ChrisK wrote on 2026-03-02, 13:37:

Have ordered the same ones for an insane price...
They will take some weeks to arrive.

To continue this task:
I've received the two replacement RAM ICs for the TI4200, along with a hugh heap of sh..t. I really wonder what people around the world are thinking about what they're doing, but on that one maybe later.
Just so much: The RAM for this card seems to be legit as far as I could tell.

So, I've replaced the two RAMs where MATS (v341 I think it was) was reporting errors before and now MATS is completely fine. Zero errors.
The card now shows a clear and error free picture under DOS as well as under Windows XP.
However, more testing showed there're still errors in 3D, in form of mostly black dots, more or less, all over the picture:

The question is: Does this still look like a RAM related issue and if so how could this be evaluated in more detail (again: MATS is OK but does it test all the RAM or only parts of it?). Or could this be a fault in the GPU itself?

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory errors and the pattern suggests one of the eight ×32-bit GDDR3 paths may be faulty.

I'm trying to find an ATI MEMID version compatible with R4xx/R480/R481. I have R3MEMID 1.07, but it rejects the R481. I found references to R3MEMID 1.13, apparently supporting R4xx/X800, but I cannot access the Chinafix download.

Do you happen to have R3MEMID 1.13, or know of another ATI memory diagnostic suitable for R480/R481/X850?

Thank you!

v_yordanov wrote on 2026-08-21, 21:33:

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory […]
Show full quote

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory errors and the pattern suggests one of the eight ×32-bit GDDR3 paths may be faulty.

I'm trying to find an ATI MEMID version compatible with R4xx/R480/R481. I have R3MEMID 1.07, but it rejects the R481. I found references to R3MEMID 1.13, apparently supporting R4xx/X800, but I cannot access the Chinafix download.

Do you happen to have R3MEMID 1.13, or know of another ATI memory diagnostic suitable for R480/R481/X850?

Thank you!

Hi again, I have made some progress with the X850 XT PE (R481).

I managed to obtain R3MEMID 1.13, which supports the card enough to run the memory tests. It detects PCI ID 0x4B4C and the full 256 MB VRAM.

I ran:

R3MEMID.EXE -NOCFG -LOG

The simplest Fill test gives:

failing bit : MDA0 0 1 2 3 4 5 6 7 ...

I also ran the complete test suite. MDA0 appears consistently in the failing results, although some of the more complex tests also report bits on MDB0/MDC0/MDD0.

I will attach the complete R3MEMID log to this post.

The card has 8 × Samsung K4J55323QF-GC16 GDDR3 chips, 256 MB total / 256-bit bus.

Does anyone know the memory-channel/bit mapping for R480/R481/X850? In particular, which physical VRAM IC corresponds to MDA0 0–... on this board?

I noticed in your Radeon 9700 repair reports that you were able to map results such as MDC0 0... or MDD0 32... to a specific physical memory IC using the schematic. Do you know if an X800/X850/R480/R481 schematic or memory-channel diagram exists that would let me do the same thing here?

Also, would the schematic be enough to trace the R3MEMID MDA/MDB/MDC/MDD data-bit groups to the individual GDDR3 chips, or is an R3MEMID board-specific CFG/mapping also required?

Thanks!

v_yordanov wrote on 2026-08-22, 13:57:

Hi again, I have made some progress with the X850 XT PE (R481). […]
Show full quote

v_yordanov wrote on 2026-08-21, 21:33:

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory […]
Show full quote

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory errors and the pattern suggests one of the eight ×32-bit GDDR3 paths may be faulty.

I'm trying to find an ATI MEMID version compatible with R4xx/R480/R481. I have R3MEMID 1.07, but it rejects the R481. I found references to R3MEMID 1.13, apparently supporting R4xx/X800, but I cannot access the Chinafix download.

Do you happen to have R3MEMID 1.13, or know of another ATI memory diagnostic suitable for R480/R481/X850?

Thank you!

Hi again, I have made some progress with the X850 XT PE (R481).

I managed to obtain R3MEMID 1.13, which supports the card enough to run the memory tests. It detects PCI ID 0x4B4C and the full 256 MB VRAM.

I ran:

R3MEMID.EXE -NOCFG -LOG

The simplest Fill test gives:

failing bit : MDA0 0 1 2 3 4 5 6 7 ...

I also ran the complete test suite. MDA0 appears consistently in the failing results, although some of the more complex tests also report bits on MDB0/MDC0/MDD0.

I will attach the complete R3MEMID log to this post.

The card has 8 × Samsung K4J55323QF-GC16 GDDR3 chips, 256 MB total / 256-bit bus.

Does anyone know the memory-channel/bit mapping for R480/R481/X850? In particular, which physical VRAM IC corresponds to MDA0 0–... on this board?

