Radeon 9000 Pro AGP Flashing

The ROM I’ve used sets the VRAM to 250MHz, and the VRAM is rated at 3.6ns/275MHz. So it’s already clocked conservatively. Don’t you think it’s more likely that one of the chips is bad?
I didn't know how your VRAM was rated sorry, if it is faster than needed it shouldn't be that.

Although my current video card project has some VRAM that is not performing at its rated speed.
 
I didn't know how your VRAM was rated sorry, if it is faster than needed it shouldn't be that.

Although my current video card project has some VRAM that is not performing at its rated speed.
I put the card back in my PC and restored the original BIOS - the same corruption is there, works fine at cold boot - symptoms appear after 20 mins and get progressively worse.

So it’s just a dud card.
 
The seller was nice – they refunded it and let me keep the card.

Screenshots attached. I wonder if it's more likely VRAM or the GPU.
 

Attachments

  • IMG_0873.JPG
    IMG_0873.JPG
    2.2 MB · Views: 8
  • IMG_0871.JPG
    IMG_0871.JPG
    4.7 MB · Views: 8
The seller was nice – they refunded it and let me keep the card.

Screenshots attached. I wonder if it's more likely VRAM or the GPU.
Time to start pinching components between your fingers and see which components change the artefacting. Try with both a static screen and some simple motion.
 
On my first attempt I got a screen full of garbage too. My card looks the same as yours except that it has a switch populated at SW1 next to the VGA connector. I see your card has the pads, but no switch populated. Anyway, referring to @Jacques photos above, I moved switch #2 to the ‘on’ position and then the card worked fine on the next boot attempt. I didn’t have to adjust the resistors under the heatsink. So you may want to look at bridging the top two pads of your SW1 footprint (and potentially putting back R227/R229).
Nice work! The switch thing is also mentioned at The Mac Elite. My PC Radeon 9000 Pro card might be a bit of an oddball. Here's its switch footprint under magnification:IMG_20260505_215606_378.jpg

As you can see, the top pads do not connect to anything! Even if one of those dual switches was populated, the top switch (#2) wouldn't do anything. The bottom switch looks like it would ground whatever that via connects to. I did try to short that with a wire before I moved the resistors, and it had no effect.

If you ever remove the heatsink on your card, I'd be very curious to see whether it already has R228 and R230 populated vs. R227 and R229 :)

Another thing to inspect is the backside of the board - these cards get shoved in drawers or boxes and damaged. Mine had a few resistors knocked off and a couple of VRAM pins bent/shorted together when I received it off eBay.
Great point. Lots of these cards, especially when sold as "untested" but sometimes even "pulled from a working environment" are e-waste and have physical damage from time spent in scrap bins.
 
I put the card back in my PC and restored the original BIOS - the same corruption is there, works fine at cold boot - symptoms appear after 20 mins and get progressively worse.

So it’s just a dud card.
Bummer. Since you have it in a PC, you might be able to boot into DOS and use R6XMEMID to narrow down which DRAM channel(s)/bit(s) are having errors. I'm not sure what the mapping of channels to physical chips is on these cards, but there's a Radeon 9200 schematic posted at VOGONS and the 9000 is really similar, so it might not be too hard to figure out.
 
Lots of Radeon 9000 news today--I got a package from Japan and this was inside:
IMG_20260505_185955_054.jpg
IMG_20260505_190009_325.jpg
I'm not sure whether this is an OEM or retail card but it has an Apple part number sticker with 630-4254, so I guess that means OEM? Would've come in an MDD then, I think. Anyway, it has the 3.3ns Hynix RAM chips. It worked well in my Quicksilver. I dumped the ROM and it's an earlier version than any of the others I have seen, 122. Clock speeds are the same as all the rest, 275MHz core, 250MHz RAM.

I flashed the ROM from this card onto my PC card with the same 3.3ns Hynix RAM, hoping that the ROM would have different timings for the 3.3ns chips. Nope. Same artifacts, same behavior as any of the other ROMs. Improves at higher clock speeds, 290MHz is almost clean but not quite. I guess there's just something wrong with this card then, but it's subtle and not an issue I've ever seen before. R6XMEMID gives it a clean bill of health.

Oh well, back in the box of misfit video cards for a while longer.

In case they're useful to anyone, I've attached the ROMs from three cards with known RAM vendors/speeds; the Hynix-33 is mine and the other two were graciously provided by @Phipli.
 

Attachments

Back
Top