"the provided PTX was compiled with an unsupported toolchain" error using CUDA

My current CUDA version is:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Fri_Sep__8_19:17:24_PDT_2023
Cuda compilation tools, release 12.3, V12.3.52
Build cuda_12.3.r12.3/compiler.33281558_0

As I can see CUDA version here is 12.2. If that is the case how do I update it to 12.3, then?
The same was as I installed previous CUDA version?

nvidia-smi is part of the NVIDIA driver package. Per the included output it tells you that the installed driver package supports CUDA versions up to and including CUDA 12.2.

In order to run CUDA versions newer than 12.2, simply update your driver to the latest available for this GPU from NVIDIA’s website.

As of now, the latest available NVIDIA driver for RTX 3050Ti Laptop version is 535.129.03, which is installed already

In that case, the conclusion seems obvious.

No idea why no driver supporting CUDA 12.3 is available for a GPU that is not EOLed. That strikes me as highly unusual, in particular since CUDA 12.3 was released a while ago.

Apparently, NVIDIA these days implements something called “minor version compatibility”, which, if I understand correctly, allows CUDA version X.Y to run on a driver that supports CUDA version X.0. Read all about it here:

Since I don’t have any experience with this feature, I retract my earlier comments in full, and state for the record that I have no idea why you do not have an operational environment.

R545 drivers are available with the CUDA toolkit installer for CUDA 12.3. See here at the moment pointing to CUDA 12.3 update 1. If you follow the instructions there carefully, you will end up with a R545 driver that supports CUDA 12.3.

Because of the complexities of notebook graphics, if it were my machine, I would consider carefully whether I want to or need to go to the bleeding edge. If it were me, I might consider whether I might want to stick with CUDA 12.2.

This question belongs in the CUDA Setup and Installation forum, not CUDA Programming and Performance, so I will be moving the question there.