MattRocks wrote on 2026-05-10, 14:17:

... It was after WinXP had logged on, recognised that the VGA card had changed, completed device detection, and was applying automatically selected drivers - the moment when you'd expect the resolution to successfully change is the moment the screen went blank and keyboard lights went out. Total hard crash. On the next POST the artefacts were present.

It feels like something shorted in the graphics card.

Yes, does feel more and more like bad caps.

At least with artifacts now "most" of the card is working. Have you pulled the caps and seen if any are shorted and if they measure close to their marked values?

Could also be a bad ROM or RAM ... Are the artifacts consistent? ... if they come/go/move it's more likely noise/bad caps, if consistent "bad spots" I'd be more suspicious of memory.

But given it's age, I'd definitely check caps first.

tehsiggi wrote on 2026-05-10, 14:26:

That's pretty normal behavior for having a memory faulty GPU loading the driver and trying to actually access / use the affected memory banks more intense.

I "mostly" agree, except... in my experience, "hard crashes" due to bad memory don't usually come back to normal operation with just "artifacts"... but you never know how the GPU is testing/allocating it's memory at startup.

Also, if the artifacts weren't visible till after drivers load .. text memory used during POST was probably good. Now artifacts appear during POST, the basic text memory is affected.
Always possible that things got significantly worse quickly, but IME - this is more common with CAPs.

Like all "old stuff", it really comes down to a debugging exercise!

DaveDDS wrote on 2026-05-10, 14:43:

I "mostly" agree, except... in my experience, "hard crashes" due to bad memory don't usually come back to normal operation with […]
Show full quote

tehsiggi wrote on 2026-05-10, 14:26:

That's pretty normal behavior for having a memory faulty GPU loading the driver and trying to actually access / use the affected memory banks more intense.

I "mostly" agree, except... in my experience, "hard crashes" due to bad memory don't usually come back to normal operation with just "artifacts"... but you never know how the GPU is testing/allocating it's memory at startup.

Also, if the artifacts weren't visible till after drivers load .. text memory used during POST was probably good. Now artifacts appear during POST, the basic text memory is affected.
Always possible that things got significantly worse quickly, but IME - this is more common with CAPs.

Like all "old stuff", it really comes down to a debugging exercise!

I had one and the same card with faulty memory appear nearly perfect on one start and on another day with the same system having vertical stripes. No caps involved, just broken memory. Of all the 9500s, 9700s and 9800s the only caps I had to change were ones that were physically damaged.

---

I would not go overly academic on this issue here now. If you've removed the fan and haven't changed the compound that was one there so far, just put a little paste onto it as well to fill in the gaps for now. For testing, this is sufficient. The card will not magically burn out because it's 10°C warmer in idle just because the thermal compound now sucks.