• His Excalibur (Platinum)Radeon 9600 Dual-DVI 256MB Videocard with AGP slot. Other spelling variants of the same card are: HIS Excalibur Radeon 9600 Dual DVI, HIS 9600 Dual DVI
• Different versions:
  • Active cooling: http://www.revioo.com/articles/a4179_0.html (Cooling is only present on prototype/pre-series production models)¹⁾
  • Passive cooling with a Silicon Bridge DVI compliant transmitter: http://www.ati-news.de/HTML/Berichte/Enmic/R9600-Dual-DVI/His-R9600-Dual-DVI-Seite1.shtml ²⁾
  • Passive cooling with a THine Electronics DVI compliant transmitter ic from THine Electronics 'THC63DV164'³⁾

• Offical manufacturers webpage from Hightech.com.hk
• General info

• Flickering horizontal lines on second head of a dual-DVI system
• Last post: strong artifacts in 640x480 mode
• [radeon] can't do dual-link tmds, or second tmds transmitter …I have a picture over the DVI now, but it flickers pretty badly…
• ...The LCD still displays the vertical lines but I'll assume that goes away once loaded into and OS...

I found this problem is not only ATI/AMD related, but NVidia chipsets also seem to suffer from this problem, according to a google search with the following terms: “dvi output” “horizontal lines” “dual dvi”

• Gainward Ultra/780 XP with NVidia FX 5600 running both outputs at 1600x1200
• GeForceFX 5600, Second DVI shows gray lines

The artifacts with the HIS videocard with RV350 chipset tested with Ubuntu 8.04 are worse than under windows. While playing a videofile on the first screen, the 2nd screen is showing artifacts quite heavily.

Initially I suspected glitches were caused by jitter on data from video memory. At resolutions below 1600×1200 the video was fine on both outputs. The glitches would only manifest at high resolution above 1600×1200 pixels. So memory timing might be more critical. If that would be the case, cleaner power may solve the problem. So I added extra 100nF decoupling capacitors on top of 4 of the 8 memory chips.

As decoupling did not change anything at all, I had to look further. A second experiment would be slightly increasing the memory voltage, from 2.50 Volt to 2.55 Volt, by adding a resistor divider at a LM431 reference voltage ic (SOT-23). The voltage increase did not improve anything either.

Because from the two DVI outputs, only the second one had problems, it crossed my mind that somehow a memory timing issue would not be a logical explanation. If it would be, then likely both outputs would suffer and not only one of them. Due to the differences in the DVI output circuitry, it made even more sense that the problem would be in something specific one of the circuits had which the other didn't have. So I started to measure power supplies for the DVI transmitter, a THine THC63DV164. At the closest capacitor (C1421), I measured 3.3V with about 200mV noise, which was far too much in my opinion.

The 3.3V power supply to the DVI transmitter suffered from subtantial noise, coming from the motherboard. The 47uF capacitor C1421, was unable to block this noise. A solution I found was just increasing the capacity. After trying out several type of capacitors, with different capacities, it seemed that 470uF would solve it. It's critical that this should be a low-esr type. Smaller values and other kind of capacitors may not filter the power supply enough. The easiest replacement capacitor I could find and would fit as well, was a 1000uF 6.3V capacitor from an older motherboard. After the replacement all troubles with glitches were solved. Both in Linux as in Windows. The noise I measured was now less then 10mV. The voltage across the capacitor is only 3.3V, so most capacitors will be fine. You may even improve it slightly, by adding a small smd ceramic capacitor (at the bottom) of about 100nF.