NVIDIA – Finally… Up to Date..!
After considerable delays, NVIDIA have finally released up-to-date drivers that support 3.11 and 3.12 kernels, without the need for patches..
I have tested 319.72 and 331.20, and both compile OK, on kernels 3.11 and 3.12..
Hopefully NVIDIA will be able to maintain their Linux driver/kernel compatibility, after this uncharacteristic lapse..
Robert Gadsdon November 6, 2013.
I’m unable to compile the driver at all. It fails while compiling unified-memory with OpenSuSE kernel 3.12.0-34.ge8fa6b4.
Did you have recognized or heard feom others about that?
Not seen this issue myself, but I have been using the mainstream kernel.org kernel for about the last 14 years or so!
Actually, I ought to mention this on my articles again..
A good place to look is the DevZone forum, and they have a thread on the new 331.20 driver, here: https://devtalk.nvidia.com/default/topic/632643/linux/-331-20-whql-long-term-driver-discussion/
Please post more data, are you using 64 or 32 bit kernel.
I’m getting already reports that 64 bit 331.20 is failing, ie, not building. 319.72 is reported to work fine on the same system, which makes sense given 331.20 introduces the new option.
Some users in
report builds with errors, but works. In sgfxi, which exits on installer error, of course the builds are reporting as failed since the module build failed:
nvidia-uvm.ko is added:
331.20 propr. driver from NVIDIA auto compiles and installs nvidia-uvm.ko.
The patch on devtalk among other things changes a license flag from MIT to GPL, which sounds like kernel made one of their apis gpl only, a trick they like doing to mess specifically with things like nvidia non free blob.
Using the nvidia .run installer command line option: –no-unified-memory
is said to fix it, or at least allow the build to occur without error, but I don’t know what that actually does so reluctant to test or recommend that. Failures reported on new 3.12 kernels 64 bit so far, but too many people don’t post their actual data so it’s hard to know for sure what makes some installs work and some fail.
There is no patch necessary for compiling driver 331.20 on kernel 3.12. Compiles and installs fine on x86_64, with kernel compiled from kernel.org. The posts on devtalk are very confused, and many are misleading.. The old patch for earlier drivers that changed the licence from NVIDIA to GPL was actually illegal, and not needed since later 3.12-rc versions, as that problem was fixed by ‘reverting’ a previous kernel patch introduced with 3.12-rc1.
Apparently I did not write clearly enough, or you read too quickly. The problem is that users are reporting compile failures, in other words, while your observation that in your case the new drivers compiled on your system, other users are reporting failures to compile.
No patches are involved in the failures. The trick here is to collect enough data to discover why it is failing on different 3.12 kernels, not to evoke the classic ‘it works fine on my system so it’s therefore fine’, something I would not have expected from you to be honest since I’ve used your blog as a resource many times because it has generally good data.
There is a patch on the devtalk forums that users claim allowed the driver to compile when it had failed without it, and there is also the option to use the installer flag, as I indicated in my first posting, which also apparently allows the compile to complete, since the nvidia-uvm.ko module is simply not built.
So on a vanilla 64 bit linux kernel, in your case it builds, good, that’s one more data point anyway.
The tone of your response is not helpful, at all. This is not a product support forum.
The place to discuss the particular problems you refer to would be the NVIDIA DevZone/devtalk forum..
I’ve posted already in the devtalk forums, but since your site tends to rank high in google for nvidia/linux driver issues, patches, etc, I thought I’d let you know that in fact there are issues with this driver for some kernels/systems, why, is unknown currently, so obviously reports of failures and of success both form data points in trying to track down the issue. I’m not looking for support from you, obviously, just gathering/distributing data. In general, it’s useful to actually read postings before you respond to them, particularly in tech areas. No need to publish this response, just letting you know. When people google on this issue and then find a report that says no issues, when issues in fact exist, one would assume that it’s worth noting this fact to help direct people to finding solutions, at least this is my approach. If you have another notion that’s fine, and I won’t bother commenting here again.