View Single Post
Old 01-10-10, 05:07 PM   #1
hasfill
Registered User
 
Join Date: Apr 2009
Posts: 5
Default 195.30: same EDID interpreted differently on VGA and DVI connectors! Must be a bug?

Using NVIDIA-Linux-x86_64-195.30 and a Dell SP2309W monitor (native resolution 2048x1152),
the same EDID data is read (up to 4 bytes):-

hexdump of edid.bin when using VGA cable:
--------------------------------------------------------

0000000 ff00 ffff ffff 00ff ac10 d01b 5253 3044
0000010 132c 0301 330e 781d 18ea a955 3753 25ad
0000020 5013 a554 004b 4f71 8081 00b3 0101 0101
0000030 0101 0101 0101 3d3b a000 8080 4021 2030
0000040 0035 22fe 0011 1a00 0000 ff00 5400 3439
0000050 4630 4139 3054 5244 0a53 0000 fd00 3800
0000060 1e55 145c 0a00 2020 2020 2020 0000 fc00
0000070 4400 4c45 204c 5053 3332 3930 0a57 4b00
0000080


hexdump of edid.bin when using DVI cable:-
-----------------------------------------------------


0000000 ff00 ffff ffff 00ff ac10 d01c 5253 3044
0000010 132c 0301 3380 781d 18ea a955 3753 25ad
0000020 5013 a554 004b 4f71 8081 00b3 0101 0101
0000030 0101 0101 0101 3d3b a000 8080 4021 2030
0000040 0035 22fe 0011 1a00 0000 ff00 5400 3439
0000050 4630 4139 3054 5244 0a53 0000 fd00 3800
0000060 1e55 115c 0a00 2020 2020 2020 0000 fc00
0000070 4400 4c45 204c 5053 3332 3930 0a57 db00
0000080


$ diff edid.bin_vga_hexdump edid.bin_DVI_hexdump
1,2c1,2
< 0000000 ff00 ffff ffff 00ff ac10 d01b 5253 3044
< 0000010 132c 0301 330e 781d 18ea a955 3753 25ad
---
> 0000000 ff00 ffff ffff 00ff ac10 d01c 5253 3044
> 0000010 132c 0301 3380 781d 18ea a955 3753 25ad
7,8c7,8
< 0000060 1e55 145c 0a00 2020 2020 2020 0000 fc00
< 0000070 4400 4c45 204c 5053 3332 3930 0a57 4b00
---
> 0000060 1e55 115c 0a00 2020 2020 2020 0000 fc00
> 0000070 4400 4c45 204c 5053 3332 3930 0a57 db00

------------------------------------------------------------------
$ cmp -l edid.bin_vga edid.bin_DVI
11 33 34
21 16 200
100 24 21
128 113 333

--------------------------------------------------------------------------
After much experimentation, I found that

Option "ModeValidation" "noMaxPClkCheck"

allowed the monitor to use its native 2048x1152:
"2048x1152"x0.0 156.75 2048 2096 2128 2208 1152 1155 1160 1185 +hsync -vsync (71.0 kHz)


The problem seems to be that the DVI EDID reports a maximum pixel clock
of 155MHz:

(--) Jan 09 20:40:08 NVIDIA(0): DELL SP2309W (DFP-0): 155.0 MHz maximum pixel clock
(--) Jan 09 20:40:08 NVIDIA(0): DELL SP2309W (DFP-0): Internal Single Link TMDS


but 2048x1152 requires 156.75MHz.

I found a suggestion that this was a monitor firmware bug in a report on a similar
problem with a Samsung Syncmaster:
Manufacturer: SAM Model: 21e Serial#: xxxxxxxxxx

That report was more than two years ago, so it seems that there are various monitor models
with the same problem.

Can someone confirm that this really is a bug rather than a valid variation/extension of the EDID
specification? If a bug, cannot nvidia get the monitor manufacturers to fix their firmware? Or should we blacklist Samsung and Dell monitors for use with "other" operating systems?

I can try to get a firmware upgrade from Dell myself, but need to confirm that the DVI EDID
block above (or is it a separate TMDS encoder bit rate) is actually wrong.

There is more information towards the end of http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=498632

Help please!

ael
hasfill is offline   Reply With Quote