Could the ram on the card be bad?

talk about scam, just wow:) At least it was directed at Taiwanese domestic market.
Is the ram daughterboard removeable or soldered permanently?

rasz_pl wrote on 2022-11-03, 06:53:

Is the ram daughterboard removeable or soldered permanently?

Good question, but probably irrelevant to this problem. The Virge/VX uses VRAM for the primary frame-buffer memory, and DRAM for additional texture memory. Unless you are using 3D software, the extra DRAM is not used. The memory on the add-on card is (EDO)-DRAM without VRAM features. If the problem is a memory problem, it would be related to the VRAM that's soldered on-board.

By the way: Virge/VX cards are known to be sensitive on their clocks. Don't try to run a Virge/VX at PCI clocks above 33MHz, and don't try to overclock the core. If you want to toy around with overclocking, get a Virge/DX instead.

You could try underclocking the RAM or the GPU core.. Maybe the artifacts will go away then.
The ViRGE has that capability. There are S3 utilities for DOS/Windows 9x which can do that easily.

Edit: https://www.vintage3d.org/virge.php 😀

rasz_pl wrote on 2022-11-03, 06:53:

talk about scam, just wow:) At least it was directed at Taiwanese domestic market.
Is the ram daughterboard removeable or soldered permanently?

Its removable , it has three connectors that it sockets into

mkarcher wrote on 2022-11-03, 08:01:

rasz_pl wrote on 2022-11-03, 06:53:

Is the ram daughterboard removeable or soldered permanently?

Good question, but probably irrelevant to this problem. The Virge/VX uses VRAM for the primary frame-buffer memory, and DRAM for additional texture memory. Unless you are using 3D software, the extra DRAM is not used. The memory on the add-on card is (EDO)-DRAM without VRAM features. If the problem is a memory problem, it would be related to the VRAM that's soldered on-board.

By the way: Virge/VX cards are known to be sensitive on their clocks. Don't try to run a Virge/VX at PCI clocks above 33MHz, and don't try to overclock the core. If you want to toy around with overclocking, get a Virge/DX instead.

Well I'm using the card stock I haven't touched any overclocking stuff

Jo22 wrote on 2022-11-03, 08:02:

You could try underclocking the RAM or the GPU core.. Maybe the artifacts will go away then.
The ViRGE has that capability. There are S3 utilities for DOS/Windows 9x which can do that easily.

Edit: https://www.vintage3d.org/virge.php 😀

I guess ill give that a go then

underclocking is a good idea
next is magnification, needle and going over all ram pins on the main pcb trying to gently move them one by one

afaik the way FMV was being encoded in DOS is mostly compression methods that only encode difference from previous image, so running FMV will only write to VIDEO RAM cells that need changing. Running something like duke3D or DOOM you are constantly blasting full framebuffer rewrites. mkarcher says this uses VRAM so mere act of scanning out video to screen wouldnt trigger memory refresh. Maybe your card has something wrong with memory refresh?

rasz_pl wrote on 2022-11-03, 11:36:

mkarcher says this uses VRAM so mere act of scanning out video to screen wouldnt trigger memory refresh.

Scanning out does trigger refresh even on VRAM. Every kind of read from the dynamic memory cells to the row buffer has to refresh them. It doesn't matter whether we just pick a single 16-bit word from the row buffer (those are x16 chips) in a /RAS+/CAS cycle, we pick a multitude of 16-bit-words from the row buffer using FPM access with /CAS cycles while /RAS is active all the time, or we copy the data from the row buffer into the "serial access memory", a shift register of 512 16-bit-words used for scan-out. Many people are mislead by the (true) fact that scanout of VRAM has no noticable effect on drawing bandwidth and expect that scanning out does not interact with the dynamic RAM part at all. In fact, even VRAM requires recurring cycles to reload the serial access memory, often called "transfer cycles". These transfer cycles are made from the primary port of the VRAM that also is used for drawing and random CPU reads of the framebuffer. The big win is that we just need one cycle to transfer 512 words (or 256 words if we use split transfers that just reload half of the serial access memory), instead of one cycle per word (as we would need with DRAM).

damn /slaps forehead, what was I thinking 😀 I somehow quickly brainfarted into imagining serial buffer being so large it can hold whole framebuffer 😀
normal ram needs refreshing every ~60ms, VGA refresh is <16ms, at least framebuffer ram is being refreshed no matter what.
There are white dots, there are black dots, some in regular pattern like gaps. It does look like one of the ram chips is flaky.

Okay so the duplicate card came in today and it has no artifacts , I guess there was something screwy with the original boxed card.

It sees the card as a 4mb card in windows , is that normal? like with say the stb velocity 3D?, does that show up as a 4mb or a 8mb card?

