[elrepo] nvidia K1 issue

[Steve Cleveland]

This is reviving an old thread, so I'm not sure if best to top-post, but 
I wanted there to be context to the question.

At the time (pre el7.4), to support a K1 card, I just need to load the 
mainline kmod-nvidia package.  But in 7.4, that driver (367.57) does not 
work.

I assume they have not provided a legacy 367 driver?  Would it be 
possible to build their last 367 release as the kmod-nvidia-367xx 
version?  And can it be available on 7.4 or are we past that window?

We run these servers behind on updates (just upgrading to 7.4), but we 
do have several to support.

Thanks as always,

  - Steve



> On 07/24/2017 11:10 PM, Phil Perry wrote:
>> On 25/07/17 00:31, Steve Cleveland wrote:
>>>
>>>
>>> On 07/24/2017 12:27 PM, Phil Perry wrote:
>>>> On 24/07/17 19:12, Steve Cleveland wrote:
>>>>> I believe my issue is the same as this:
>>>>>
>>>>> https://elrepo.org/bugs/bug_view_page.php?bug_id=724
>>>>>
>>>>> But I can't see how to reopen it or comment on it.  I have a K1 
>>>>> card. nvidia-detect is telling me I need kmod-nvidia-367xx, which 
>>>>> doesn't appear to exist.  The difference is the K1 card does not 
>>>>> appear to be supported in the 375.66 driver like the K2 card.
>>>>>
>>>>> # rpm -q nvidia-detect
>>>>> nvidia-detect-375.66-1.el7.elrepo.x86_64
>>>>>
>>>>> # nvidia-detect -v
>>>>> Probing for supported NVIDIA devices...
>>>>> [102b:0534] Matrox Electronics Systems Ltd. G200eR2
>>>>> [10de:0ff2] NVIDIA Corporation GK107GL [GRID K1]
>>>>> This device requires the legacy 367.xx NVIDIA driver kmod-nvidia-367xx
>>>>> [10de:0ff2] NVIDIA Corporation GK107GL [GRID K1]
>>>>> This device requires the legacy 367.xx NVIDIA driver kmod-nvidia-367xx
>>>>> [10de:0ff2] NVIDIA Corporation GK107GL [GRID K1]
>>>>> This device requires the legacy 367.xx NVIDIA driver kmod-nvidia-367xx
>>>>> [10de:0ff2] NVIDIA Corporation GK107GL [GRID K1]
>>>>> This device requires the legacy 367.xx NVIDIA driver kmod-nvidia-367xx
>>>>>
>>>>> Has there been a 367.xx legacy release from nvidia?
>>>>>
>>>>> Thanks,
>>>>>
>>>>>   - Steve
>>>>>
>>>>
>>>> Hi Steve,
>>>>
>>>> I believe nvidia dropped support for your GRID K1 card in the latest 
>>>> 375.xx series driver, so you need to stay on the last 367.xx series 
>>>> driver.
>>>>
>>>> The last 367.xx series driver released by nvidia was 367.57.
>>>>
>>>> Apparently nvidia will support your card going forward with legacy 
>>>> releases to the 367 series, but as of yet they have not released any 
>>>> updates. So the most recent (current) 367 series driver is the 
>>>> 367.57 release. When nvidia release a legacy update we will release 
>>>> that as kmod-nvidia-367xx legacy driver, or at least that was the plan.
>>>>
>>>> I updated nvidia-detect to report the 367.xx series driver as the 
>>>> correct driver for those cards anticipating nvidia would make a 
>>>> legacy release imminently but they haven't. Sorry for the confusion 
>>>> that has caused.
>>>>
>>>> For now, you just need to install and stick with the last 367.57 
>>>> driver release.
>>>>
>>>> Assuming you have yum-plugin-nvidia installed, the latest 375 series 
>>>> driver should be blacklisted on your system and not available to 
>>>> install with yum as it does not support your hardware, so simply doing:
>>>>
>>>> yum install kmod-nvidia
>>>>
>>>> should pull in the latest 367.57 driver automatically for you. In 
>>>> other words, by the magic of yum-plugin-nvidia, yum _should_ just do 
>>>> the right thing for you (please tell me this works for you!). And 
>>>> because future (version > 367) releases are blacklisted by yum, yum 
>>>> should never offer you updated driver versions that do not support 
>>>> your hardware.
>>>
>>> I have a local mirror, and it had some 375.xx versions before the 
>>> latest that were causing issues.  I removed them and it's 
>>> successfully installing the 367.57-3 version.
>>>
>>
>> Brilliant!
>>
>> Yes, the blacklist for K1/K2 devices no longer supported were added to 
>> later 375.xx series releases and earlier releases were removed from 
>> the repository for exactly this reason.
>>
>> Older versions are always available in the archive.
> 
> I greatly appreciate all of the work you put into maintaining this 
> stuff.  nvidia certainly doesn't make it easy.  I support approximately 
> 600 desktops, most having nvidia GPUs.  It would be much more difficult 
> without your tools.  I'm able to use nvidia-detect and puppet together 
> to make sure the correct nvidia packages are installed.
> 
> In the case of the K1 that does cause a slight issue.  nvidia-detect 
> returns a '6' code, which equates to the (currently) non-existent 367xx 
> packages.  After your information, in puppet I'm treating '6' and '1' 
> the same and installing kmod-nvidia.  But I'm not sure what the best 
> approach is.  I understand you don't want to return a '1' for the K1 as 
> it's not really using the 'current' driver.  But installing kmod-nvidia 
> is the correct solution for now.  But returning a package that doesn't 
> exist isn't ideal either.
> 
> I'm sorry I don't really have an answer (other than my own work-around), 
> but I wanted to make sure you're aware of the issue. Not sure there are 
> enough K1 cards out there to worry about.  I only have the one, so you 
> don't need to make sweeping changes on my account.
> 
>   - Steve