GPU72 / MISFIT use for 100M digit range?

[Uncwilly]

So, could I put Misfit on a machine that is only running Prime95 (not a Macfakx) and set it for the desired range and assignment type and let it keep the machine fed for all cores? So I would not have to manually feed them for 332M range factoring assignments.

[swl551]

Originally Posted by Uncwilly

So, could I put Misfit on a machine that is only running Prime95 (not a Macfakx) and set it for the desired range and assignment type and let it keep the machine fed for all cores? So I would not have to manually feed them for 332M range factoring assignments.

Sorry, MISFIT cannot do what you specified. If there is a large need for such a utility I would consider looking into the requirements. Maybe a new thread is needed to capture an audience and get the specs ironed out.

Scott

[chalsall]

Originally Posted by swl551

Sorry, MISFIT cannot do what you specified. If there is a large need for such a utility I would consider looking into the requirements. Maybe a new thread is needed to capture an audience and get the specs ironed out.

If there's a desire for this, a proposal...

1. GPU72 could bring in a range of 332M candidates which Uncwilly advises on.
2. These are then exposed in a separate table on all the various reports, but would really simply be LLTF (although in a higher range).
3. MISFIT and/or the fetcher could be used to grab the assignments. By simply modifying the ranges the user(s) could ask for the 332M candidates.
4. Because it's ill-advised to modify the worktodo.txt file when Prime95 is running, the worktodo.add feature is leveraged instead.

Uncwilly -- I've offered before to have GPU72 manage the 332M range, but only with your permission. This is another offer.

Scott -- would any of this require any additional work on your part? For me it would only be a couple of hours.

Edit: Just realized... To support Prime95 you'd need to add a "Number of Workers" functionality.

Edit #2: Also just realized... For Prime95/mprime clients the proxy could be used to assign such work if the "LMH" option is set.

[kracker]

Originally Posted by chalsall

1. GPU72 could bring in a range of 332M candidates which Uncwilly advises on.
2. These are then exposed in a separate table on all the various reports, but would really simply be LLTF (although in a higher range).
3. MISFIT and/or the fetcher could be used to grab the assignments. By simply modifying the ranges the user(s) could ask for the 332M candidates.
4. Because it's ill-advised to modify the worktodo.txt file when Prime95 is running, the worktodo.add feature is leveraged instead.

+1!

[swl551]

Originally Posted by chalsall

If there's a desire for this, a proposal...

1. GPU72 could bring in a range of 332M candidates which Uncwilly advises on.
2. These are then exposed in a separate table on all the various reports, but would really simply be LLTF (although in a higher range).
3. MISFIT and/or the fetcher could be used to grab the assignments. By simply modifying the ranges the user(s) could ask for the 332M candidates.
4. Because it's ill-advised to modify the worktodo.txt file when Prime95 is running, the worktodo.add feature is leveraged instead.

Uncwilly -- I've offered before to have GPU72 manage the 332M range, but only with your permission. This is another offer.

Scott -- would any of this require any additional work on your part? For me it would only be a couple of hours.

Edit: Just realized... To support Prime95 you'd need to add a "Number of Workers" functionality.

Edit #2: Also just realized... For Prime95/mprime clients the proxy could be used to assign such work if the "LMH" option is set.

Well, I already have a complete GIMPS fetching code base that can HTTP/GET 100m assignments. I'm not sure there is a need to get the assignments from GPU72. I can build to either. Adding data to WorkToDo.add should be simple. Since it appears 100m is done with P95 and not mfaktX I think a commandline utility is better than trying to get MISFIT to comingle with P95.

Thoughts?

[chalsall]

Originally Posted by swl551

I'm not sure there is a need to get the assignments from GPU72.

I have developed some tools and techniques which allow "Spidy" to get assignments from Primenet which it wouldn't ordinarily assign (because they're already TFed to or past the level thought "profitable" for CPUs).

But Uncwilly has managed this range for years -- I consider him to be one to "Go/No Go" on GPU72 getting involved.

[Uncwilly]

Let me think and get back to you with my thoughts.

[LaurV]

Like one of the "biggest" producers for TF-LMH (100M digits), I would like very much and feel very comfortable (and maybe contribute more ) if GPU72 would offer assignments in 332M+ range. Like "exponent first" or "bit first" (I usually do bit-first, i.e. take everything to 69, then take everything to 70, etc). I sometime get out 200-500 exponents per day, and that would be a lot of manual work, therefore I just generate a long list with exponents from my personal database, or sometime from the server (manually, using something like this and copy/paste), and just start working them without any reservation. Therefore, about 5% to 10% of the work I do is in vain, because sometimes, somebody else does some exponents and reports them before I am reaching them in my list, and when I am reaching them and report them, I get "error 40 - result no need" or something like that. I don't have time to check continuously which exponents are done and which are not, so I prefer to let it running and duplicate some work.

Therefore, it would be wonderful if Misfit+GPU72 would offer a way to get those assignments "legally" and "automatically". It would even convince me to start using Misfit

So, think fast, Uncwilly...

[Aramis Wyler]

I imagine the workfetcher would work too if 332M numbers were available from gpu72. It might be tricky to get the initial commitments in correctly from Uncwilly's current system.

[Uncwilly]

Ok, after thinking about it and how it might work, here is what I think.

Sure, let GPUto72 work in the range. The help from some GPU's will be great.
As previously noted be some others, there is 2 types of work that folks tend to do, bit-wise (pick a range and move all of the exponents up one or 2 bits) and depth-wise (take the smallest exponent available to a set depth, generally to depth specified here: http://mersenneforum.org/attachment....achmentid=9126). GPUto72 should allow for both of those. It should also leave some exponents available for the CPU folks to pick-up.

At least until further notice, the bit-wise testing should stay in the 332 range, until the minimum bit level above 332192831 reaches at least 74.

The depth-wise commitment should be at least 78 (at first while demand is high for exponents to LL), ideally 80 or 81 (later). If there is anyone that wants to do P-1's, that would be great too.

If there is a way to leave some exponents for CPU folks in PrimeNet to work on, that would be great.

It might be tricky to get the initial commitments in correctly from Uncwilly's current system.

I use PrimeNet, as noted here: http://mersenneforum.org/showthread.php?t=11308

I will probably continue to use that method on at least 2 of my machines. If I can get to of my borged machines to use an automated system to fetch work, I would love to.

Feedback? Ideas?

[swl551]

If GPU72 manages the ranges and has a method to request work, I can build the fetching utility.