I also tried running croc at 1024x768 but it wouldn't let me run at that res for some reason after selecting the option in the menu,
even though I attached the additional 4mb ram board.

Maybe the addon ram is also faulty?

AppleSauce wrote on 2022-11-03, 04:53:

Could the ram on the card be bad?

Most likely soldering issue somewhere.

It sees the card as a 4mb card in windows , is that normal?

No, the module is not working. Probably loose connection between pins or oxidation, try to reinstall it.

The Serpent Rider wrote on 2022-12-03, 11:53:

Most likely soldering issue somewhere. […]
Show full quote

AppleSauce wrote on 2022-11-03, 04:53:

Could the ram on the card be bad?

Most likely soldering issue somewhere.

It sees the card as a 4mb card in windows , is that normal?

No, the module is not working. Probably loose connection between pins or oxidation, try to reinstall it.

Okay so after pulling the card out and reseating the module the computer detected it (I did notice a piece of debris on one of the ram chips legs , not sure if that had something to do with it?)

Probably should also mention that I had to bend some pins back into place to get the module on in the first place.

It did run in 1024x768 but there's some weird artifacts only at that res which leads me to believe there might have been something that happened to the original card that the add on board came from(maybe the owner fried it?)

You can see at 800x600 there are no artifacts

Bit of a bummer , but I guess not the biggest deal in the world the game runs like a slide show at 1024 anyway.

AppleSauce wrote on 2022-12-03, 14:35:

It did run in 1024x768 but there's some weird artifacts only at that res which leads me to believe there might have been something that happened to the original card that the add on board came from(maybe the owner fried it?)

The symptom clearly looks like texture memory failure.This means that either some memory chip on the add-on board is bad, some trace on the add-on board is broken or the add-on board still doesn't make good contact to the base board. You only get the failure at 1024x768, because at lower resolution the on-card memory is sufficient for both the framebuffer and the textures, and only at 1024x768 the add-on board gets used for texture memory.

As the symptom seems to be consistent over the whole memory, I consider a broken trace, solder joint or bad contact more likely than a broken RAM chip.

Well guess what , I ended up removing the ram board bending some pins a bit more and reseating the board for a third time and now it works all of a sudden.

I don't know if it was oxidisation or the pins not being quite aligned but at least it works now.

After using the card again its now doing this , hopefully its not the monitor itself but it looks like it can't properly draw the image , it goes away after a while so maybe its the caps?

AppleSauce wrote on 2022-12-17, 03:36:

it goes away after a while so maybe its the caps?

"goes away after a while" indeed makes caps in switch-mode voltage regulators suspect. Those caps perform better when they are warmer, so when they are marginal, the fail to work properly when cold, but work good enough after warmup. On the other hand, you are using a ViRGE card. The ViRGE/VX chip runs on plain 5V with no switch-mode regulators, so I don't expect the caps on the card to be the culprit, although it's not impossible that the 5V supply is indeed not stable enough.

I'm more inclined to suspect the PC power supply and the monitor power supply, both of which contain switch-mode regulators.

mkarcher wrote on 2022-12-18, 14:31:

AppleSauce wrote on 2022-12-17, 03:36:

it goes away after a while so maybe its the caps?

"goes away after a while" indeed makes caps in switch-mode voltage regulators suspect. Those caps perform better when they are warmer, so when they are marginal, the fail to work properly when cold, but work good enough after warmup. On the other hand, you are using a ViRGE card. The ViRGE/VX chip runs on plain 5V with no switch-mode regulators, so I don't expect the caps on the card to be the culprit, although it's not impossible that the 5V supply is indeed not stable enough.

I'm more inclined to suspect the PC power supply and the monitor power supply, both of which contain switch-mode regulators.

Well I tried the monitor with my other pc and it had no issue , also a swapped the gpu for another one and the problem went away.
After inspecting the gpu I did notice that it had a LM317 voltage regulator on it in a TO92 package , I could try swapping it from the other duplicate card with bad ram and see if that fixes it.

AppleSauce wrote on 2023-01-03, 16:54:

Well I tried the monitor with my other pc and it had no issue , also a swapped the gpu for another one and the problem went away.

This likely rules out the monitor, but does not yet rule out the PC power supply.

AppleSauce wrote on 2023-01-03, 16:54:

After inspecting the gpu I did notice that it had a LM317 voltage regulator on it in a TO92 package , I could try swapping it from the other duplicate card with bad ram and see if that fixes it.

It's worth a shot. Also, when you are swapping parts, you might also try whether swapping C13 and/or C31 (the two electrolytic caps near the LM317) helps with that problem.