Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - StyM

Pages: [1] 2
1
GIMPS / Re: Linux GIMPS trial factoring on Pascal Nvidia GPU card
« on: 14 May, 2017, 09:15:37 PM »
how do i run this program on a SLI system ? it seems to only use 1 card.

In the readme doc they have this, but not sure if that applies to SLI:

#########
# 7 FAQ #
#########

Q Does mfaktc support multiple GPUs?
A Yes, with the exception that a single instance of mfaktc can only use one
  GPU. For each GPU you want to run mfaktc on you need (at least) one
  instance of mfaktc. For each instance of mfaktc you can use the
  commandline option "-d <GPU number>" to specify which GPU to use for each
  specific mfaktc instance. Please read the next question, too.

Q Can I run multiple instances of mfaktc on the same computer?
A Yes! You need a separate directory for each instance of mfaktc.
The following users thanked this post: StyM

2
BOINC / Re: Checking BoincTasks from browser VIA IP#
« on: 05 May, 2017, 06:04:00 PM »
windows, i'm going away next week and i want to control my rigs. :)

I use the "Team Viewer" application, it's free for non commercial use and it works with Windows, Linux and Android OSes. See here: https://www.teamviewer.com/en/?pid=google.tv.download.s.ca&gclid=Cj0KEQjwxbDIBRCL99Wls-nLicoBEiQAWroh6vSKWI1AX8f7L9fmffZaIswsX_4NT42B6TmmZOnAzCkaAmcV8P8HAQ

You simply install it  on your main rig to create a TEAM Viewer account.  You then install Team Viewer on all your other rigs and use your Initial Team Viewer email address and password to get them to join your manage group. For android devices you download the Team Viewer app from google Store to your phone or Tablet and then use the afore mentioned email and password credentials to join your computer group. (Just insure that you authorize remote access during the Team Viewer Installations.)

Team viewer will let you manage all aspects of any device on your computer group. When away I use a Samsung Tab S2 - 10 inch tablet to monitor and access my computer group. Will also allow you to remotely access relatives or friends computer to troubleshoot or chat.
The following users thanked this post: StyM

3
BOINC / Re: Checking BoincTasks from browser VIA IP#
« on: 05 May, 2017, 05:16:10 PM »
Win, Linux or both?

On Ubuntu I installed Wine and then installed BoincTasks. Run Boinctasks click extra settings mobile and enter port and PW. I used 8080 for port.

On Win10 installed Boinctasks and clicked extra settings mobile and entered port and PW.

On my Unbutu box edit gui_rpc_auth.cfg and input a PW. Edit remote_hosts.cfg and and allowed hosts ip.

See here for help http://boinc.berkeley.edu/wiki/Controlling_BOINC_remotely

In whichever maching u want to be the host click computer and find in Boinctasks. and add computers.

Mel
The following users thanked this post: StyM

4
GIMPS / Re: Linux GIMPS trial factoring on Pascal Nvidia GPU card
« on: 06 April, 2017, 01:01:19 AM »
I found a script to get work and send results back for the program mfaktc. It works great.

Go to the following link to get it. A few lines down on that page there is a tab called "raw" on the right. Click on it to get the script in a better format. Copy and paste the script into a text file and name it mfloop.py

   https://github.com/teknohog/primetools/blob/master/mfloop.py

Put it in the same file as the mfaktc program.

To give it permission to run it I did:

   cd mfaktc-0.21
   chmod u+x mfloop.py

An example on how to run it:

  ./mfloop.py -u username -p password -n 10

      username is your user name on mersenne.org that had been created
      password is your password
      -n 10 tells the program to have a cache of 10 work units. If you leave it off (-n 10) it just gets 1.

The default of the program sends results and gets work every hour. There is a -t option to change this.

There are other options. You can see these in the program around line 340.
The following users thanked this post: StyM

5
GIMPS / Linux GIMPS trial factoring on Pascal Nvidia GPU card
« on: 01 April, 2017, 03:20:48 PM »
This is the procedure I used to do trial factoring on a GTX 1050ti using program MFAKTC on Linux. With that card get over 300 Ghz-Days.

Go the following post and download and extract the attached file that user attached to his post. File name is mfaktc-0.21.linux64.cuda80.tar.gz. This compiled version fixes the error "ERROR: cudaGetLastError() returned 8: invalid device function"

   http://www.mersenneforum.org/showthread.php?p=454575&highlight=invalid+device#post454575

Create a text doc called WORKTODO.TXT in the mfaktc-0.21 directory. The readme file section 3 contains an example of a couple of entries you can put in there as a trial test. The program will put the results of the test in RESULTS.TXT. After the trial test and before real work you can erase the trial work.

To run the program start a terminal and:

   cd mfaktc-0.21
   LD_LIBRARY_PATH="./lib/" ./mfaktc.exe

Hopefully your trial test worked by finding factors. You will data in RESULTS.TXT. You can delete the results file with the trail data.

To get the real work log on to:

  https://www.mersenne.org/

(Edit 4/5/2007 - See the 3rd post in this thread for script that automatically will get work and send results back. end Edit)

There will be a tab called Manual Testing. Go to Assignment to get assignments.
In the preferred work type specify "Trial Factoring". You only need 1 worker.
Paste the work it gives you to your WORKTODO.TXT file and run program as you did above.

The program put the results in RESULTS.TXT. You will need to upload them on Mersenne using the Results tab under Manual Testing.

I have been adding more work to the WORTODO.TXT file while it is running. I just do it while it is in the middle of a factor search. For the upload I also just rename the RESULTS.TXT file and upload that. MFAKTC will created a new one when it gets another result.

   



The following users thanked this post: StyM

6
Hardware info, tips & tricks / Ryzen 7 1700 PPW comparison
« on: 16 March, 2017, 01:55:34 AM »
Ok, I had the Ryzen 7 1700 only about 3 days of running on Windows 10 Pro.
I'll do a quick comparison  PPW between the 1700 and my dual E5-2695 es V4.
While these numbers may not 100% correct but I think it should be close.
Need to run like a month or something to get the perfect numbers.
Anyway, I'll pick the points for today 3/15/2017 on both machines:

1700 (16 threads) @ 3.7ghz = 165W
WCG points = 71,655
PPW = 71,655/165 = 434 wcg ppw or 62 boinc ppw

Dual E5-2695 ES V4 (56 threads) @ 2.8ghz = 305W
WCG points = 235,038
PPW = 235,038/305 = 770 wcg ppw or 110 boinc ppw

Note : Both systems are running FAAH but the 2P is running Linux mint and the 1700 is running windows 10.
I'll finish up the cache WUs then I'll install Ubuntu and compare again... Linux to Linux.
If any1 is running a single xeon V4, please post your numbers to compare with the Ryzen.
Ryzen 7 1700X and Ryzen 7 1800X's owners are also welcome  {{{}

The following users thanked this post: StyM

7
Hardware info, tips & tricks / Re: Optimal Memory Config
« on: 05 March, 2017, 02:50:50 PM »
i put 8 x 4gb per system. or 4 x 4 GB per CPU

The C612 chipset is capable of quad channel RAM so 4 DIMMS per CPU is optimal.
However not all projects are RAM speed dependent.
VietOZ runs his dual xeons in dual channel on WCG and sees no points reduction.

8 DIMMs per CPU will reduce the max speed from 2133 to 1866.
i only have 1 machine with 16 DIMMs and i do not see any points difference in DC projects due to speed reduction.
The following users thanked this post: StyM

8
Hardware info, tips & tricks / Re: Optimal Memory Config
« on: 05 March, 2017, 02:12:31 PM »
I try to build with 1G per thread so I can run CEP at any time (I always try to run just one project on a rig) but have a couple with just 16G in 32 thread machines.... No issues.

I have read of others running less. I think you might get away with 16 in a 56core not so sure about a 72. but I am sure someone has tried

Fuzzydice talked about running 24t on 8G

Most projects seem to need <200MB per thread and often less than that so add what the OS uses to 11.2 and 14.4 and you have your minimum for the above rigs
The following users thanked this post: StyM

9
Crunchers café / Re: XS DC Competition (Jan - Feb 2017)
« on: 07 February, 2017, 01:29:07 AM »
which one is distributed.net ?

Cruncher - optimal golumb ruler
The following users thanked this post: StyM

10
Crunchers café / Re: XS DC Competition (Jan - Feb 2017)
« on: 06 February, 2017, 09:44:00 PM »
golumb ruler
The following users thanked this post: StyM

11
BOINC / Re: Boinc not using all cores on win 10
« on: 12 January, 2017, 06:40:19 PM »
i remember this problem now, it has to do with NUMA balancing.

once you get over 64 cores Windows creates a second NUMA group. 

BOINC is not programmed to correctly allocate the processes to the different NUMA groups.
This script was created for me by a fellow XS member, i can't remember who now.
http://73.155.178.123:81/images/NUMA_Balancer_WCG_1_0.zip

it will re-allocate the WCG processes across the NUMA groups.

If i remember correctly, you have to set it up in the process scheduler to run every 5 minutes or so to keep everything balanced.

Because of the above issue, all my machines with >64 cores run Linux with ZERO issues.
The following users thanked this post: StyM

12
BOINC / Re: Boinc not using all cores on win 10
« on: 12 January, 2017, 06:10:41 PM »
On the WCG web site under Device Profiles there is a setting to control maximum number of cores used. Might want to check it. There are also settings for cc_config.xml to control cores used.
The following users thanked this post: StyM

13
BOINC / Re: Boinc not using all cores on win 10
« on: 12 January, 2017, 06:03:41 PM »
run two folders till you fix it maybe?
The following users thanked this post: StyM

14
The following users thanked this post: StyM

15
Hardware info, tips & tricks / Re: ES + QS
« on: 14 October, 2016, 10:26:52 PM »
i doubt it would work, but maybe if they are the exact same stepping and revision.

on OldChap's note. 
I had 2 identical SPEC ES CPU's one stamped USA and one stamped Malaysia. They would not work together, separately they were fine.
The following users thanked this post: StyM

Pages: [1] 2