Greetings from a newbie

[moebius]

Originally Posted by piforbreakfast

Speaking of which, how does GIMPS handle a job if our computer shuts itself off for several hours (I just woke up about an hour ago and I'm about to head to work now) and we have to reboot? Do they just put a project's progress on hold if for whatever reason we're offline for a short time? I'm brand new to this and just trying to figure things out. I'm 49 percent complete with the job I'm on at the moment and Prime95 is running again.

You can continue the calculation at the respective exponent at any time. You just have to note that an exponent expires after a certain time, which can be seen at https://www.mersenne.org/workload/. With Advanced> manual communication
you can update the progress of the exponent on the server.

[piforbreakfast]

So I've just been reading over some of the posts geared toward new members, and I see where it says that it's recommended that at least 1/5 of your work time be dedicated toward double-checking earlier results, and in fact the job I have in progress right now is a double-checking one. Is there a "best" job, i.e., most productive one, that we should always gravitate toward first? Should someone reserve most of their work toward double-checking unless otherwise suggested? One thread I saw about double-check posts said there was a backlog of jobs of more than a year. I'm just trying to get off to a productive start to help the effort make progress.

[kriesel]

Originally Posted by piforbreakfast

how does GIMPS handle a job if our computer shuts itself off for several hours (I just woke up about an hour ago and I'm about to head to work now) and we have to reboot? Do they just put a project's progress on hold if for whatever reason we're offline for a short time? I'm brand new to this and just trying to figure things out. I'm 49 percent complete with the job I'm on at the moment and Prime95 is running again.

Prime95 provides two different estimates of completion; one in Test, status, based on average rate of progress, and one in the worker window, based on recent rate of progress. Your PrimeNet account pages such as "cpus" will correspond to average rate of progress.

If you want it to run unattended without shutting itself down while you sleep, see Power settings https://www.mersenneforum.org/showpo...87&postcount=8

Time to expire is a week, month, or more, depending on whether the assignment has been reported as started and what category or type of work it is. See https://www.mersenne.org/thresholds/ https://www.mersenne.org/workload/ and other parts of the website.

Before PRP was adopted, the way to confirm a newly set up GIMPS system was reliable was to run several matching LLDC. More LLDC is always welcome, since the backlog is several years long. https://www.mersenneforum.org/showpo...4&postcount=15

[VBCurtis]

Originally Posted by piforbreakfast

So I've just been reading over some of the posts geared toward new members, and I see where it says that it's recommended that at least 1/5 of your work time be dedicated toward double-checking earlier results, and in fact the job I have in progress right now is a double-checking one. Is there a "best" job, i.e., most productive one, that we should always gravitate toward first? Should someone reserve most of their work toward double-checking unless otherwise suggested? One thread I saw about double-check posts said there was a backlog of jobs of more than a year. I'm just trying to get off to a productive start to help the effort make progress.

We appreciate your view and concern, but do what interests you most about the project. If you employ your GPU to make factoring progress, that helps the project quite a bit.

We test each candidate to make sure it's not divisible by small factors before sending it out for the kind of test you're currently running. That pre-testing is done by GPU these days. It cannot find a prime, but it helps the overall project by reducing the number of candidates that need the lengthy test.

mfaktc (CUDA/nvidia cards) or mfakto (OpenCL/AMD cards) are the software packages to run factoring on GPU. Those threads take some time to digest, but learning awaits!

[Runtime Error]

Originally Posted by piforbreakfast

So I've just been reading over some of the posts geared toward new members, and I see where it says that it's recommended that at least 1/5 of your work time be dedicated toward double-checking earlier results, and in fact the job I have in progress right now is a double-checking one. Is there a "best" job, i.e., most productive one, that we should always gravitate toward first? Should someone reserve most of their work toward double-checking unless otherwise suggested? One thread I saw about double-check posts said there was a backlog of jobs of more than a year. I'm just trying to get off to a productive start to help the effort make progress.

Like others have said, only do what makes you happy. At the end of the day, it is your time and resources going towards the project. With the new development of PRP-proofs, new tests of that nature are no longer adding to the double-check backlog. Eventually, the backlog will cease to exist. And the main folks who have not switched to PRP-proof from LL don't tend to contribute much to double-check work. Don't hold back from doing first time tests!

