S3 Virge GX 385 Video Noise Recap?

[ahyeadude]

So my S3 Virge GX 385 (4MB SGRAM) PCI has always had some shimmering (best way to describe how the pixels move) video signal noise especially at higher VESA resolutions. I confirmed it's not any other issue because my Cirrus Logic 5434 board works fine with absolutely no noise in the same machine. I am finally annoyed enough to do something, so I am thinking a recap might help?

Has anyone had success recapping an S3 card to reduce noise?

Just the electrolytics?

[rasz_pl]

probably badly designed filtering, back in the day people used to rip filters altogether from cheap S3 cards

[ahyeadude]

I don't even know where to start ripping filters out. I'm not having any luck on google either. Are there some basic guides to how these things were designed? Is it a power filtering issue or just a analog video signal filtering issue? Is the DAC integrated into the silicon or is it external on the board?

[rasz_pl]

analog vga signal, pi topology + loading resistors
http://www.calculatoredge.com/electroni ... 20pass.htm
I remember reading in the nineties on usenet about people just shorting inductors to bypass it and reporting "more vibrant colors"

[ahyeadude]

Thanks. I also picked up a Dell STB Nitro 3D 1.3 (S3 Virge/GX 385 4MB EDO) and this card has perfectly clean video output. So, I think I'll take a look at the STB output design and see if there is a way to replicate it on the cheapo Virge card I have.

[rasz_pl]

wouldnt expect any different form STB card
post pictures of analog sections if you need any help

[Tiido]

One Virge board needs a fix to produce non jittery image. I had two identical boards with the exception of the fix, I applied it to the other and jitter disappeared.

U3 pins 3+4 and 12+13 need to be shorted to fix the jitters.

Break the ferrite off the inductors and remove the little capacitors on RGB lines to reduce horizontal blur. It will not fix it but the image improves slightly.

[ahyeadude]

Sounds promising. I'm going to try to remove the inductors and capacitors on the output. I don't have an empty U3 pad, I do have a similar pad U15, but its populated with an IC (see pics). Before I go all medieval on it, here are some pictures of the cheapo Virge/GX with the jitter/shimmer issue.

https://imgur.com/a/E1DJ0Jz

[ahyeadude]

So I removed the SMD caps and inductors from the RGB lines. I then shorted where the inductors were. It's slightly better, but still get the shimmer/jitter. Especially when the CPU is loaded at high res (starting up a linux boot cd).

Any idea where on my board are the equivalent pin 3+4, 12+13? Also, what are those? I've got a datasheet for the Virge/GX chip, so I can trace it out on my board.

Also, looking at the STB, it looks like there are two inductor/capacitor circuits in series for each color channel.

[rasz_pl]

maybe you need to recap this card after all
+CPU affecting power plane can mean bad/tired main power supply in need of recap/replacement
do you have an oscilloscope to look at the 5V line(and 3.3V directly on the card) during all this?

Also, looking at the STB, it looks like there are two inductor/capacitor circuits in series for each color channel.

I dont see it, which model exactly are you looking at? I mean you could use higher order filter, but its not needed, its there to cut aliasing and EMI, I bet monitors have own filters anyway.

[Tiido]

I looked more closely at the board I have here and the modification involves only monitor ID lines, not actual H and V sync and it doesn't seem like it does anything beyond enabling DDC to be used... strangely without it, there were jitters in the image with that particular board.
Changing few of the capacitors in strategic areas might produce some positive result.
You could also try clearing out filter(s) on the horizontal sync line, in theory the sharper edges of the sync will produce more stable image...

[ahyeadude]

I've got an osilloscope on order. Nevermind, answered my own question, the metal tab on the 3.3V regulator is keyed to the middle pin, which is Vout.

[ahyeadude]

So maybe we are on to something here... The STB board is completely 5V and the cheapo board is 5V/3.3V (presumably for 3.3V SGRAM). Could the linear voltage regulator be picking up some noise from other components? See pics of the STB board below, you can also see the inductor/capacitor pairs for RGB...

https://imgur.com/a/9Dz5Tt2

[ahyeadude]

I am a total newb with a scope, but below is a video of startup probing the 3.3V regulator output on the cheapo Virge board. Looks noisy (+/- 0.5V) and the scope output corresponds to the noise on the screen. 5V off the PSU looks fine (sine wave +/- 4%).

Note, the grid lines are 0.5V.

https://youtu.be/dghRtgcj53k

[rasz_pl]

DSO Nano v3 - good enough for the task
Weird that 3.3 would jump almost all the way to 4V. H1084 is a cheap, probably much inferior clone of LM1084, might be worth replacing with genuine LM1084IT-3.3 along with capacitors.

[ahyeadude]

Going to replace the electrolytics and the linear regulator as suggested. Also going to test another PSU just to be sure. I'll report back.

[ahyeadude]

Okay, so tried the following...

- Replace PSU with another working AT PSU and also a new modern ATX PSU. Same issue.
- Replace linear regulator with new LM1084IT-3.3. Same issue.

Will recap the board tomorrow.

[ahyeadude]

Replaced all the electrolytics with new higher quality and it has improved the noise probably 70% or so. Interesting aside, they had all drifted higher, some 50%. I've read that is typical of using 16V caps on lower voltages.

I really think it is on the power filtering side. I'm going to try to figure out the regulator circuit design and see if it can be improved. Might get the remaining 30%.

[ahyeadude]

Taking a look at the design it has 3 internal power planes, one ground, one 3.3V, and one 5V. They seem to just strategically place caps at points around the board. However, this board had quite a few unpopulated capacitors. I think this was probably a cost savings measure and also that the board did not have the other 4MB of SGRAM populated. I poked around on the net for a similar board, and I think my board is a knock off of this Venus Virge GX board. All of the component labels and layout are almost identical.



The Venus Virge clearly has these populated, even in 4MB form. So, I just went ahead and populated out the remaining caps.



This has improved it nearly to 90% from the initial condition. There are still some unpopulated SMT caps, that might get the extra 10%. I might get around to adding those some day. Thanks for all the help, hopefully this helps someone with these knockoffs in the future.