cooling fan can whisk up a nice vanilla custard sauce as good as any French chef.

Try it!
With all the heat from my GPUs and the cooling fans, we could cook pancakes and scrambled eggs with a vanilla custard sauce topping for breakfast.[/OT]

Last fiddled with by flashjh on 2013-03-29 at 00:21
flashjh is offline   Reply With Quote
Old 2013-03-29, 01:42   #69
kracker
 
kracker's Avatar
 
"Mr. Meeseeks"
Jan 2012
California, USA

23·271 Posts
Default

Quote:
Originally Posted by flashjh View Post
[OT]

With all the heat from my GPUs and the cooling fans, we could cook pancakes and scrambled eggs with a vanilla custard sauce topping for breakfast.[/OT]
You could try preheating food like a microwave on top.
kracker is offline   Reply With Quote
Old 2013-04-06, 02:19   #70
Chuck
 
Chuck's Avatar
 
May 2011
Orange Park, FL

3×5×59 Posts
Default Work balance question

If there are 6 lines remaining in one worktodo file, 1 in another, and "Balance when difference >" is set to 10, will the smaller file run out of work and the instance stop?

Or is there some sort of safety valve to ensure that none of the files can run out?

I am seeing this situation since I manually added a bunch of short double checks; normally this would not have happened.
Chuck is offline   Reply With Quote
Old 2013-04-06, 16:27   #71
swl551
 
swl551's Avatar
 
Aug 2012
New Hampshire

23·101 Posts
Default

Quote:
Originally Posted by Chuck View Post
If there are 6 lines remaining in one worktodo file, 1 in another, and "Balance when difference >" is set to 10, will the smaller file run out of work and the instance stop?

Or is there some sort of safety valve to ensure that none of the files can run out?

I am seeing this situation since I manually added a bunch of short double checks; normally this would not have happened.
Chuck your conclusion is correct: No such fail-safe exists. It is a good idea though.
swl551 is offline   Reply With Quote
Old 2013-04-06, 18:42   #72
TObject
 
TObject's Avatar
 
Feb 2012

6258 Posts
Default

Also, with a very slow card (GTX 440) and long assignments (to 74 bits), with MISFIT set to minimum GHz-D/D settings, it normally has two exponents in the work to do file and two exponents in the staging file.

Usually this works fine, but sometimes mfaktc runs out of the exponents in the work to do and quits. MISFIT thinks that mfaktc stalled, and all work stops even though there are still exponents in the staging file.
TObject is offline   Reply With Quote
Old 2013-04-06, 19:59   #73
swl551
 
swl551's Avatar
 
Aug 2012
New Hampshire

23·101 Posts
Default

Quote:
Originally Posted by TObject View Post
Also, with a very slow card (GTX 440) and long assignments (to 74 bits), with MISFIT set to minimum GHz-D/D settings, it normally has two exponents in the work to do file and two exponents in the staging file.

Usually this works fine, but sometimes mfaktc runs out of the exponents in the work to do and quits. MISFIT thinks that mfaktc stalled, and all work stops even though there are still exponents in the staging file.
That is why I have stated a best practice of keeping at least two days of work on hand. One days worth in your live workToDos and one days worth in your reserve. Given a proper seeding the auto-fetching and draining system works fine.
swl551 is offline   Reply With Quote
Old 2013-04-06, 20:01   #74
TObject
 
TObject's Avatar
 
Feb 2012

34·5 Posts
Default

Quote:
Originally Posted by swl551 View Post
That is why I have stated a best practice of keeping at least two days of work on hand. One days worth in your live workToDos and one days worth in your reserve. Given a proper seeding the auto-fetching and draining system works fine.
Four exponents to 74, at the current TF front, is well over two days of work for non-overclocked GTX 440.

Last fiddled with by TObject on 2013-04-06 at 20:05
TObject is offline   Reply With Quote
Old 2013-04-06, 20:21   #75
Aramis Wyler
 
Aramis Wyler's Avatar
 
"Bill Staffen"
Jan 2013
Pittsburgh, PA, USA

23·53 Posts
Default

Quote:
Originally Posted by Chuck View Post
If there are 6 lines remaining in one worktodo file, 1 in another, and "Balance when difference >" is set to 10, will the smaller file run out of work and the instance stop?

Or is there some sort of safety valve to ensure that none of the files can run out?

I am seeing this situation since I manually added a bunch of short double checks; normally this would not have happened.

Maybe the minimum 'balance when difference >' setting should be a function of the number (in units or ghzd) fetched, so that one can't really fetch less than one can balance.
Aramis Wyler is offline   Reply With Quote
Old 2013-04-06, 20:54   #76
swl551
 
swl551's Avatar
 
Aug 2012
New Hampshire

23×101 Posts
Default

Quote:
Originally Posted by Aramis Wyler View Post
Maybe the minimum 'balance when difference >' setting should be a function of the number (in units or ghzd) fetched, so that one can't really fetch less than one can balance.
I don't understand your proposal. One row in a file is one assignment. I cannot move part of that row.

Also I'm still not seeing a problem that cannot be resolved by modifying configuration to address the situation. Don't want work to run out, put 10 days of work in your queue. It won't hurt anyone. There is enough work to go around.
swl551 is offline   Reply With Quote
Old 2013-04-06, 21:05   #77
chalsall
If I May
 
chalsall's Avatar
 
"Chris Halsall"
Sep 2002
Barbados

974710 Posts
Default

Quote:
Originally Posted by swl551 View Post
Also I'm still not seeing a problem that cannot be resolved by modifying configuration to address the situation. Don't want work to run out, put 10 days of work in your queue. It won't hurt anyone. There is enough work to go around.
Sure... But can be argued this is sub-optimal. And requires a human in the loop.

If MISFIT has knowledge about how many GHzDays/day an instance can do, then it should be able to balance the worktodo.txt files such that each instance has a reasonable amount of work based on that.
chalsall is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
released: MISFIT for MFAKTx version 2.11.0 (31-Jan-2017) swl551 MISFIT 19 2020-11-18 14:29
MISFIT 2.9 swl551 MISFIT 44 2014-12-15 12:46
MfaktX Controller (Windows) Mini-Geek Software 52 2014-02-18 18:27
MFAKTx speed menu (windows) swl551 Software 8 2013-03-07 00:16
(archive)mfaktXHelper - utility for managing mfaktX installations swl551 MISFIT 57 2012-08-12 03:37

All times are UTC. The time now is 08:29.


Tue Jul 27 08:29:20 UTC 2021 up 4 days, 2:58, 0 users, load averages: 2.16, 1.86, 1.79

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.