Best type of Prime95 work for my laptops

[windowfrog]

Hi there,
I heard about the Great Internet Mersenne Prime Search in the past, but didn’t officially start using the Prime95 program until this week.

I searched on this forum and a few other sites to try to find the information I’m looking for - the best type of work for my laptops to run - but I still feel like I don’t fully understand. I’m new, but trying not to ask repetitive questions!

As of right now, I’m running trial factoring on my (very) old laptop - an Intel Core2 Duo P8400 @ 2.26GHz - and double checks on my new laptop - Intel i7-7500U @2.7GHz. Both are running about 12-14 hours a day, since I don’t feel comfortable running the program overnight.

My understanding is that trial factoring is better on GPUs, but what other short-term type of work could be done on the old laptop that won’t take months or years to complete? I attend school out of state, so having the assignments be quick is preferable. For the new laptop, is there a way to have PrimeNet assign LL/PRP/etc. tests that will still take only a month or two to complete but aren’t just double checks?

Is my thinking wrong for either or both of these, and there’s something else I should consider? I ideally want to balance the long-term distributed computing part of GIMPS with the short-term “fun” aspect of finding factors (like I did the other day, when I found a factor for an exponent and I jumped up and down in excitement!).

Any help is appreciated. Thank you!

[Uncwilly]

For the older one you can do PRP-CF (PRP tests on the cofactors of Mersenne numbers.) They tend to be shorter assignments, just 2-3 GHz-days.
You can also do ECM, those are really short. That is trying to find factors on numbers we know are composite.

There is not really any First Time checks that won't be long on the newer machine. If you have a bunch of RAM that you can spare, look at P-1. Those can be shorter, you can find factors.

Welcome to the group . Enjoy yourself.

[windowfrog]

Originally Posted by Uncwilly

For the older one you can do PRP-CF (PRP tests on the cofactors of Mersenne numbers.) They tend to be shorter assignments, just 2-3 GHz-days.

If you have a bunch of RAM that you can spare, look at P-1. Those can be shorter, you can find factors...

Thank you for the response! So the PRP-CF test would try to find co-factors of an exponent that’s already a known composite? Does that have a higher success rate than ECM?

For the P-1 test, how much RAM is needed? If I have 16 GB on my machine, is 8 GB enough? If the P-1 test only finds factors/cofactors, would it make sense to run PRP tests on the really big numbers on the newer laptop, while searching for cofactors on the old?

[axn]

Originally Posted by windowfrog

So the PRP-CF test would try to find co-factors of an exponent that’s already a known composite?

Not quite. PRP-CF tests the cofactor to see if it is (probable) prime, i.e. whether the Mersenne number is fully-factored.
Just to clarify, if the mersenne number M(p) has a factor f, then the cofactor is M(p)/f.

Originally Posted by windowfrog

For the P-1 test, how much RAM is needed? If I have 16 GB on my machine, is 8 GB enough?

Give it as much as you can spare without compromising the usability of the system. 12GB out of the 16GB should be safe to give. But even 8GB is pretty good.

Originally Posted by windowfrog

If the P-1 test only finds factors/cofactors, would it make sense to run PRP tests on the really big numbers on the newer laptop, while searching for cofactors on the old?

Sure. PRP-CF & ECM have small runtimes, and comparatively less memory requirement. P-1 has small runtime, but during stage 2, benefits from lots of memory. Beefier hardware can do doublechecks or even first time tests. For the DC/FC, and even P-1, use a multi-threaded configuration, where all your cores work on a single test.

[Uncwilly]

Originally Posted by windowfrog

So the PRP-CF test would try to find co-factors of an exponent that’s already a known composite? Does that have a higher success rate than ECM?

PRP-CF tests the non-trivial co-factors to see if they are prime. If they are, we know that the Mersenne number is fully factored. If not other testing can be employed to find factors.

ECM is one of the tests that is used in sequence to try to find factors of a number. It is not as likely to find a factor for a number as TF or P-1 (which are done first in that order). But, it is searching a larger area. The task size that is handed out is small.

For the P-1 test, how much RAM is needed?

4 GB is plenty

If the P-1 test only finds factors/cofactors, would it make sense to run PRP tests on the really big numbers on the newer laptop, while searching for cofactors on the old?

