Where should I start?

[christian_]

Hey everyone! I'm new to this project, and so far I'm enjoying it. Matt Parker's video on standupmaths introduced me to GIMPS. Maybe I'll be the lucky person to find M50, and if not, I'll make some contribution I'm sure.

Anyways, I have both a GTX750 and an Intel Pentium G3258 to contribute, however I don't know what use they would best suit.

I have my 2 cores on the CPU set to 'Whatever makes sense' in Prime95 (both are currently doing DC's), and my GPU is running mfaktc with GPUto72 deciding what to choose (currently doing a LL TF).

However, I have also got CUDALucas installed, which I think is better at doing first-time LL's. With all this in mind, does anyone have any suggestion where my computing power would be most helpful?

I'm very new to all of this terminology so please forgive and correct any errors. :)

[kladner]

Your CPU is probably in a good place, at least until you get a couple of good double checks completed.

For the GPU, you might want to look here, for mfaktc performance,

and here, for CUDALucas performance. The comparisons might be helpful, though I can tell you that right now, doing LLTF as assigned by GPU72 is especially valuable to the project.

It also seems that your GPU probably does better at TF work.

These are just opinions, and off-the-cuff ideas. I am sure that others here will chime in, especially if I misstated anything.

[Batalov]

Originally Posted by christianpoland

However, I have also got CUDALucas installed, which I think is better at doing first-time LL's. With all this in mind, does anyone have any suggestion where my computing power would be most helpful?

Running a DC on each new device at least once is a good thing to do - both for the project and for your own benefit. When the result will match, you will know that the system is working well. When/if it will not match, it will help you to tune the system.

And after one or two matched DC, absolutely, do move on to LLs.
That is what the server will in fact try to do for you if 'Whatever makes sense' is selected.

Good luck, and welcome!

[christian_]

Originally Posted by kladner

Your CPU is probably in a good place, at least until you get a couple of good double checks completed.

Originally Posted by Batalov

Running a DC on each new device at least once is a good thing to do - both for the project and for your own benefit. When the result will match, you will know that the system is working well. When/if it will not match, it will help you to tune the system.

And after one or two matched DC, absolutely, do move on to LLs.

Thanks for the advice! The GPU is expected to finish in an hour or so, then I'll assign a couple DC's.

[Uncwilly]

Welcome to the project. As the others pointed out, GPU's excel at Trial Factoring. They are so good at it, that in general that is the best thing to have them do. Doing a couple of Double Checks on them is a good test.

You may find that you like one type of work more than another. Doing what makes you feel good is a fine thing.

[kladner]

Originally Posted by Uncwilly

Welcome to the project. As the others pointed out, GPU's excel at Trial Factoring. They are so good at it, that in general that is the best thing to have them do. Doing a couple of Double Checks on them is a good test.

You may find that you like one type of work more than another. Doing what makes you feel good is a fine thing.

And indeed, Welcome!

[pinhodecarlos]

Can I make a few questions?

1) What's the most efficient way to LL, using CPU or GPU?
2) When it will be possible to run GPU (AMD or Nvidia platforms) commanded by Prime95 instead of using a third party software (manual reservation).
3) On Prime95 how do I set an LL to be run on all cores of my CPU instead of running 1 instance of LL per core (HT off)?

Thank you in advance,

Carlos

[LaurV]

Originally Posted by pinhodecarlos

Can I make a few questions?

1) What's the most efficient way to LL, using CPU or GPU?
2) When it will be possible to run GPU (AMD or Nvidia platforms) commanded by Prime95 instead of using a third party software (manual reservation).
3) On Prime95 how do I set an LL to be run on all cores of my CPU instead of running 1 instance of LL per core (HT off)?

Thank you in advance,

Carlos

1. Most efficient is still with the CPU. The GPU is "fastest" (unless you have hundreds of CPU cores like Madpoo or Chalsall ). For example a Titan worth about 10 cores or so, of the newest CPUs (i.e. is so fast as a 10 cores CPU, for example), but it consumes more than double on the power bill.
2. Never. There are tools for automatic assignment for GPU, if that is your problem, and having the cuda/cl/Lucas as separate applications helps with the "independent" verifiction/double checks. It does not worth the effort to integrate all in one, and it will not be easy to maintain, as people doing them have different specializations, and they have real life too. It is not the same to code for cpu or gpu.
3. in the Test/Workers... menu: Set the numbers of workers to 1 and the number of cpu's to use to your number of cores, then restart (from test/exit) the p95 program

[christian_]

Originally Posted by LaurV

There are tools for automatic assignment for GPU, if that is your problem, and having the cuda/cl/Lucas as separate applications helps with the "independent" verifiction/double checks. It does not worth the effort to integrate all in one, and it will not be easy to maintain, as people doing them have different specializations, and they have real life too. It is not the same to code for cpu or gpu.

Thanks for the advice, although for another person. Would you be able to link me to the tool you mention for automatic assignment?

[kladner]

Originally Posted by christianpoland

Thanks for the advice, although for another person. Would you be able to link me to the tool you mention for automatic assignment?

This link is to the latest version of MISFIT created by Scott, SWL551. It handles assignments for Trial Factoring, and includes many useful automated features.
http://www.mersenneforum.org/misfit/...FIT-2.10.0.zip