I have a pre-production tesla installed in the first X16 slot and a geforce 8600GT in the 2nd x4 slot that i use for my video. I have the 8600 driver options set in xorg.conf for BusID “PCI:2:0:0”, my bios set to initialize PEG2 and my desktop works well. This is only a 256mb card so cuda does not use it since my desktops take up most of that. I do not have any device configuration in xorg.conf for the tesla on PCI:1:0:0.
When my machine starts from a power up, without X starting, the tesla gpu sets up at 599mhz and full speed fan. When X starts, it initializes the tesla, drops the fan back to 26% manual instead of auto and the gpu sets to 499mhz. the shaders and memory clocks all drop as well.
I want to make it so X when it starts does not touch the tesla at all in any way.
Are there any xorg options or nvidia driver options to accomplish this?
Is this possible? if so how?
my nvidia driver is version 180.60 and I am using Gentoo X86_64 o/s
the kernel module enumerates all devices. why would you want this to not happen? as soon as a CUDA app starts it will increase the clocks to where it should be (and the fan is dependent on temperature).
Unfortunately that is not the case. I have to manually set the fan back to auto using nvclock since the --info display shows it set to manual settings not automatic like it is on powerup before X inits. And when seti cuda app runs the clocks do not change back.
So, I want to eliminate the initializing of the card and just use it as it is on power up. I suspect it would work a heck of a lot better. The tesla is not disturbed by loading the nvidia module since I do that during boot and the cuda app works well when left as a ‘server’ with no X running. Only when X starts up is when it bothers the tesla and the cuda app execution times slow noticably. I wish to keep X from touching it in any way.
oh, I just noticed pre-production Tesla. Yeah, don’t use that. It’s not really supported at all anymore because we know of ways where it just doesn’t work (hence it being pre-production).
hmm… can you give me basic info about what it either can or cannot do? even though it has video i do not use that. i wish to use it strictly for seti@home number crunching. i have had some problems with it locking up but i suspect it may have been too weak a psu. I just installed an 850w psu this morning but since the project is down I have no way of testing whether it now works or will still lock up after 30 min or so of crunching. temps are not a problem, it never gets beyond 68c and even then it only hits that on momentary peaks.
my boss picked this up for me from somewhere, so if it has severe issues that will prevent me from using it at all, i would like to be able to list things and point saying “this is official”. we do understand it is not supported but i initially assume it does work hence i am trying to be as ‘generic’ as possible in my approaches to problems to weed out problems with this tesla or problems with my installation.
If it has video out, it is a very early sample and is broken in a whole lot of ways. If you file a bug on that silicon, we will not support it at all because we already know it has quite a few problems. Assume it doesn’t work until proven otherwise–that’s more accurate to describe what that silicon is.