(archive)MISFIT - Page 6

kracker · 2012-11-29, 19:17

Looking good, thanks!

swl551 · 2012-12-01, 21:34

A very basic prototype is avail and is basically proof of concept. I was able to get MISFIT CuLu aware in just over an hour. A full conversion of MISFIT to fully support all the same features it does for mfaktX will take many hours. Feedback and interest in MISFIT for CuLu will determine its future.

READ CAREFULLY.

1. it assumes you are working DoubleCheck work. Test work won't get detected.

2. many of the MISFIT features are disabled as they need adaptation.

3. you can fetch DoubleCheck work from gimps using the Add Work button. Sorry gimps allows a MAX of two rows to be fetched at a time.

4. You can upload results as normal

5. can't balance

6. can't detect stalled process

7. can use remote control as normal

8. very little testing... BE CAREFUL.

This install should go into a different directory than your existing MISFIT installation.

http://www.mersenneforum.org/misfit/

flashjh · 2012-12-02, 00:00

Have it working remote just like MISFIT.

Had to rename the executable so it would open on the same system as MISFIT for TF. Also need to change the name internally so it can be distinguished between the two versions in the notification bar.

It's working quit well so far. I had it send the current results file just for a test and it submitted correctly.

Formatting is a little off for some reason.

Retrieved one assignment from PrimeNet with no trouble.

For info in the boxes, 'Factor' should be LL Rows, LL Done, Prime [#], Not Prime [#]

I have an LL-DC that will complete in about 5 hours and I have it set to auto submit after that. I'll post on how it goes.

swl551 · 2012-12-02, 19:07

Screenshot of MISFIT-CuLu managing four instances of CUDALucas. (two local, two remote)

Development of MISFIT-CuLu is against CUDALucas 2.04 beta.

Stef42 · 2012-12-02, 22:11

Looking good!

Off the record: how can you process exponents so fast?! N = 50k seems a bit low

swl551 · 2012-12-02, 22:21

Quote:

Originally Posted by Stef42

Looking good!

Off the record: how can you process exponents so fast?! N = 50k seems a bit low

I've asked Dubslow and Flashjh to confirm my throughput as it seems much higher in the beta version.

Allowing the beta to auto-select the FFTLength caused what normally took 2 hours to go to 27 hours so I started setting the lengths manually. Using 50k reduced my GPU load to <50% allowing me to run a second instance on the same GPU while each instance was running the expected ~2 hours to do an LL test.

I have confirmed this behavior on two GTX 570s and one 560.

Dubslow · 2012-12-02, 22:41

Quote:

Originally Posted by swl551

I've asked Dubslow and Flashjh to confirm my throughput as it seems much higher in the beta version.

Allowing the beta to auto-select the FFTLength caused what normally took 2 hours to go to 27 hours so I started setting the lengths manually. Using 50k reduced my GPU load to <50% allowing me to run a second instance on the same GPU while each instance was running the expected ~2 hours to do an LL test.

I have confirmed this behavior on two GTX 570s and one 560.

Yeah, something's definitely wrong. Those interim residues aren't changing (they should look like random hex, not ffffffffff00000000000). I've just sent a PM with more details.

Btw, a DC exponent should take about a day on a 570. It takes around 40 hours on a 460.

Stef42 · 2012-12-02, 22:56

For that given range, your N= should be 1600K at least, not 50. The error rate should have stopped the program already, I wonder why it did not. I'm running the same range on my GTX570, it's taking about 26 hours.

If you have been reporting assignments which took the 2 hours you mentioned, they are all incorrect I suppose.

swl551 · 2012-12-02, 23:00

Quote:

Originally Posted by Stef42

For that given range, your N= should be 1600K at least, not 50. The error rate should have stopped the program already, I wonder why it did not. I'm running the same range on my GTX570, it's taking about 26 hours.

If you have been reporting assignments which took the 2 hours you mentioned, they are all incorrect I suppose.

I suspect they are. Who do I give the list to so they can be reversed out?

swl551 · 2012-12-03, 00:44

1. Fetch DC work from gimps
2. Assign/Manage/monitor work to instances on local machine or over network share (samba for Linux users)
3. Balance work assignments
4. Detect stalled instances of CuLu and email out
5. Email when work gets low
6. Perform remote control START/STOP of CuLu instances on remote machines
7. Have results automatically uploaded to GIMPS

Actually the only thing that MISFIT CuLu lacks in comparison to MISFIT is the calculation of GHZDays right in the app for remaining work and completed work.

