I installed Ubuntu 8.10 x64 on my new machine (q6600, GTX280 )

The install went great, I got everything going, and I was able to install Boinc through the add.remove programs in linux.

However that edition was 6.2.12, which did not recognize by GTX280

This could be a few things,

first off, Im not sure if I have the right drivers for the card on there
I downloaded the new drivers, and had issues installing them, then Ubuntu popped up and said to activate the 177 drivers for nvidia(it does not say what version of drivers (ie. 177.88 ~ or whatever)

Now, when I tried to install the new version of BOINC (6.4.5) it will load and it will load the 6.2.12 version

I really want to stick with Linux (stronger crunching), but I am so used to windows ( I am a BIG Linux NEWB ) that I cannot make my way around linux


Some google searching resulted in me figuring out how to install .run(Nvidia drivers), and .sh(BOINC) files. But I still do not know if I am doing it right.

I just started a day of work so Ill will be here for another 8 hours , and until then I cannot touch the machine.


But I would REALLY apprecaite any help you can give now, I know there is lots of guys in here using linux, hopefully you'll understand what Im talking about




Thanks !!!! This machine should be 12k + RAC so I am really excited to get it going! (I ended up staying up till about 3am trying to get this to work - and its still not!)

@Jon Boy UK (Shout Box), that is the method I used, however it still loads the 6.2.12 version.

I went into my home folder and erased the BOINC install that Ubuntu made, however when I run the run_manager in the "new" BOINC folder that it makes on the desktop it still loads 6.2.12.

Tony the 6.2.xx series doesn't support GPU crunching.

Download the 6.4.5 (best version available in my opinion) from Berkeley. Delete the 6.2 from the package manager.

put 6.4.5 on desktop open a terminal type sh then space then drag the d/l into the terminal hit enter and it will install in the home folder.

In the Boinc folder find run client right click and run then manager the same way (this is how I start Boinc on my GPU machines.


The reason it keeps opening 6.2.12 is because it is installed thru the package manager and auto loads on boot....you can't use that install is all and you are going to the wrong place to find BOINC as now it is loaded in 2 different locations in the file system Your Best option is to uninstall Boinc completely from the synaptic manager.

There we go!! Thank You!!

That is what I could not find last night(synaptic manager), I knew if had 2 installs on there but I could not get the 6.2 off.

Thanks Jeff!

I'm gonna run home on lunch and see if I cant get that going!

any recommendations on remote login software for linux?

Tony,

I had the same problem myself a little while ago and Jeff kindly walked me through the process
Jeff has answered your question - Happy days !

The only drawback to using Linux for GPU crunching is as far as I can find there is no utility available for the newer cards for OCk of the GPU.I have gotten around this sorta by buying factory OCk cards the highest I could find....hopefully NVIDIA will come out with a utility that works soon

As you know Tony I haven't done remote access yet and I assume now by your question that that link you gave me for remote access was for Win only ? Since most of my hosts are Linux if you find a good remote login manager please let me know....

Overclocking Nvidia cards in linux - http://www.linuxhardware.org/nvclock/

Just found some remote software too! Tested it out in windows and it seems pretty easy so far.... Ill let you know how it goes in linux

RealVNC - http://www.realvnc.com/index.html

I have not used either of the programs in linux yet, but I sure will be tonight!

Tony NVclock,NQclock,and NGclock do not work on the newer cards with CUDA2.0.....just posted a query at GPUGrid...maybe someone has found something that works

NQclock does allow me to change fanspeed on the 9800GT but doesn't work on the GTX260

ahh, well I guess its good that I know that now, and not after 4 hours of trying to figure it out tonight

Im really hoping they get something working too! I have my GTX280 in my windows machine running at 671 now, and the linux machine is only at 616, that will prob make quite the difference...

the new version claims to support G200 chips

Hey Jeff.... you seen this yet??

quoted from NVClock's website

Sunday January 4 2009
Due to huge demand I'm releasing a new version of NVClock. NVClock 0.8 (beta4) offers support for the latest cards and improves functionality for current models like smartdimmer support for Geforce8/9 boards and fanspeed adjustment / temperature monitoring on most Geforce8/9/GT200 cards. I would like to thank Michael Vassiliadis for his smartdimmer debugging and neldoreth for testing the fanspeed code. Further I would like to thank everyone who performed tests for me as myself I don't have access to much Nvidia hardware (mostly nv4x cards and a fanless 8600gt.

