RTX3060 + RTX4070Ti Linux CUDA: How to add it?

Hi all.

I’m upgrading a workstation that already have a RTX3060 12Gb by adding to it another card, a RTX4060Ti 8Gb.

In a first attempt, I just plugged the new card, booted it up and checked nvidia-smi: seems ok but certainly it is only seeing the old card, not the additional one, since it only reports a 12Gb card in the system.

Since installing the card on my opensuse system was already a huge pain, I would like the less effort route to increase the TFlops available for my gromacs simulations (I plan to run single simulations on each card, and not a single simulation between both cards together).

Can someone please help me on this?

Thanks a lot in advance! :)

What is the output of nvidia-smi?

Hi rs277: Thanks for the reply.

lspci, dmesg and nvidia-smi outputs follows:

kondepudi:~> sudo lspci | grep -i vga
03:00.0 VGA compatible controller: NVIDIA Corporation Device 2788 (rev a1)
08:00.0 VGA compatible controller: NVIDIA Corporation GA106 [GeForce RTX 3060 Lite Hash Rate] (rev a1)
kondepudi:~> dmesg | grep NVRM
[   11.645237] NVRM: loading NVIDIA UNIX x86_64 Kernel Module  525.60.13  Wed Nov 30 06:39:21 UTC 2022
[   29.650996] NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x30:0x56:1005)
[   29.651047] NVRM: GPU 0000:03:00.0: rm_init_adapter failed, device minor number 0
[   29.738440] NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x30:0x56:1005)
[   29.738485] NVRM: GPU 0000:03:00.0: rm_init_adapter failed, device minor number 0
...
kondepudi:~> nvidia-smi
Tue Dec 17 14:30:42 2024       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.60.13    Driver Version: 525.60.13    CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:08:00.0 Off |                  N/A |
| 37%   36C    P0    39W / 170W |      0MiB / 12288MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

Any ideas?

Nothing obviously wrong, but the 4060 isn’t happy. All I can suggest is to try updating the driver to the latest available.

Given that the driver installation in these systems are not a simple feat for some reason, can I ask:

  1. All I should do is just plug and the system should run smoothly directly, nothing else needed?

  2. Should those two be able to operate in the same machine without any hassle?

  3. Would it be a good idea to try to swap both in positions, or place only the 4060, or risk something else?

Thanks!

Yes, there should be no problem with both cards in the system together and the driver currently in use should support the new card.

No harm in trying the 4060 on it’s own, but before doing that, one thing to double check, would be that the power connector is securely attached to the card and if you are using any sort of power adapter, eg 8p → 16p, that it is working correctly.

Sorry, it looks like the 4060Ti is not supported with this driver - see here:

Driver Details | NVIDIA

It needs to be at least 525.147.05

Hi rs277.

Ok, the link you provided does not mention the RTX4060Ti in the supported list, so that means that unfortunately I will have to reinstall driver and probably cuda.

Can I be Hopeful that I’ll not have to recompile the softwares I’ve already compiled with the older driver? :(

I’ll begin to give it a try later this week. I’ll keep you posted, right now just crossed fingers.

One question: would it be better or not to reinstall the driver with the board plugged on it? I just noticed that the Xwindows does not come up with it on, for obvious reasons now. :P

You should not need to reinstall cuda, as long as it’s version 11.8 or greater - when support for 4XXX series cards was added.

It should not make any difference whether the card is resident or not when updating the driver.

Hi all.

Sorry for taking too long to answer back.

In the end, the trick was at the driver, but given that opensuse leap 15.2 (with nvidia dirvers + cuda installed the “hard way”) could not support an updated driver for the card (and 15.4 was a no go at the time, t simply didn’t work with cuda), I had to upgrade it to 15.6.

However, I’ve had the pleasant surprise that it worked almost “automagically” with the rpms, what is a great improvement. :)

Anyway, thanks a lot for the help, I would had taken a long time more to verify the card X driver version issue. :)