On Fri, 7 Dec 2001, Alan Hourihane wrote:

> On Fri, Dec 07, 2001 at 05:27:06PM +0000, Simon Harvey wrote:
> > 
> > Having searched through tdfx_driver.c in CVS, and looked at the
> > voodoo3_spec pdf, I'm pretty sure the problem is due to the memory delay
> > settings.
> > 
> > Specifically Bit13 (sg_clk_nodelay) is clear, not set as in tdfx_driver,
> > and depending on the BIOS, various delays are imposed on reading from the
> > SGRAM (set by bits 19-16). The oflop_del_adj bits differ too. I've a
> > V3-2000PCI, BIOS 2.15.06, btw.
> > 
> Ooops. Noticed you've got a V3 and not a Banshee.
> 
> If you change that ifdef to PCI_CHIP_VOODOO3 does it help your problem ?
> 
> Alan.
Not sure where you mean. I've been looking at

http://cvsweb.xfree86.org/cvsweb/xc/programs/Xserver/hw/xfree86/drivers/tdfx/tdfx_driver.c?rev=1.86&content-type=text/x-cvsweb-markup

Around line 498 the DRAMINIT1 register is written to for anything other
than a Banshee. I guess I could change this and try again, but I've never
bothered a full compile of XFree86 on my Celeron; linux kernel takes long
enough.

I emailed Daryll Strauss too, the original author, and just got:

-- Quote --
The problem is that each vendor used different DRAM in their
boards. The actual 3dfx code, would take the value for these registers
from the Windows registry. So, every hardware vendor had different values
they determined and set from software. Leaving them entirely alone from
the BIOS doesn't work, because not all vendors set them correctly in the
BIOS. (That never worked on my cards). We tried picking the "least common
denominator" type of number, but even that wasn't successful because
setting it too high would break some boards and setting it too low others.

About the only option anyone could come up with was to have a database
of manufacturers and decide what values to use for each one. Mike Harris
(at Red Hat) started to do that by collecting feedback from users, but I
don't think he got much input and hasn't had time to implement what he
did get. It's just a mess.
-- EndQuote --

Do you know if this is still the case? If so I guess the only solution is
cooling...



_______________________________________________
Xpert mailing list
[EMAIL PROTECTED]
http://XFree86.Org/mailman/listinfo/xpert

Reply via email to