[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#998298: mesa-vulkan-drivers overrides nvidia driver



Hi,

I have found the same problem with my Intel chipset.

My laptop now implicitly switches to software rendering for Vulkan applications because VK_ICD_FILENAMES remains empty by default.

After setting:

export VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/intel_icd.x86_64.json:/usr/share/vulkan/icd.d/intel_icd.i686.json

the application now uses hardware rendering at full speed without tearing my CPU apart. I think the value for VK_ICD_FILENAMES should have a sane default on every system, depending on the ICD Debian wants to use for a particular chipset.

Tested on Debian Bullseye.

Best regards,
Jelle

On Tue, 2 Nov 2021 19:28:17 +0100 =?UTF-8?Q?Andrzej_Zi=c4=99ba?= <a-zieba@go2.pl> wrote:

> W dniu 02.11.2021 o 10:32, Michel Dänzer pisze:
> > On 2021-11-02 00:25, Andrzej Zięba wrote:
> >> Package: mesa-vulkan-drivers
> >> Version: 21.2.4-1
> >> Severity: important
> >> X-Debbugs-Cc: a-zieba@go2.pl
> >>
> >> Dear Maintainer,
> >> Upgrade of mesa-vulkan-drivers overrides the Nvidia Vulcan driver.
> >>
> >> * What led up to the situation?
> >> I have upgraded the package and noticed that Warthuner stopped working due to
> >> reported freeze.
> >>
> >> * What exactly did you do (or not do) that was effective (or ineffective)?
> >> WT uses Vulkan so I figured out that this package is probably to blame and
> >> indeed downgrading it was fixing the problem.
> >
> > Can you provide the output of vulkaninfo (without setting VK_ICD_FILENAMES) for both cases?
> Attached.
>
> >
> >
> >> To circumvent the problem with the mesa-vulkan-drivers upgraded I had to set
> >> VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/nvidia_icd.json
> >> in my Xession init scripts to force Nvidia Vulkan driver.
> >
> > It's rather doubtful that there's a bug here.
> >
> > Without setting VK_ICD_FILENAMES, there's no well-defined order in which Vulkan devices are enumerated. Most Vulkan applications just use the first enumerated device, so if the enumeration order changes, they'll end up using a different one.
> So we end up with random driver each time there is update?
> I think the distribution should have some mechanism to make sure that
> the expected driver is used, update-alternatives or something.
>
> Best Regards,
> AZ
>
-- 
---------------------------------------------------------------------------
             Jelle Haandrikman          System Tester
       mail: jhaand@freedom.nl          mobile: +31 (0)6 1918 0130
              twitter: @jhaand          website: http://jhaand.nl

Reply to: