GPU to 72 status... - Page 380

chalsall · 2018-07-04, 18:40

Quote:

Originally Posted by penlu

Now I have another question---why is gpu72 handing out TFing for ~93M only up to 72 bits, esp. when TF to a given bitlevel is easier at higher exponents?

I know you've already answered your own questions, but to expand for context...

Some like to do higher candidates to lower bit levels because more factors are found per GHzDays, as well as "wall-clock time". This might be because they have slower GPUs, or because they just like finding lots of factors.

And this is perfectly fine. The rule around the GIMPS project is everyone's allowed to do whatever they want (this is a volunteer effort, after all), so long as it doesn't negatively impact on anyone else. As in, all work is appreciated, except for "poaching" (read: doing work assigned to someone else).

GPU72 was created to facilitate the coordination of GPU Trial Factoring; originally such efforts were actually coordinated via emails, and occasionally "toes were stepped on".

TL;DR: TF'ing to 72 bits in the 90s is still valuable to the GIMPS project in the long term, but going to 76 in the high 80s is more immediately useful.

chalsall · 2018-07-22, 21:07

Hey all. Just a quick update.

Thanks to everyone who stepped up to provide some fairly serious GPU compute to help with the "to 76 bit" Trial Factoring crunch.

We now have about a three month buffer of candidates "optimally trial factored" ahead of Primenet's P-1'ers.

petrw1 · 2018-07-22, 21:30

Quote:

Originally Posted by chalsall

Hey all. Just a quick update.

Thanks to everyone who stepped up to provide some fairly serious GPU compute to help with the "to 76 bit" Trial Factoring crunch.

We now have about a three month buffer of candidates "optimally trial factored" ahead of Primenet's P-1'ers.

That was efficient and effective.

If anyone is so inclined you are welcome to redirect your efforts to GPU72 DC TF and help the DC progress

kladner · 2018-07-24, 20:37

Quote:

Originally Posted by petrw1

That was efficient and effective.

If anyone is so inclined you are welcome to redirect your efforts to GPU72 DC TF and help the DC progress

Chris let me know the status of things, so I have DCTF enqueued for when the LLTFs finish.

chalsall · 2018-07-25, 20:21

Quote:

Originally Posted by kladner

Chris let me know the status of things, so I have DCTF enqueued for when the LLTFs finish.

Yeah, I saw that. Thanks.

GPU72 is also managing DCTF'ing ahead of the DC wavefront. And it's currently targeted to help Wayne's (aka petrw1) sub-project of getting 0.1M ranges below 2,000 un-factored candidates (while at the same time helping the GIMPS DC effort).

Interestingly, Oliver Kruse completed his 1,000+ manually reserved P-1 assignments in the 87M range a couple of days ago, and then reserved another 1,000+ in 88M. Excellent!

While this tightens the temporal buffer a bit, we're still looking good.

Thanks for the compute everyone!

Chuck · 2018-07-25, 20:37

Quote:

Originally Posted by chalsall

GPU72 is also managing DCTF'ing ahead of the DC wavefront. And it's currently targeted to help Wayne's (aka petrw1) sub-project of getting 0.1M ranges below 2,000 un-factored candidates (while at the same time helping the GIMPS DC effort).

I switched MISFIT over to Double Check, Let GPU72 Decide. It has already begun downloading DC TF.

chalsall · 2018-07-25, 21:02

Quote:

Originally Posted by Chuck

I switched MISFIT over to Double Check, Let GPU72 Decide. It has already begun downloading DC TF.

Thanks Chuck.

The nice thing about DCTF is there is a finite amount to complete; unlike LLTF which will go on forever (or, at least, as long as most of us are still using deterministic compute)....

kladner · 2018-07-25, 22:01

The nice thing about LLTF is the big spike in Work Saved when a factor is found.

James Heinrich · 2018-07-25, 22:26

Quote:

Originally Posted by kladner

The nice thing about LLTF is the big spike in Work Saved when a factor is found.

M9999999491 has a factor: 19999998983
"you just saved 551,667 GHz-days"

kladner · 2018-07-26, 02:32

Quote:

Originally Posted by James Heinrich

M9999999491 has a factor: 19999998983
"you just saved 551,667 GHz-days"

I am not so ambitious. I was just doing wave-front LLTF. While it is insignificant in real terms, I got a real charge from 2 or almost 3 factor days in LLTF.

penlu · 2018-08-06, 03:14

I have just received a couple of exponent assignments I found kind of strange since they're so large: 166802869, 223488289

It looks like these were recently (as in today) TF'd up to 70 bits; I got the assignment to go up to 80. They're associated with an assignment ID and show up in my mersenne.org but not gpu72.com assignments. I am using primetools to get work. I assume gpu72 went down briefly and so primetools fell through to primenet...?

I have a couple of questions. Is this associated with mersenne.org going down recently? Why are these exponents so large?

2018-07-22, 21:07	#4171
chalsall If I May "Chris Halsall" Sep 2002 Barbados 3³·19² Posts	Quick status update... Hey all. Just a quick update. Thanks to everyone who stepped up to provide some fairly serious GPU compute to help with the "to 76 bit" Trial Factoring crunch. We now have about a three month buffer of candidates "optimally trial factored" ahead of Primenet's P-1'ers.

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

2018-07-25, 22:01	#4177
kladner "Kieren" Jul 2011 In My Own Galaxy! 2×3×1,693 Posts	The nice thing about LLTF is the big spike in Work Saved when a factor is found.

2018-08-06, 03:14	#4180
penlu Jul 2018 31 Posts	I have just received a couple of exponent assignments I found kind of strange since they're so large: 166802869, 223488289 It looks like these were recently (as in today) TF'd up to 70 bits; I got the assignment to go up to 80. They're associated with an assignment ID and show up in my mersenne.org but not gpu72.com assignments. I am using primetools to get work. I assume gpu72 went down briefly and so primetools fell through to primenet...? I have a couple of questions. Is this associated with mersenne.org going down recently? Why are these exponents so large?