Re: XFree86 (Knoppix) and NVidia 5200 Ultra...
On Thu, 11 Sep 2003, Daniel Stone wrote:
> (CCing debian-knoppix, debian-x, and Mark Vojkovitch).
>
> On Thu, Sep 11, 2003 at 09:15:42AM +0930, behoffski wrote:
> > I'm using the version of Debian XFree86 contained in Knoppix 5.2
> > (2007-07-26).
>
> Hmm, interesting. Knoppix guys: do you know about XFree86 4.3.0-0pre1v1
> in experimental, which is newer than my -0ds series? It's the
> team-maintained version, by myself, Branden Robinson, and ISHIKAWA
> Mutsumi.
>
> > I've used this package with four identical systems containing
> > NVidia FX 5200 Ultra cards (using boot parameter "xmodule=nv" to
> > avoid the problem that the card's ID isn't known). I've found
> > that the operation of XFree86 is unreliable -- ome machine fails
> > to start X at all -- others are flaky.
That's kindof vague. I've had no problem with the 5200 series
and XFree86 4.3.0, and don't know of any problems other than the
usual 4.3.0 nv driver confusion about which head it's supposed
to come up on (usually only a problem with DFP). Not starting
is usually a configuration issue. There were no outstanding
fatal issues with 4.3.0 except for a segfault on the old Riva128
when doing DDC detection.
> >
> > I've noticed that in xfree86.org's online CVS archive, there has
> > been a reworked nv module added in the last fortnight -- with
> > CVS comments indicating that "Some newer GeForceFX cards have a
> > new PLL design which needs to be programmed differently in order
> > to get a stable display." This comment appears to tie in very
> > neatly with my observations.
"Newer" meaning not released yet. The 5200 is what I would consider
"old" GeForceFX hardware. The PLL changes are not applicable.
> >
> > http://cvsweb.xfree86.org/cvsweb/xc/programs/Xserver/hw/xfree86/drivers/nv/
>
> Hmm, interesting. We've already backported the SiS and Savage drivers,
> so I don't see why we couldn't pull nv back as well. OTOH, we might want
> to wait for the SDK stuff - XSF?
>
> > My main interest in this is to have a Knoppix CD that works well
> > with the 5200 Ultra systems. So two questions:
> >
> > 1. If I remaster the existing Knoppix CD, and merely replace the
> > NV driver with an object compiled from the current CVS sources
> > (nv_drv.o), what is your estimate of the likeihood of success
> > (or, in other words, how compatible is this set of sources from
> > the XFree86 CVS archive with the XFree86-4.3.0-ds package contained
> > in Knoppix)?
>
> Umm, it *should* work, but no guarantees.
CVS drivers that use Xv will not work with 4.3.0 due to Marc La France
moving duplicated Xv clipping and data-copying code into the core
server to be shared by all. The CVS "nv" driver therefore calls Xv
helper functions which are not present in 4.3.0.
>
> > 2. Would it be feasible to fast-track these sources into upcoming
> > XFree86-4.3.0 Debian experimental packages -- or would this be
> > too much against the grain of Debian's emphasis on stable code?
>
> Given we've already dragged back the SiS and Savage stuff, I don't see
> why we couldn't drag the nVidia driver back, as ewll. Mark, how stable
> is the new nv code?
CVS has no bug fixes over 4.3.0 except for:
1) Better "which head do I use" detection.
2) Fix some fullscreen video overlay static using DFP on GeForceFX.
3) Fix a DPMS problem with some laptop flat panels (regression from 4.2)
4) Fix a segfault on Riva128 during DDC probing (regression from 4.2)
5) Video overlay works correctly in doublescan modes now.
6) Fix a problem misdetecting 256 Meg cards as 16 Meg cards.
About 5 weeks ago I COMPLETELY REWROTE all acceleration in the
driver and split Riva128 support out into a separate module. While
I'm not aware of any outstanding problems with this code, it hasn't
seen much testing beyond my own testing. The magnitude of the changes
was substantial and is therefore risky. Note also, that the driver is
now split over two modules. Improvements (not bugfixes) in CVS over 4.3.0
include:
1) GeForce4 and newer support scaling of small modes on a larger DFP instead
of just centering it.
2) Add interlaced mode support for GeForceFX.
3) Better 2D performance overall due to the rewrite, particularly on
GeForce3, GeForce4 Ti and GeForceFX processors. The rewrite was
done not so much for performance but to get a head start on the
major rewrite necessary for some future hardware which is dropping
some old features used by the 4.3.0 drivers.
4) A new XvImage adaptor for TNT and up which uses a YUV blits.
Mark.
Reply to: