You are not logged in.
Pages: 1
Hello I'm trying to install arch on a kind of supercomputer (dual cpu xeon e5 and nvidia tesla k20c).
I had some troubles with UEFI so I decided for a classial MBR boot (still with GPT). For booting I need to use the option nomodeset otherwise the screen is messed up.
With lspci the K20 is not listed under vga (nothing is found grepping vga), but as 3D controller. I installed the proprietary nvidia drivers (from the nvidia site Tesla cards should be supported). When I try to start X I get the fatal server error: no screens found.
I runned a GParted live without problems so there should be a way to start X.. but maybe it used the vesa driver.
Any idea/experience? Thanks!
Last edited by DarioP (2013-05-29 10:35:31)
Offline
Actually I was wrong: grepping VGA (capitals) I found a Matrox MGA G200e and installing the vesa driver and X started..
The next step I have in mind is to try to access the Tesla card with cuda or openCL, I would like to understand if it works as it should.
Offline
I got crazy trying to run some OpenCL kernels, but finally I have been able to set it up. Here are the steps I followed:
1) nvidia-smi (nice tool!) told me that there was not enough power.. The technician forgot to plug the cables to the card! Silly, but fundamental!
2) as the graphical output is provided by that crappy matrox chip, which requires vesa, nothing tells to load the nvidia driver. Again a simple call to nvidia-smi after the boot enables the card, to have it on automatically I wrote a simple oneshot rule for systemd calling /usr/bin/nvidia-smi -pm 1 which enables persistence too.
Here I found a simple code to check the setup of OpenCL (it prints out the error codes so you can see where and why the program stops)
http://stackoverflow.com/questions/8969 … e-the-same
...and here some additional infos that may be useful
http://stackoverflow.com/questions/5133 … untu-10-04
This should be all!
Offline
Pages: 1