Hopefully this helps motivate more people to move back to LL work since the TF work is so far ahead of the curve.

flashjh · 2012-12-03, 01:05

I am currently working an exponent that I broke into two because it's so large:

Code:

Factor=2012-11-21,332240563,77,78
Factor=2012-11-21,332240563,78,79

It's telling me they're a duplicate even though they're not. Can you make it also check the from and to for duplicate checking?

Quote:

Originally Posted by swl551

SURE...give me about an hour

I need to find a way to get more time so I can program ;)

2012-11-29, 19:17	#56
kracker "Mr. Meeseeks" Jan 2012 California, USA 2³×271 Posts	Looking good, thanks! Attached Thumbnails

2012-12-01, 21:34	#57
swl551 Aug 2012 New Hampshire 1450₈ Posts	MISFIT for CUDALucas (proof of concept) A very basic prototype is avail and is basically proof of concept. I was able to get MISFIT CuLu aware in just over an hour. A full conversion of MISFIT to fully support all the same features it does for mfaktX will take many hours. Feedback and interest in MISFIT for CuLu will determine its future. READ CAREFULLY. 1. it assumes you are working DoubleCheck work. Test work won't get detected. 2. many of the MISFIT features are disabled as they need adaptation. 3. you can fetch DoubleCheck work from gimps using the Add Work button. Sorry gimps allows a MAX of two rows to be fetched at a time. 4. You can upload results as normal 5. can't balance 6. can't detect stalled process 7. can use remote control as normal 8. very little testing... BE CAREFUL. This install should go into a different directory than your existing MISFIT installation. http://www.mersenneforum.org/misfit/

2012-12-02, 00:00	#58
flashjh "Jerry" Nov 2011 Vancouver, WA 1,123 Posts	Have it working remote just like MISFIT. Had to rename the executable so it would open on the same system as MISFIT for TF. Also need to change the name internally so it can be distinguished between the two versions in the notification bar. It's working quit well so far. I had it send the current results file just for a test and it submitted correctly. Formatting is a little off for some reason. Retrieved one assignment from PrimeNet with no trouble. For info in the boxes, 'Factor' should be LL Rows, LL Done, Prime [#], Not Prime [#] I have an LL-DC that will complete in about 5 hours and I have it set to auto submit after that. I'll post on how it goes. Attached Thumbnails Last fiddled with by flashjh on 2012-12-02 at 00:02

2012-12-02, 19:07	#59
swl551 Aug 2012 New Hampshire 2³·101 Posts	MISFIT-CuLu Screenshot of MISFIT-CuLu managing four instances of CUDALucas. (two local, two remote) Development of MISFIT-CuLu is against CUDALucas 2.04 beta. Attached Thumbnails

2012-12-03, 00:44	#65
swl551 Aug 2012 New Hampshire 2³×101 Posts	So what can you do with MISFIT CuLu 1. Fetch DC work from gimps 2. Assign/Manage/monitor work to instances on local machine or over network share (samba for Linux users) 3. Balance work assignments 4. Detect stalled instances of CuLu and email out 5. Email when work gets low 6. Perform remote control START/STOP of CuLu instances on remote machines 7. Have results automatically uploaded to GIMPS Actually the only thing that MISFIT CuLu lacks in comparison to MISFIT is the calculation of GHZDays right in the app for remaining work and completed work. Hopefully this helps motivate more people to move back to LL work since the TF work is so far ahead of the curve.

2012-12-02, 22:11	#60
Stef42 Feb 2012 the Netherlands 2·29 Posts	Looking good! Off the record: how can you process exponents so fast?! N = 50k seems a bit low

2012-12-02, 22:56	#63
Stef42 Feb 2012 the Netherlands 2·29 Posts	For that given range, your N= should be 1600K at least, not 50. The error rate should have stopped the program already, I wonder why it did not. I'm running the same range on my GTX570, it's taking about 26 hours. If you have been reporting assignments which took the 2 hours you mentioned, they are all incorrect I suppose.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
questions about misfit	jschwar313	MISFIT	1	2016-02-02 00:32
Help with Misfit	TheMawn	MISFIT	18	2015-06-21 22:00
Misfit feature request.	Aramis Wyler	MISFIT	15	2015-03-04 21:13
MISFIT 2.9	swl551	MISFIT	44	2014-12-15 12:46
MISFIT for MFAKTx	swl551	MISFIT	186	2014-02-05 23:27