# Changes: Initial GT200 support
# Smartdimmer support for various Geforce8/9 laptops from Sony, Samsung, HP and Apple
# Added an smartdimmer program to replace the original smartdimmer 0.1 program by Erik Wahling
# Bios dumping using PRAMIN to prevent flickering
# Bios parsing fixes for newer Geforce9/GT200 bioses
# Fanspeed adjustment on various Geforce8/9/GT200 cards
# Use of video bios sensor calibration data on Geforce6-GT200 cards
# Improved support for internal GPU diode temperature on G84/G86/G94/G96 cards
# Fixed NV-CONTROL crash in nvclock_gtk
# Many small bugfixes

Cool!Thanks Frank and Tony....heading to there now to d/l and try....hopefully the new support does allow OCk....at least fanspeed control is a plus

Have any luck with it?

Limited success.....not sure I installed it correctly....tried it on the 9800GT and it does OCk with the utility but its not holding.

It is not an easy install procedure and I am not a Linux guru

That makes 2 of us I will do my best on it tonight, hopefully I can get it going.....

Any thing I can come up with Ill be sure to bring back here!

Got the linux machine up and running last night.....

I was REALLY low on time, so no OC on the CPU or the GPU yet

The machine is pulling down GPU and CPU work and knocking it out like crazy!
?????????????
From what I have read, in windows you run 3 cores & 1 GPU,

In Linux it is OK to run 4 cores & GPU right? I was monitoring progress by start time and % completed with a nice calculator I made in and the progress seemed on par with that of my windows machine with 3/1

******

The linux machine..... check out the summary page.

It is only reporting half the power of my other quad....
It is completing jobs in the right amount of time, but the benchmarks are not reporting properly.

Is this going to effect my credit for this machine? Its running PG, milkyway, & GPUGRID

4+1 in Linux works because the cpu uses like 2% for GPU.

Everything looks good from what I see...

Don't worry about the benchmarks on the projects you are running as they aren't used (on PG it depends what sub-project)but they do seem low even for Linux.

I was going to re-run them but I really didnt care that much..... Ill do it tonight

heres an interesting one... in windows the GPU reports .03 CPU 1 CUDA
...................................... in linux the GPU reports .06 CPU 1 CUDA - at first I was only getting 2 tasks, but it jumped to 4.

So far the Linux box has finished 3 WU's from GPUGRID, one from each credit range, they have all taken almost exactly 5.5 hours!

And wow, BOINC linux is excellent for communicating with projects. Right when a WU finishes it uploads, communicates with GPUGRID, and I get a new one, all within minutes of the completion.

I think its time I start warming up to linux for all my machines..... just gotta figure it out first

not good - they should show different runtimes - credit is granted according to work done..



this is not a linux thing - my windozes are doing exactly the same since 6.4.5.

and by the way: of course you can run 3+1 or 4+1 on win, but 4+1 slows down the GPU-task. how much? depends on the speed of CPUs and the kind of project you are running on CPUs.

And they do, I should have been more clear, the average is around 5.5 hours, and I have found that the CPU time that GPUGRID displays is off pretty badly (I have been monitoring start times). It seems like the CPU Time that GPUGRID reports is the CPU Time that BOINC said, which is wrong.

Here is my linux box, if you notice the times they are all over the place, but I looked over the message log to get a better idea of run times. I remember being told that GPU's finish in 3 hours, I have never had a sucessful GPU WU finish that quickly. And Im pretty sure my GPU is working 100%, Im getting 10 to 12k a day, which is a good average to have for a GTX280...... atleast compared to others.


It might not be a linux thing, but my windows machine with 6.4.5 does not update right away, nor is it great at getting work, I have to check that computer every 6 hours or so just to make sure I dont lose my queue of work.

When I run 4+1 on windows it takes 10 hours if not longer to finish a GPU project, (milkyway, PG on CPU). And thats not worth it, the credit from GPUGRID is a better "value"

I think your getting lucky with the updating on your windows computer, cause I sure am not. I have 3 windows machines with GPU's (2 are way to slow to matter(switched then to seti)) but all three are terrible at updating themselves and getting work.....

CPU-time is not telling much - when i run 4+1, if find CPU-times of some miuntes..

you need to check the task-details: http://www.gpugrid.net/result.php?resultid=202010




it wasn't some days ago, but last 3 days were smooth.



as far as i can see, those *GPUTEST* WU's do need much less CPU-work than the other types.

So Frank, with the work that finished so far, my math tells me that computer will do ~11,500 cred/day. Is that what I should be expecting from a GTX280?

I cant totally argue there, it is much better now than it was before the first of the year. Except for the 24 hour delay if they have no work available at the time. Still dont really understand that.

So it would be smart for me to keep the windows machine at 3+1 then?

This is kinda a weird situation, I have to leave my BOINC prefs to 100% CPU, and it will only run 3 cores and the GPU, but if any of the projects goes into high priority then I have to lower it to 90% CPU usage so it only uses 3 cores. Otherwise it uses all 4 and starves the GPU of the CPU cycles that it needs.