I noticed in your Radeon 9700 repair reports that you were able to map results such as MDC0 0... or MDD0 32... to a specific physical memory IC using the schematic. Do you know if an X800/X850/R480/R481 schematic or memory-channel diagram exists that would let me do the same thing here?

Also, would the schematic be enough to trace the R3MEMID MDA/MDB/MDC/MDD data-bit groups to the individual GDDR3 chips, or is an R3MEMID board-specific CFG/mapping also required?

Thanks!

Hey there, I'm currently out and can check some schematics on monday. I'll come back to you.

v_yordanov wrote on 2026-08-22, 13:57:

Hi again, I have made some progress with the X850 XT PE (R481). […]
Show full quote

v_yordanov wrote on 2026-08-21, 21:33:

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory […]
Show full quote

Hi @tehsiggi, I'm repairing a Radeon X850 XT PE AGP (R481) with severe VRAM artifacts. VMTCE shows millions of repeatable memory errors and the pattern suggests one of the eight ×32-bit GDDR3 paths may be faulty.

I'm trying to find an ATI MEMID version compatible with R4xx/R480/R481. I have R3MEMID 1.07, but it rejects the R481. I found references to R3MEMID 1.13, apparently supporting R4xx/X800, but I cannot access the Chinafix download.

Do you happen to have R3MEMID 1.13, or know of another ATI memory diagnostic suitable for R480/R481/X850?

Thank you!

Hi again, I have made some progress with the X850 XT PE (R481).

I managed to obtain R3MEMID 1.13, which supports the card enough to run the memory tests. It detects PCI ID 0x4B4C and the full 256 MB VRAM.

I ran:

R3MEMID.EXE -NOCFG -LOG

The simplest Fill test gives:

failing bit : MDA0 0 1 2 3 4 5 6 7 ...

I also ran the complete test suite. MDA0 appears consistently in the failing results, although some of the more complex tests also report bits on MDB0/MDC0/MDD0.

I will attach the complete R3MEMID log to this post.

The card has 8 × Samsung K4J55323QF-GC16 GDDR3 chips, 256 MB total / 256-bit bus.

Does anyone know the memory-channel/bit mapping for R480/R481/X850? In particular, which physical VRAM IC corresponds to MDA0 0–... on this board?

I noticed in your Radeon 9700 repair reports that you were able to map results such as MDC0 0... or MDD0 32... to a specific physical memory IC using the schematic. Do you know if an X800/X850/R480/R481 schematic or memory-channel diagram exists that would let me do the same thing here?

Also, would the schematic be enough to trace the R3MEMID MDA/MDB/MDC/MDD data-bit groups to the individual GDDR3 chips, or is an R3MEMID board-specific CFG/mapping also required?

Thanks!

Hey there, sorry for coming back to you that late. If it's still needed, I took two schematics and verified the memory channel + IC arrangement for R480 - it's the same as for R300 (which doesn't surprise me a lot).

Hey, no problem at all about the delay.

I think I can confirm what you posted. I also checked the R480 schematic, and it shows that MDA0 is connected to U52.

I’m attaching the schematic here, as it may also be useful to someone else working on an X850/R480/R481 card in the future.

I measured MVDDQ, MVDDC, VREF and VREF #2 around U52 and compared the readings with the known-good U54. Everything seems normal and the readings are very similar.

So at this point I suspect either U52 itself / its BGA connection, or the GPU BGA / MDA0 path on the GPU side. I’m still not sure which one is actually responsible.

I’m thinking of replacing/reballing U52 first, before doing anything to the GPU. Do you think that would be a reasonable first repair attempt?

Thanks again for the support!

P.S. The attached image shows the artifacts on the faulty card. The artifact pattern changes when I change the screen resolution.

v_yordanov wrote on 2026-09-03, 19:06:

Hey, no problem at all about the delay. […]
Show full quote

Hey, no problem at all about the delay.

I think I can confirm what you posted. I also checked the R480 schematic, and it shows that MDA0 is connected to U52.

I’m attaching the schematic here, as it may also be useful to someone else working on an X850/R480/R481 card in the future.

I measured MVDDQ, MVDDC, VREF and VREF #2 around U52 and compared the readings with the known-good U54. Everything seems normal and the readings are very similar.

So at this point I suspect either U52 itself / its BGA connection, or the GPU BGA / MDA0 path on the GPU side. I’m still not sure which one is actually responsible.

I’m thinking of replacing/reballing U52 first, before doing anything to the GPU. Do you think that would be a reasonable first repair attempt?

Thanks again for the support!

P.S. The attached image shows the artifacts on the faulty card. The artifact pattern changes when I change the screen resolution.

Any behavior change if you apply pressure on U52? I'd say your plan is good.

Did you verify the resistor arrays on the GPU<>vRAM traces? (if your card has them) They might just be cracked/broken.