8500 is not FCBGA. Could be caps or dead RAM chip.

The 8500 core? Sure it is.

You can check the memory directly with ATIs software for that Re: VGA Repair report collection

Make sure there's no thermal solution (paste) spread over tiny SMD elements around the chip if there are, if so get something soft like qtips, placing one from above and remove it very gently by spinning it around an SMD.

Thank you so much for your help.

Regarding the ATI testing tools, R6XMEMID should be suitable for my 8500LE isn't it ?
In the topic you mentioned, it deals with freeDOS to run it but could it work with "simple DOS" from a win98 boot disk ?

Cheers,
JB

bimole wrote on 2025-10-20, 19:45:

My 8500LE has 4ns Hynix TSOP RAM chips. Pushing on the RAM chips and/or on the corners of the GPU does not change anything on ar […]
Show full quote

My 8500LE has 4ns Hynix TSOP RAM chips.
Pushing on the RAM chips and/or on the corners of the GPU does not change anything on artifacts.
I tested the card with VMTCE, which reports a lot of memory errors.
I resoldered all the RAM chips leads with flux without any success, artifacts are still there.

TSOP RAM chips rarely fail, at least from that era.
Resoldering them is usually more likely to damage something than to help, so best left for last resort. All moot now since you already did it, but just to keep it in mind for the future.

bimole wrote on 2025-10-20, 19:45:

When I got the card I noticed that the heatsink fan was almost blocked. Could it be possible that a bad cooling could have damaged the core or BGA with overheating ?

Yes, it's possible if the GPU overheated badly that something could have been damaged in it.
All in all, wire-bonded chips tend to be very tough, but even they can fail if ran too hot for too long.
That said, if the fan was not stuck, I'd say the GPU probably never overheated that badly.

bimole wrote on 2025-10-20, 19:45:

Is this Radeon generation prone to BGA failures ?

The "Radeon generation" is quite a broad term here.
Generally speaking, no - at least on ATI-built cards. With 3rd party OEMs, that's not always certain, though.

Radeon 8500 LE is 2nd gen Radeon (R200 series -based) and pretty tough for the most part.

bimole wrote on 2025-10-20, 19:45:

What about the VRM? Should I also check the electrolytic capacitors ?

Who's the builder of your card? Is it an ATI-made card or some 3rd party? The former (made by ATI) should have some pretty high quality caps on there, so extremely unlikely for those to be the issue. But if it's some 3rd party -made card, YMMV with the quality of the caps. Perhaps show us a picture if it's a 3rd party-made card.

In any case, it won't be the VRMs - those either work or they don't on older cards. If not working, your card won't boot.

I would say start by removing the card and inspecting it very carefully under good light. Maybe post some high-quality pictures of both sides of the card here. If you're not the original owner and you got it used, there's always a chance someone before you mis-handled it and possibly broke some small SMD components on the back of the card. Look for chipped/cracked/missing termination resistors around the RAM chips, as these are the most likely to cause artifacts.

bimole wrote on 2025-10-21, 19:28:

Thank you so much for your help. […]
Show full quote

Thank you so much for your help.

Regarding the ATI testing tools, R6XMEMID should be suitable for my 8500LE isn't it ?
In the topic you mentioned, it deals with freeDOS to run it but could it work with "simple DOS" from a win98 boot disk ?

Cheers,
JB

Yeah, you should be able to run it under regular DOS just fine. DOS4GW is needed iirc.

I don't have a logical explanation, but I have a 8500LE with 3.3ns memories that was a mess, any small frequency change would artifact. I flashed the VBIOS of a retail 8500 and now the card is great, even at 275/275.
Not sure if yours can do 275 on the memory, but I would try a different VBIOS to check if gets more stable.

Hi,

Here are some news.
Tests with R6XMEMID (launched from win98 bootdisk with r6xmemid.exe -nocfg -nordfload, only 6 first tests done) confirm memory issues :

Always with "MDB0". I guess it is a memory chip index ? But which one?! Or maybe a bad resistor array ?
Numbers 24, 25 ,26, etc... correspond to address lines right ?

Here are two pictures of the card (I removed the fan) :

Your help is very appreciated 😉

JB

If I compare the memory layout of the R200 used on the Radeon 8500 with the RV280 (used on 9200, for which we've got schematics), I'd guess the MDB will be the top four chips on front and bottom. (Not the ones next to the GPU + AGP slot.

MDB0 means, memory channel B, rank 0 - this is to be expected, the cards R200 has two memory channels, A and B. Each is 64 Bits wide. Per memory channel there are 4 Chips with 16 bits each. Now don't be confused why errors start with bit 24 and not 17 or 32 (like a multiple of 16) - it's not uncommon to go with a wild mix of data lines on the memory chips, for layout reasons.

The bits tell you which data lines are affected. You'll have to map the data lines via the schematic to the corresponding chip.

See this 9200 example (probably different to the 8500, as the R200 and RV280 are completely different chips layout wise):

I'll go and see if I can figure out more with the 9100s I got laying around for you, but I'd say your best bet is to check these ICs and their surroundings.

Edit: Just looked at one picture.. is that dirt?