Re: GPU driver blocklist things regarding NVIDIA

Anssi Hannula <[email protected]> writes:

> 10.04.2012 08:56, Eirik Byrkjeflot Anonsen kirjoitti:
>> Anssi Hannula <[email protected]> writes:
>> 
[...]
>> That does sound a bit sweeping.  If the intention was to say that the
>> NVS Quadros have poor performance, I'd be inclined to think that it
>> could possibly be true.  The NVS series seems to be designed for people
>> that have no use for graphics performance (e.g. the NVS 295 having about
>> half the performance of a GF 210).  But I would expect at least some of
>> the professional Quadros to have acceptable performance for our use.
>
> Yeah, most NVS Quadros seem to be based on the same chips as low-end
> GeForce cards, however, there seem to be one or two low-middle-end ones,
> like NVS 5100M (based on same core as GeForce GT (220, 230M, 240M, 325M,
> 330M), 315).
> Anyway, though, if the NVS cards are blacklisted, the respective GeForce
> devices should be blacklisted as well, to be consistent.

Yes, I think we probably need to have another look at that rule.

> However, the "Quadro FX" match:
>  "OpenGL-renderer": { "regex": "(GeForce [67]|Quadro FX).*" },
> seems wrong anyway, as Quadro FX are not limited to low end cards, but
> contain mid and high-end cards as well (like Quadro FX 5800 that has
> same core as GeForce GTX 260/275/280/285).
> I assume the above regex wants to only match against all Geforce 6+7
> family Quadro FXs - in that case, you can use the second number in the
> Quadro FX device name to detect the family, with some exceptions (untested):

Hard to say.  The only stanza I find with FX is one that disables
multisampling for some reason.  I wonder if the error here is rather
different: Maybe it is meant to match the GeForce 5 series, which nvidia
named "FX".  And the "Quadro" part is just a mistake due to the similar
naming.

[...]
>>> I also see that all versions of nouveau are blocked from 3d due to
>>> "crash when compiling some fragment shaders". Is there any additional
>>> information about that available (testcase or an upstream bugreport)?
>> 
>> I thought that applied to Mesa 7.10.2 and earlier.  I thought that was
>> fixed in Mesa 7.10.3.
>
>
> Well, all versions seem blocked in blacklist version 1.0.5:
>          "list entry": {
>                  "OpenGL-vendor": { "regex": "nouveau" },
>                  "OpenGL-version": { "regex": ".* Mesa .*" },
>                  "OpenGL-renderer": { "regex": "(Gallium|Mesa) .*" },
>                  "status2d": "supported",
>                  "status3d": "blocked driver version",
>                  "reason3d": "Nouveau drivers crash when compiling some
> fragment shaders."
>          },
>
> (as noted, I'm running Mesa 8.0.2)

That does look rather odd.  I can't immediately find the reason why this
was added.  But it seems to be rather old, so maybe it is the same bug
as the 7.10.2 issue, and we just haven't realized that this should be
removed.

We were supposed to document the reason for all of the blocklist
entries, but I worry that we may not have done that properly.

eirik
--
Opera-Linux: https://list.opera.com/mailman/listinfo/opera-linux
More lists: https://list.opera.com/mailman/listinfo/
Unsubscribe: mailto:[email protected]?subject=unsubscribe