Hi Everyone,
I'm running 2 GPU's in parallel, usually for GPUGrid.

I also run LHC. I dedicate 8 out of the 12 cores to BOINC.

Up until a couple of days ago, LHC was taking 6 CPU cores, running one LHC WU each.
GPUGrid is running tasks on my 2 GPU's and take 1 CPU core each. (so one GPUGRID WU per GPU and 1 CPU core per WU)

This has been running flawlessly for ages.

Recently however, LHC cannibalises one more CPU Core, effectively depriving GPUGrid from running a second WU on
my second GPU.

I've been playing with the resource settings, but in vain.

Can anyone help how to restore the balance to 2 GPU WU's + 2 Cores for GPUGrid and 6 CPU cores for LHC (and 6 WU's) ?

Your support would be most appreciated !

BelgianEnthousiast.

Just to complement : I also observed the same behaviour with MilkyWay WU's on the GPU.
Only one GPU is working, the second one remains idle while LHC keeps controlling 7 CPU cores...

For LHC@home you can set the max # of CPUs (cores) to utilize:

Login-> Click on your name in upper right if you are taken to Your Account Page -> Click at 'Preferences for this project'

You'll be here: https://lhcathome.cern.ch/lhcathome/prefs.php?subset=project


There you can set the max # of CPUs (Edit Preferences). Don't forget to SAVE!

This won't affect any work CURRENTLY in progress. If going forward there are still too many cores being used then you will need to let them know via their forum so they can investigate.

~Y

Thanks for the discussion..
I have several rambling questions:
a)
Even though I set up my preferences to allow GPU, it has not happened.
Is there a way to check server status to see which sub-projects of LHC@home use the GPU?

b) I have been running Einstein@Home for years - doing GPU work.
Do I have to allow No Work from Einstein to let the lhc@home work to get to the GPU?
Or, can LHC@Home "share" my queue of GPU WUs?

c) Einstein@home needed a CPU "free" to handle getiing data to/from the GPU.
How is this done with the LHC@Home work?

d) I noticed that lhc@home does not honor the network limits that are made available by the BOINC preferences.
Is there a way to do this?
I'm in the US and use DSL for both internet and cell-phone wifi.
LHC@Home WUs kick me off my phone. :-(
(Yes, I can set Quality of service on the cell phone MAC address, but that is a pain in the .... )

e) I was running the LHCb task and was set up to run 5 tasks.
AS the memory utilization crept up, Boinc held back a WCG task with an error messge that said "Not enough memory."
I have 11.6 GiB of memory and only (hah-hah) 79% was in use.
I dropped LHC@Home down to 4 tasks to deal with this.
Anyone care to shed light of how available memory is calculated and required?


f) Sometimes, LHC@Home wouldn't run because BOINC did not see Virtualbox.
I got around this by:
1) suspending all BOINC activity
2) Exiting BOINC Manager
3) a group of commands:

sudo service  --status-all

sudo service boinc-client stop
sudo service virtualbox stop
sudo service virtualbox start
sudo service boinc-client start

Restarting BOINC

My question is:
Is there a good reason to let VB autostart - even when I am not running BOINC???
( or Days when I disable LHC@home.)


g) [soapbox]
Running 4 LHC@home CPU tasks did NOT use much cpu at all - in comparison to other projects. :-(
So why can't CERN write tasks that run in a BOINC environment that don't use VirtualBox?
[/soapbox]

I have an 8 kernel AMD64 that is running Ubuntu Mate 17.10.

Please share your opinions and experiences.
T H A N K Y O U,
Jay

-- edit -- fix typos and clarify MAC address. Add (obvious) step of restarting BOINC.

In general, it's best to take project-specific questions to the project message boards, since the people there know it best.

LHC does not currently have any GPU applications; you can check the applications page of a project to see what architectures it will run on. If there is a GPU app, you will see something like CUDA, ATI/AMD, OpenCL or the like listed. https://lhcathome.cern.ch/lhcathome/apps.php

---

My Detailed BOINC Stats

In Windows GPUGrid needs swan sync environment variable setup and an entire CPU core dedicated to feeding the GPU. I use Process Lasso to keep all other processes away from a thread where only the GPUGrid exe runs. I have not had an issues in Linux though.

Einstein requires a minimum of 1gb of video memory. I have a Dell Latitude that has an NVS4200M GPU chip. But it is only 512mb of memory.

Somewhere, probably in the forums, a project ought to have info on how much memory you need with your GPU chip/card.

SETI probably the most forgiving of the projects, but there might be another project that fits that bill.

There is no global setting in the manager for NO CPU. IIRC you can modify your config file manually though.

You CAN set it per project though in the PROJECT preferences.

There's a post about running GPU here. It's identical to the one at SETI. I suggest you run through it if you haven't already.

Have you looked at the official (non-exhaustive) list of BOINC projects?
http://boinc.berkeley.edu/projects.php