You are not logged in.
Pages: 1
first of all i apologize if i've used a wrong section for posting this topic.
I got the following "issue":
In my system i'm using one GPU with 2 monitors on it. No problem installing it at all everything runs fine and smooth.
For some projects of mine i need to run some CUDA applications that i've written. The point is that those applications need to run for quite some time, and the gpu usage during those runs is minimum 99%. Therefore i get some serious lagging when using the PC even for text editing, which has to do definitely with the full working gpu which has no resources to handle the rest graphic requests from the operating system. The cpu usage and the host memory usage is very low, so there are enough resources to be able to do other cpu-based tasks while the gpu task is running.
So i was thinking of buying one new very cheap GPU, so that i can connect the 2 monitors in there, and let the other GPU handle the computations.
My question is first of all if this is possible. Am i going to have any issues with drivers, or card detection and stuff like that? Is something like this suppported from the kernel or X?
Even if all of that is possible, i started questioning about games. Will it be possible to get the old GPU handle the game computations but send the framebuffers to the new GPU?
Thanks in advance
ps: I'm running Debian right now, but i guess this question is irrelevant of the distro.
Last edited by gregkwaste (2015-11-17 21:22:59)
Offline
At work I am doing exactly (well, almost) this.
The Core i7 processor has an onboard Intel GPU, which has connectors on the motherboard, and feeds my display.
Alongside this, I have installed an NVIDIA GPU for CUDA. Thanks to (not so recent anymore) changes in the nvidia driver package, it no longer depends on the nvidia libgl, and updates are painless.
Important things to keep in mind:
1. Set the motherboard to use the correct GPU as screen. Depending on your motherboard/CPU, this may mean you either need to enable the internal GPU (including something a Dell BIOS calls a Windows only feature by the name of *will find out, needs reboot as the manual is silent about this*)
2. Install the drivers for both cards, but only the libgl package for the one you use for display (in my case I'm using mesa-libgl, and do not have nvidia-libgl installed)
3. That should be it, unless I'm forgetting something
So yes, it's quite possible, I do it every day.
Offline
At work I am doing exactly (well, almost) this.
The Core i7 processor has an onboard Intel GPU, which has connectors on the motherboard, and feeds my display.
Alongside this, I have installed an NVIDIA GPU for CUDA. Thanks to (not so recent anymore) changes in the nvidia driver package, it no longer depends on the nvidia libgl, and updates are painless.Important things to keep in mind:
1. Set the motherboard to use the correct GPU as screen. Depending on your motherboard/CPU, this may mean you either need to enable the internal GPU (including something a Dell BIOS calls a Windows only feature by the name of *will find out, needs reboot as the manual is silent about this*)
2. Install the drivers for both cards, but only the libgl package for the one you use for display (in my case I'm using mesa-libgl, and do not have nvidia-libgl installed)
3. That should be it, unless I'm forgetting somethingSo yes, it's quite possible, I do it every day.
Great, what about games? Did you ever try to check if there are problems using the good gpu for the calculations?
Offline
Pages: 1