GIMPS installation guide

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,574
GIMPS uses the Prime95 app which is the same as SOB. It is a little easier to understand for setting up because you don't have to make changes to a config file.

First, you need to register an account with the website at www.mersenne.org/gettingstarted/ Make sure you add our team under your Account Settings. Our team name for this project is HardOCP



Next, click on the link to download and install Prime95 on the second step of that screen which should bring you to a page that looks like the one below.



Choose the version that best suits your system and extract the files to a folder on your computer. Remember where this folder is located. In that folder you will find a file called Prime95.exe. Run that file and you should see the app run with a Welcome message. Even though you already created an account, click Join GIMPS! button anyway.



The next screen is where you enter the user name you created with the website. I recommend also giving your computer a unique name for easy reference in the future if you add more to your list. Hit OK.



The CPU Settings and Information screen basically tells the client how long and at what times to run the app.



If you left the default memory setting at 8MB, you will get this prompt. If you click yes, it will take you back so you can change the amount to something higher.



The next screen tells the client how many workers to run. This essentially is how many cores you want used by the client and this is how many work units will run at a time. You can also change priority levels and such here.



Now you should see something similar to this. If you hover your mouse over the icon next to the clock, you should see what percentage is complete on each worker.

 
Last edited:

Gilthanis

[H]ard|DCer of the Year - 2014
Joined
Jan 29, 2006
Messages
8,574
http://hardforum.com/showpost.php?p=1037748416&postcount=114

I couldn’t find a Noddy guide anywhere so everything was a bit trial and error unfortunately. To ease the pain for others I’m willing to give a guide a bash sooooooo the steps for Nvidia Cuda graphics cards under Windows 64 bit are:

Download the mfaktc-0.17 Windows x64 client and cudart64_32_16.dll files from here (links are in the pdf guide) and create a folder into which you extract the files.
Create a empty text file within the folder named ‘worktodo.txt.’
Sign into your GIMPS account and open up the ‘manual test’ sub link ‘assignments’.
Select ‘trial factoring’ from the drop down menu, enter 1 into the number of cpus and then the number of assignments you want. I request no more than a 100 at a time – any more and you risk the server timing out. They're working on automating the assignment/results process but aren't there yet.
Copy and paste the provided assignments into the worktodo.txt file you created. The TF assignments typically only run to a bit level of 71 so if you have a fast Fermi GPU the given assignments will be swiftly completed. You can slow this down by manually increasing the bit level of the assignments by a little to say 73 or 74 which will increase the depth of the Trial Factoring. You do this by changing the last one or two digits of the assignment. From what I’ve read this is apparently normal practice although I’m not entirely sure how far you should take this.
If you’re running more than one GPU on a single PC you’ll have to run multiple instances following the above instructions for each. Create a shortcut to the mkfakt executable in each folder then right click and open properties. Add –d 0 to the target for the first gpu, -d 1 for the second etc. It’s similar to running multiple F@H GPU clients in this respect.
There are settings you can play with within the config file but best leave as is until one of us gets brave.
Run using the executable or, as appropriate, the shortcuts.
When you are ready to send results, sign into your GIMPs account, select the ‘manual testing’ sub link called ‘results’ and use the upload button to upload the results.txt file within the folder you created earlier. Always wait until an exponent has completed it’s full run to the set bit level before uploading it’s results and remember to delete the results file after a successful upload – a new one will be created when you complete more work.

With a fast GPU, your CPU will be the bottleneck so it is recommended that you use a full core with HT switched off. There is no facility to run on multiple threads so, although I’ve not tried this yet, you can run more than one instance per GPU thereby effectively doubling the amount of CPU cores supporting each GPU. This is said to increase output by over 50% and push you closer to 100% GPU usage. I’m really going to watch out for the heat output when I give this a go at the weekend.

I’m again working away so please forgive any mistakes/omissions as all the above is from memory. Don’t hesitate to shout if you hit a problem although it may be a case of the blind leading the blind I'm afraid.

Best of luck.

Bob

PS Although GIMPS is sucking up some Muon resources the serious drop in Muon output today is from my better half inadvertently cutting off the ventilation to my fastest rig which, rather unsurprisingly, caused it to power off for most of today. Normal service has been resumed.
 

plext0r

[H]ard DCOTM x3
Joined
Dec 1, 2009
Messages
780
It is recommend that if you're running Windows, install MISFIT to handle the automatic download and upload of GIMPS work units. I just installed MISFIT 2.7.5 on my first Windows GIMPS client and it was fairly simple to configure. See http://www.mersenneforum.org/misfit/ for more details.
 
Top