Video memory reported by GLX_MESA_query_renderer differs between amdgpu and radeon

[Kai Wasserbäch]

[Please CC me on all replies, I'm not subscribed to this list.]

Dear Alex,
Alex Deucher wrote on 10.12.2016 22:40:
> On Fri, Dec 9, 2016 at 3:27 PM, Kai Wasserbäch
> <kai at dev.carbon-project.org> wrote:
>> [Please CC me on all replies, I'm not subscribed to this list.]
>>
>> Hey,
>> I've been noticing that GLX_MESA_query_renderer reports less video memory when I
>> use amdgpu for my Hawaii PRO GPU, than when I use radeon. Since I'm pretty happy
>> with amdgpu (as I do not use HDMI audio) and it's the code, that gets active
>> improvements, I'd like to stay with it, but having 40 MB of video memory
>> "vanish" is not nice.
>> With radeon GLX_MESA_query_renderer reports 4096 MB RAM (as expected, the R9 290
>> was sold/advertised with that), while I'm seeing 4056 MB with amdgpu (last
>> tested kernel: 4.8.13).
>>
>> As I'm not sure, whether this is a bug or rather some always reserved memory
>> area, which gets subtracted before reporting the available memory, I didn't file
>> a bug report and rather am writing this e-mail. If it's a bug I'm happy to file
>> the report against amdgpu.
> 
> IIRC, the driver interfaces used to get the information exposed via
> that extension are slightly different.  radeon exposes the total
> amount of vram while amdgpu reports the total amount of free vram.
> The actual amount of free vram is probably pretty close.

thanks for the explanation!

But I'd like to point out that some of the automatic configuration tools shipped
with games (and probably other programs) are deciding upon the reported values
to which quality level they should default. Maybe reporting the theoretical VRAM
might be better (you could always report the actual values through sysfs or some
other channel)?

Cheers,
Kai

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 659 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161210/71ce27d2/attachment.sig>