NVIDIA 384.69 Linux Driver Released With A Few Fixes

[phoronix]

Phoronix: NVIDIA 384.69 Linux Driver Released With A Few Fixes

NVIDIA today released the 384.69 Linux driver as their latest release in the 384 "long-lived" series...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-384.69-Linux-Driver

[tildearrow]

and includes Quadro P4000 Max-Q support, an intermittent hang with Vulkan when using VK_KHR_display,

So this new driver version introduces a bug? I'm guessing you missed to add "fix".

[hrkristian]

So this new driver version introduces a bug? I'm guessing you missed to add "fix".

That's a feature.

[speculatrix]

I wish Nvidia would release a version of Cuda dev tools which work with the current stable releases of distros like Fedora.

[GdeR]

I wish Nvidia would release a version of Cuda dev tools which work with the current stable releases of distros like Fedora.

Fedora's life-cycle doesn't make things easy. New CUDA versions are generally released every year around fall and they are only compatible with the last year's major release of GCC at the time of testing, which by the way is welcomed, because that compiler is generally at a more mature bugfix update by then: think about the upcoming CUDA 9 and GCC 6.3/6.4. So basically CUDA is extensively tested on the Fedora release that ships with that version of GCC by default: Fedora 25 in this case. Problem is that Fedora 25 will basically reach EOL a few weeks after CUDA 9 will be released. This is happening each and evey year. People usually suggest me to switch to Ubuntu LTS or CentOS, but that's a heck of a big change for me just for being able to use the CUDA toolkit on a platform that is officially supported.

EDIT: Also, on Ubuntu LTS and CentOS you generally won't get easy access to a reasonably updated version of GCC by default, which is a shame because many of us just can't wait to use new features from C++, which is evolving really fast. However, CUDA is usually backward compatible up to a specific version of GCC, and GCC is the one requirement you really need in most cases, so you can still compile it yourself from source on the distribution of your choice.

[speculatrix]

Fedora's life-cycle doesn't make things easy. New CUDA versions are generally released every year around fall and they are only compatible with the last year's major release of GCC at the time of testing, which by the way is welcomed, because that compiler is generally at a more mature bugfix update by then: think about the upcoming CUDA 9 and GCC 6.3/6.4. So basically CUDA is extensively tested on the Fedora release that ships with that version of GCC by default: Fedora 25 in this case. Problem is that Fedora 25 will basically reach EOL a few weeks after CUDA 9 will be released. This is happening each and evey year. People usually suggest me to switch to Ubuntu LTS or CentOS, but that's a heck of a big change for me just for being able to use the newest CUDA toolkit on a platform that is officially supported.

thanks for that explanation.

I used Fedora on workstation at dayjob because that was mandated. I might see if I can switch to Ubuntu. I've been through several major releases of Fedora and I'm getting to the point where a fresh wipe/install is probably due anyway.

[pinguinpc]

In this drivers wine 2.15 still works