To add to what VBCurtis said above, most GPUs are *much* better at trial factoring to eliminate candidates than they are at doing primality tests. Like 40 times as good. (Basically videogames use lots of single-precision units which are good at TF but currently not ideal for primality tests). The fact that you get more "bang for your buck" with GPU trial factoring tends to tilt the scale towards factoring work on GPUs because finding factors is very satisfying!

[piforbreakfast]

Originally Posted by kriesel

Prime95 provides two different estimates of completion; one in Test, status, based on average rate of progress, and one in the worker window, based on recent rate of progress. Your PrimeNet account pages such as "cpus" will correspond to average rate of progress.

If you want it to run unattended without shutting itself down while you sleep, see Power settings https://www.mersenneforum.org/showpo...87&postcount=8

I just "woke" my computer screen and saw that I'm at 96.8 percent completion on my very first project, two days ahead of the projected completion time Prime 95 estimated when I began this job. WOO-HOO!! My computer is running at blazing speed.

So what happens after I finish the job I'm on?

[paulunderwood]

Originally Posted by piforbreakfast

I just "woke" my computer screen and saw that I'm at 96.8 percent completion on my very first project, two days ahead of the projected completion time Prime 95 estimated when I began this job. WOO-HOO!! My computer is running at blazing speed.

So what happens after I finish the job I'm on?

Assuming you are using the latest Prime95, a certificate will be generated and that will be uploaded to the server. Then your computer will start the next task, be that cached or downloaded from the server. Have a peek into worktodo.txt.

[kriesel]

First assignment on a newly installed prime95 instance is LL DC. It matches the first LL on the same exponent by someone else, or doesn't.

Later, if running PRP in V30.2 or later prime95, it will generate a large proof file and upload it to the server after sending the results.json.txt contents. Then the server produces a Cert assignment and input file (not nearly as large as the proof file) for some system and user, possibly even the same as produced the proof file, without loss of verification reliability. Depending on proof power, performing this verification assignment may take only 0.4% to 2% as much effort to confirm the correct completion of the PRP test as a full test takes, and complete within hours of submission of the proof file. (For more on the proof/cert development, see https://mersenneforum.org/showthread.php?t=25638)

Odds are given by prime95 status output for finding a prime in a single test, and are usually of order 1ppm.
At some point during the progress, if prime95's configuration for "days of work to queue up" >0, another assignment will have been retrieved from the server and stored in local file worktodo.txt for the worker in prime95.

Assuming you've created a PrimeNet account and entered the account name in prime95, computing credit will be added for the user and the system.

Work will continue with the next worktodo item for the worker, if any.

[piforbreakfast]

Originally Posted by Runtime Error

Like others have said, only do what makes you happy. At the end of the day, it is your time and resources going towards the project. With the new development of PRP-proofs, new tests of that nature are no longer adding to the double-check backlog. Eventually, the backlog will cease to exist. And the main folks who have not switched to PRP-proof from LL don't tend to contribute much to double-check work. Don't hold back from doing first time tests!

To add to what VBCurtis said above, most GPUs are *much* better at trial factoring to eliminate candidates than they are at doing primality tests. Like 40 times as good. (Basically videogames use lots of single-precision units which are good at TF but currently not ideal for primality tests). The fact that you get more "bang for your buck" with GPU trial factoring tends to tilt the scale towards factoring work on GPUs because finding factors is very satisfying!

What would make me happy is being able to contribute in any way possible. Which header would indicate exponents awaiting first-time tests?

[MattcAnderson]

Welcome to the forum. I too bought a gaming for dedicated prime searching purposes. It was a water cooled system. I find that the searching for 'a needle in a haystack' is fun because we are part of a team. I also did calculations for fermatsearch.org.

Regards,
Matt

[Uncwilly]

Originally Posted by piforbreakfast

What would make me happy is being able to contribute in any way possible. Which header would indicate exponents awaiting first-time tests?

Using Prime95 just go into "Test"->"Worker Windows" on the menu. Then "Type of work to get" select "First time prime tests".
To get the most tests done by the machine overall, generally most people run 2 to 4 workers total and each with 2-4 cores, depending on the number of CPU.cores and on chip cache. Use "Options"->"Benchmarks" to see what works well. Often times having the machine work on more numbers at once give more output overall, than working on only one at a time.