AMD deciding _now_ what to do about Linux

Dave Airlie airlied at
Fri Jun 15 23:25:12 PDT 2007

On 6/16/07, Chris Pemberton <cjpembo at> wrote:
> Dotan Cohen wrote:
> > I wrote to AMD regarding the state of ATI Linux drivers..
> >
> Perhaps this is a good time for me to ask a question:
> How many man hours would you estimate it would take for *experienced*
> developers, using mesa, dri, drm, agpgart etc to produce a driver that
> performs *better* than AMD's current drivers?  (Assuming AMD gives all
> the necessary resources).  I ask this because I would first like to
> understand the magnitude of what I'm asking for before I do the actual
> asking.  Could anyone here offer up a rough estimate?

I'm thinking you'd be up for 3-4 man years of effort do get a 3D
driver that might outperform fglrx on ATIs hardware, this is after all
ATI's hardware, there HW eng and SW eng teams have a lot of internal
interaction, a lot of time for non-ATI people would be spent waiting
for information flow and for ATI people would be waiting for legal
clearance for information release...

To be honest the Intel model has worked very well, they have also
managed to build up a sort of ecology around their driver where
improvements to it happen without their having directly funded them,
and in a lot of cases where they have provided direct funding and
manpower to work on the driver, however the driver remains the
"property" of the community, this was one problem ATI had with open
source before where they didn't really understand community and would
get annoyed when we made serious structural changes (like benh's
memmap patches...) to the driver, despite these changes making the
driver overall a lot more stable and useful across more platforms..

The thing is a 3D driver is a large beast, and I know Greg has people
lined up willing to work on drivers, but I don't see any of them
jumping into the current codebases adding features, things like TTM
and modesetting for radeon are using well known techniques on the card
that don't require specs to get a driver developed, they just require
a lot larger contribution of time than your average USB device.


More information about the xorg mailing list