Hi all,

This is I am sure a really basic question but I want to be sure. I help out on Linux servers as a student and I am just getting started.

I installed an application for someone that is CPU based. However they came back and said that it isn't much use to them because they need it to be GPU based.

They said it could be installed on one of three servers, however on each server when I run 'lspci -v' I do not believe there is a dedicated graphics card on any of these servers. Especially not one of them because it is virtual The only thing I can see is a VGA compatible controller and on the virtual, a VMware SVGA adapter.

Can I still install GPU mode for this application? One of the prerequisites for it is CUDA and I do not know if that can be installed on a server without a GPU especially not an nvidia GPU. I am assuming if I do get this installed on the server the performance will be very poor since it does not have a GPU. Can anyone elaborate on what I can expect, or what I should advise this person (also a student)?

Thanks a lot.

You can expect the application to not even start.

Can I still install GPU mode for this application? One of the prerequisites for it is CUDA and I do not know if that can be installed on a server without a GPU especially not an nvidia GPU. I am assuming if I do get this installed on the server the performance will be very poor since it does not have a GPU. Can anyone elaborate on what I can expect, or what I should advise this person (also a student)?

Thanks a lot.[/QUOTE]

Putting a graphic software on a non graphic card server will send the server to a crash, because the grapic software will be looking for the grapic card of which there will not be one to find, your, or your friends or bosses server will crash and not boot at all. Just do the web based login that it was built for, unless you would like to brick it that is.

The CUDA etc. could be installed on a client workstation with the right kind of hardware.
The application itself then can be run from the server using "ssh -X", that is: the server runs it (it probably is a faster or larger machine), but the workstation does all of the display handling.
For one specific problem (testing a very expensive nVidia card) we did install the card in the server, but as no monitor and/or keyboard was installed on it (and the server was quite some distance away) the students still had to run their applications through ssh -X from their own workstations. But now almost all of the computing was done on the server, only the visuals went over the remote X link.

Is it free software? Is CUDA truly required? Are there build options?

Anyway, CUDA is, if I remember correclty, just a library/interface to the GPU computing capabilities of the hardware, so it should compile just by having the cuda library/headers/etc on the system.

If you are installing a ready made package that wants CUDA and you don't have it, simply install the library, that should satisfy the packages requirement.

However: CUDA *does* make little sense without a GPU and if the application is graphical then, yeah, it probably relies on a graphical session, unless its maybe a command line utility but then it still wants hardware.
So...why?

Edit:
I thought I was in the Slackware forum, doesn't make THAT much of a difference but it's less likely that newcomers compile many things in most other distros.