CUDA installs drivers that don't exist yet?

This refers to a driver 495.29.05, but in the driver download section that doesn’t exists yet for K and P series cards. What does this do? Does this install a not-yet-existing driver 495.29.05, or am I expected to wait with the CUDA 11.5 installation until it exists in the regular driver download/installation location?

Linux driver for Pascal here:

https://www.nvidia.com/download/driverResults.aspx/181159/en-us

The 11.5 Release Notes state: NVIDIA Driver support for Kepler is removed beginning with R495.

I know, but it doesn’t answer any of my questions. In particular, what does it mean if a CUDA installer “comes” with a driver? Shouldn’t the driver be already installed when I install CUDA? Why bundle driver installs with CUDA installs? I wouldn’t normally “upgrade” my driver with a CUDA install. I’d take care of the driver install first, test if everything works, and then install CUDA. Am I doing this wrong?
The link is useless for me, under “supported products” it doesn’t mention any Tesla. The latest for P series Tesla cards still seems to be the 470.57.02.

I gave you an answer for Kepler and if you’d mentioned Tesla, I wouldn’t have bothered with the link.

No, the release notes state that the included driver components are 495.29.05 (Linux) or 496.04 (Windows).

It also states the minimum required driver versions are >=450.80.02(Linux) >=456.38(Windows.

It also states:

For convenience, the NVIDIA driver is installed as part of the CUDA Toolkit installation. Note that this driver is for development purposes and is not recommended for use in production with Tesla GPUs.

For running CUDA applications in production with Tesla GPUs, it is recommended to download the latest driver for Tesla GPUs from the NVIDIA driver downloads site at http://www.nvidia.com/drivers.

During the installation of the CUDA Toolkit, the installation of the NVIDIA driver may be skipped on Windows (when using the interactive or silent installation) or on Linux (by using meta packages). 

The latest Pascal and Kepler Tesla Driver version is > than the minimum required version.

Thanks, that’s helpful. But I don’t find it. Instead of copying the text it would be better to show the link. I don’t find that text anywhere with the CUDA-11.5-with-driver-495 download.

First entry at the top of this page - “Release Notes”:

https://docs.nvidia.com/cuda/index.html

It also features reasonably prominently on the Download page:

Given the link in your opening post mentions “Release Notes”, I thought it reasonable to think you had already read them.

Once upon a time, CUDA installers came without a bundled {matching | minimally required} driver in the installer package.

Result: Many frustrated complaints from CUDA users who failed to or had trouble identifying, downloading, or installing an appropriate driver, preventing them from enjoying the latest CUDA features right away.

Probable cause: Some percentage of CUDA users appear to be unwilling or unable to read and comprehend the release notes that come with CUDA {and | or} follow instructions given therein.

Solution: Bundle each CUDA release with a matching driver, so that as long as the installer runs successfully, it’s a one-stop shopping experience for CUDA users.

Result: Much fewer complaints as each new CUDA version works out-of-the-box, so happier CUDA customers. Remaining complaints now are usually that the CUDA installer overwrote the user’s existing, newer, driver. Whether this overwriting happens is actually configurable during installation, but requires a tiny bit of reading to discover. Maybe the CUDA installer should add voice output :-)