Just to be clear, there is PRP for First Time checks (this is preferred over LL now). PRP-DC to follow up on those that had a PRP as a first time. Then PRP-CF, these are working through the lower ranges of numbers.

You will have to find what works for you on your machines. The newer laptop should be ok for first time PRP's. If you find that doesn't work, then P-1 or running DC's is the next best use.

Your machine, your joy, your choice.

[kriesel]

Originally Posted by windowfrog

Any help is appreciated. Thank you!

Welcome. Some background info can be found here
The i7-7500U is ok for first time checks with patience. PRP preferred, not LL, because of stronger error checking.
For P-1 on prime95, you'll need to reset its memory allowance from the default paltry MB to something reasonable like 8000. Match the day and night settings to each other.

The i7-7500U has an igp, on which you could consider running mfakto to do some small TF assignments. Doing so would cost about half the prime95 throughput, but produce more computing credit overall. That also takes patience, since this igp is less than 1% the speed of an RTX2080. Yes 100+ times longer.
Core2 is rather costly to operate per unit throughput. I down mine for the summer. It's marginal economics in the winter when there is heating benefit.

I have systems on which I run GIMPS software 24/7, including laptops. It's easier to just let them run continuously, but costs more for electricity. I think it has little or no effect on hardware life. One is a 10 year old i3-370M.

[Runtime Error]

Originally Posted by windowfrog

I ideally want to balance the long-term distributed computing part of GIMPS with the short-term “fun” aspect of finding factors (like I did the other day, when I found a factor for an exponent and I jumped up and down in excitement!).

The GPU72 project has an option to do trial factoring with Google's Colab offering. Basically, you just open a web browser and GPU72 does trial factoring for you on Google's hardware. The setup is quite simple! Eventually, the session will end and you'll have to wait 24hrs for Google to give you more GPU time, but that's the price for it being free.

The front page of GPU72 has instructions. Also this thread is useful: https://www.mersenneforum.org/showthread.php?t=24875

(Note: all of this is found in Kriesel's linked thread, but there is a lot to unpack there)

[LOBES]

Originally Posted by Runtime Error

The GPU72 project has an option to do trial factoring with Google's Colab offering. Basically, you just open a web browser and GPU72 does trial factoring for you on Google's hardware. The setup is quite simple! Eventually, the session will end and you'll have to wait 24hrs for Google to give you more GPU time, but that's the price for it being free.

The front page of GPU72 has instructions. Also this thread is useful: https://www.mersenneforum.org/showthread.php?t=24875

(Note: all of this is found in Kriesel's linked thread, but there is a lot to unpack there)

I have an old Chromebook that is past its AUE and it just runs Colab, but I turn off the GPU and set the Runtime to NONE, that way it only runs P-1. I found that when I set it to use GPU, it times out, often after only an hour or so, due to inactivity.

However, when I set it to use CPU only, it runs for 12 hours exactly before it stops. So I only have to restart it twice a day. as opposed to GPU which might stop after an hour, and then it won't let me use another GPU anyway for a day or so.

[Uncwilly]

There is a trick that you can do to maximize the GPU time.

Once you get a good GPU (like a T4), go up to the little down arrow next to "RAM" (as seen in the attached image), once you click it, click "Connect to hosted runtime." The area where RAM was should change to "Busy". You will then get the maximum time that they will allow for the GPU. After that, you can do the P-1 for the rest of the time.
2 hours on a P100 or TF is worth a bunch to the whole effort.

[De Wandelaar]

Originally Posted by kriesel

Welcome. Some background info can be found here
I have systems on which I run GIMPS software 24/7, including laptops. It's easier to just let them run continuously, but costs more for electricity. I think it has little or no effect on hardware life. One is a 10 year old i3-370M.

Do you have an idea of the running temperature of your laptops (CPU and GPU) ?

For my part, I'm trying to limit the CPU (intel) and GPU (nVidia) temperature under 73° C (actual data :+/- 1 % of time > 72 ° C). I'm looking for a safe long term temperature but I wonder if 73 ° C is too agressive (impact on the hardware lifetime) or too conservative ?
Thanks,
Yves