I currently have a Titan Xp installed, running linux system Fedora 29. I’d like to add a Tesla V100 to the system for running Tensorflow 1.14.0, which is tested for CUDA 10.0. But am I right that the driver coming with CUDA 10.0 doesn’t support Titan Xp? If so, what would be your recommendation? Thanks.
There can only be one NVIDIA driver active in a system. But a single driver can support a multiple GPU architectures and thus a large number of different GPU models. Historical observation indicates that driver support spans roughly the past five years worth of GPU models.
Simply install the latest driver package available from NVIDIA for your platform, Fedora 29. It has support for all GPUs from the Kepler architecture on to the latest shipping GPUs, including the Pascal architecture used by the Titan Xp.
I have a GT 610 2GB and a GTX 1060 3GB that I want to run together. Would I be able to do so with the 2 separate drivers installed or will I have to lose functionality of the 1060 and drop to the 610 driver. I need to run the 610 as a Physix card and the 1060 as the primary card.
You can’t ever, under any circumstances or any OS, use or have active or installed 2 different NVIDIA GPU drivers in the same system.
Fermi devices are obsolete. They do not work with current drivers or current GPUs. You cannot use Fermi devices with GPUs that use or require newer drivers.
Now my question is why does the card I have support geforce experience with DirectX 12 support to an extent?
It’s quite possible that Fermi devices are or were covered by GeForce experience and provide DX12 support with whatever was the final driver branch that supported them, I believe it was R390 drivers.
[url]https://www.anandtech.com/show/12624/nvidia-moves-fermi-to-legacy-ends-32bit-os-support[/url]
Why trash all of the Fermi cards when they could at least enable support for them to serve as Physix cards. If R388 is able to support the 1060 and the 610 at the same time why can’t they make 441 and higher just support certain features as in compute support and extra ram?
Because carrying forward support in newer drivers incurs costs that cannot be recouped, and most people who owned such hardware carried their Fermi-based GPU(s) to an approved e-waste recycler years ago. I know I did.
I am curious: How much would you be willing to pay for a support contract that keeps outdated hardware supported?
BTW, your Fermi GPU should still work just fine with the driver you installed for it four years ago, so it is still functional and not trash unless you choose to trash it.
Greetings I’m trying to utilizy 2 GPUs now.
One is RTX 3070 and the other is 1080 Ti.
But they seem to connflict with one another.
If I just use RTX 3070 it instlls drivers for itself and works fine.
But when i also but 1080 Ti in another PCi slot it feels like RTX loses all driver powers (only one monitor is working etc) and thats all.
How should I proceed. My main goal was just to use RTX 3070 as main GPU and use 1080 Ti for Rendering
Hello,
I’m seeking assistance with installing drivers for a dual GPU setup on my x99 motherboard. Here are the details of my build:
- Motherboard: x99 with two PCIe x16 sockets
- CPUs: Dual Intel Xeon E5-2680 v4
- Primary GPU: NVIDIA Tesla M40 (for compute tasks)
- Secondary GPU: (To be determined for video output)
I plan to use the Tesla M40 for machine learning and compute tasks. However, I also need a secondary GPU for video output. My questions are:
- Is it possible to use a secondary GPU for video output alongside the Tesla M40 on this setup?
- What type of GPU should I choose for the secondary slot to ensure compatibility and effective performance?
- Which NVIDIA driver should I install to support both the Tesla M40 and the secondary GPU?
- Are there any specific settings or configurations I should be aware of to make this setup work smoothly?
Thank you in advance for your guidance!
Yes.
It shouldn’t matter much. Tesla M40 is a maxwell generation GPU, so I would choose maxwell or newer.
I would generally install the most recent driver available for the GPU. Use the driver wizard. Obviously to some extent this depends on the other GPU you choose.
I’m not aware of any. But you can find various reports on these forums of people who have struggled, and settings changes they made.
As a reminder: you never install two separate drivers for NVIDIA GPUs. Regardless of how many NVIDIA GPUs you have in a system or what kind they are, you can only install one NVIDIA GPU driver. Choose one that is compatible with both or all cards. The latest NVIDIA drivers still support Maxwell GPUs. This will obviously eventually change sometime in the future. I cannot advise about future events.
A few more comments:
-
NVIDIA recommends that Data center products (including Tesla M40) only be installed in platforms that were qualified for their use. You putting it in a random motherboard is unlikely to meet this expectation. So it may not work at all. Or, to get it to work, you may need to make various BIOS changes, update the the latest BIOS for your motherboard, or it may not work at all. I don’t have a recipe for you and won’t be able to respond to further requests if you run into trouble.
-
If you get it to “work”, a data center GPU like the M40 requires server-assisted flow-through cooling, that is integrated with feedback from the motherboard. If you fail to provide adequate cooling (e.g. via noisy, powerful external fan, or whatever) then the GPU may overheat during operation and result in a miserable experience.
Thank you for the guidance! I successfully set up my build with the Tesla M40 and a GTX 750 Ti.
The system works well after I made BIOS changes, including
- enabling above 4G decoding,
- disabling CSM.
I have a follow-up question regarding the Tesla M40’s working temperature:
- What is the normal operating temperature range for the Tesla M40?
- What is considered the maximum critical temperature for safe operation?
Your advice would be much appreciated!
I don’t happen to know this info. My suggestion would be to run nvidia-smi -a and study the resultant output. Depending on the GPU and driver version it may contain info like the following:
. Temperature
GPU Current Temp : ---
GPU T.Limit Temp : ---
GPU Shutdown T.Limit Temp : ---
GPU Slowdown T.Limit Temp : ---
GPU Max Operating T.Limit Temp : ---
The values reported above where the dashes are (—) would be your best source of info. If that info is not present, listed as N/A, or otherwise not available, then you should generally be in a “safe” operating range if you keep your GPU temp below about 80C. That should be safe for most GPUs. For best performance, you would want to keep your operating temp below the slowdown temp, if that is reported/available. in my experience, slowdown temp limit for most GPUs are at or above 80C.
Hi all,
I just bought a Lenovo P3 with an RTX A2000 12GB card in it. It has a 2nd PCIe slot and 8pin power (6+2); I have a GTX 1080 from an old system I tried to add to the current setup (A2000 in first PCIe slot, GTX 1080 in 2nd slot) but I get a bluescreen on startup. Tried updating to latest RTX drivers, looked promising but Windows then indicated it was “uninstalling this device” on the A2000 in device manager, and then proceeding to bluescreen.
Am I doing something wrong? Thanks for any help!