Singeing a GPU with LL work?

[Rodrigo]

Hello,

I just installed a PNY GeForce GT 430 in my new HP Slimline S5-1010. I have Brain's PDF with instructions for using CUDALucas (thanks!), but I'm not sure what the next step is. How do I get the GPU "signed up" so that it gets credit for work done?

Guess I've been spoiled by Prime95, which automates all this for the user.

Is the procedure the same as (or equivalent to) the "manual assignment" process for folks who don't use Prime95?

Maybe I'm getting ahead of myself, and it will all become clear when I actually download and launch CUDALucas?

Rodrigo

[moebius]

You can use the "manual assignment" page to assign exponents and submit the results in mersarch.txt manually, but you can also calculate a by prime95 already assigned exponent on your gpu, the exponent disapears from your assignments after submitting your result.

[Rodrigo]

moebius,

Thank you for the information.

Let me make sure that I understand what you said: I could adopt an LL assignment that was already given to one of my CPUs (but not yet running), then run it on the GPU in CUDALucas, and then (when I reported the result) that exponent would disappear from the CPU's list of assignments?

Rodrigo

[firejuggler]

yup thats it. this is how i do with TF on mfaktc

[Rodrigo]

Originally Posted by firejuggler

yup thats it. this is how i do with TF on mfaktc

Thanks, firejuggler.

Now, the thing is, my CPUs only get new assignments from Prime95 when the current LL test is already at like 90%, and they only get one new test assugnment ahead of time. So if I "adopt" the one assigned new LL for the GPU, that leaves the CPU with nothing to do at the end of its current test. Is Prime95 (or PrimeNet) smart enough to assign a new LL test to the orphaned CPU?

Also, I guess that I'd have to remove the adopted exponent from the CPU's worktodo file, else it too might start working on it, once the current LL is finished.

Right, or not really? I'm just starting to grope my way through this process.

Rodrigo

[ckdo]

Originally Posted by Rodrigo

Now, the thing is, my CPUs only get new assignments from Prime95 when the current LL test is already at like 90%, and they only get one new test assugnment ahead of time.

Consider increasing the value under "Options -> Preferences -> Days of work to queue up". The maximum is 90, which should give you several exponents (per core) unless you have a really slow machine.

Originally Posted by Rodrigo

Is Prime95 (or PrimeNet) smart enough to assign a new LL test to the orphaned CPU?

Yup.

Originally Posted by Rodrigo

Also, I guess that I'd have to remove the adopted exponent from the CPU's worktodo file, else it too might start working on it, once the current LL is finished.

Not "might" but "will".

[Rodrigo]

Very good, ckdo -- thank you for filling me in!

Rodrigo

[Christenson]

Originally Posted by firejuggler

yup thats it. this is how i do with TF on mfaktc

And I do the opposite...only use the "manual assignments" and "manual results" form...it's a matter of practice and preference...and the last time I did it for an LL test, for my GPU, I forgot to sign in first....if you want to know which is better, you'd better give us both a !

Then again, there's the checkout and results threads for Operation Billion Digits...there, you have to figure out what you are doing for yourself; there is no Automatic assignments!

Finally, ckdo has been feeding me most of my TF assignments lately...he's got an interesting little project that has been saving a few LL-D tests going.

[Rodrigo]

Christenson,

Glad you mentioned OBD. For me, ironically, the OBD assignment and reporting process is so unautomated that it's easy...

I'm running a test LL on the GT 430 (CUDALucas, of course). Got two questions so far:

1) What happens at the end of the test -- does it write a result file, like FactorX for OBD, or do you have to type it off the last line in the DOS box?

2) What happens if the power goes out during the run -- is the information saved somewhere, or do you have to start all over again? I see that my test has generated two files prefixed with "c" and "t" and then the exponent, but they are both 16,385 KB in size, so I'm not sure what their purpose is. (Opening "t" with Notepad displays a whole bunch of garbage.) I don't see any files that appear to be comparable to FactorX's "results.txt" and "status.txt".

Rodrigo

[Rodrigo]

And another question:

I just finished running a test LL for a known MP (http://primes.utm.edu/mersenne/index.html#known) (859433). This had the advantage of finishing quickly so that I could soon see what kind of result CUDALucas puts out.

The output is rather cryptic:

 
M( 859433 )P, n = 524288, CUDALucas v1.2
1:10 real 0:00 real CUDALucas: Could not find a checkpoint file to resume from

There is also a new file in the directory, mersarch.txt, whose entire contents is the first of these two lines. This appears to answer my question #1 in the previous post, but it leads to the following new one(s):

Can you decipher this for me? Had this been an actual LL run, what would I have reported to PrimeNet?

Rodrigo

[ckdo]

Originally Posted by Rodrigo

 
M( 859433 )P, n = 524288, CUDALucas v1.2

Can you decipher this for me? Had this been an actual LL run, what would I have reported to PrimeNet?

The manual results form will accept that line. It means: It was found that M859433 is prime using a 512k FFT.