Gotta drive back home first, but here are some first inputs:

- What message do you get from R3MEMID? How do you run it?
- The darker pin is iirc 5V, used for the memory VRM. Try to clean it up very well. If the 5V rail is not powerful enough / voltage is too low, a 9700 will spit out artifacts like there is no tomorrow. So make sure to have that one clean.

I assume the missing caps from the picture are all in place again, right?

Edit:
There is one missing cap (bulk one, probably 22µF, would have to check the schematic) below U53 on the back of the card. Not sure if you've replaced that one as well. That's the first thing I've seen.
The crazy garbled video output is something I've seen when testing a 9700Pro with bad external power connectivity / too low 5V rail. That causes the memory voltages to become unstable.

tehsiggi wrote on 2025-09-23, 06:50:

Gotta drive back home first, but here are some first inputs: […]
Show full quote

Gotta drive back home first, but here are some first inputs:

- What message do you get from R3MEMID? How do you run it?
- The darker pin is iirc 5V, used for the memory VRM. Try to clean it up very well. If the 5V rail is not powerful enough / voltage is too low, a 9700 will spit out artefacts like there is no tomorrow. So make sure to have that one clean.

I assume the missing caps from the picture are all in place again, right?

tehsiggi, I was really hoping you would see this post 😉

Here is R3MEMID

I cleaned that pin as much as I could (still looks like it was shorted / black color). I replaced the PSU (and so the 4 pin connector) but still the same issue.

Yes, the caps are restored in the picture.

tehsiggi wrote on 2025-09-23, 06:50:

Edit:
There is one missing cap (bulk one, probably 22µF, would have to check the schematic) below U53 on the back of the card. Not sure if you've replaced that one as well. That's the first thing I've seen.
The crazy garbled video output is something I've seen when testing a 9700Pro with bad external power connectivity / too low 5V rail. That causes the memory voltages to become unstable.

Added the quote since your post and the edit came at the same time 😉

How do you run R3MEMID with the garbled output? Did you use a second video adapter? R3MEMID only works when using the Radeon as the primary video card.

tehsiggi wrote on 2025-09-23, 07:13:

How do you run R3MEMID with the garbled output? Did you use a second video adapter? R3MEMID only works when using the Radeon as the primary video card.

Oh no, well that's the issue. I'm running a PCI card and assumed R3MEMID would "look" for the ATI. Arrrg! That explains why other software detects the card.

fyi, I have these waiting to go...but want to confirm it's 100% memory issue first.

Well, you get the card "running" right? Even with bad output? System booting and "doing" stuff?

Then I'd suggest the following: Get the boot media (DOS) ready so that it directly executes R3MEMID.EXE -LOG -GENREF -NOCFG. Then you'll have to blindly press "1" to run the first test. I think it even supports some switches to run the tests directly. I'll check for you when I'm back home.

You'll get a R3MEMID.LOG file written that you can look at afterwards. Have done similar things in the past.

Any of the memory modules pressure sensitive (apply some good amount of force by pinching the memory front/back with your thumb and index finger) ?

tehsiggi wrote on 2025-09-23, 07:18:

Well, you get the card "running" right? Even with bad output? System booting and "doing" stuff? […]
Show full quote

Well, you get the card "running" right? Even with bad output? System booting and "doing" stuff?

Then I'd suggest the following: Get the boot media (DOS) ready so that it directly executes R3MEMID.EXE -LOG -GENREF -NOCFG. Then you'll have to blindly press "1" to run the first test. I think it even supports some switches to run the tests directly. I'll check for you when I'm back home.

You'll get a R3MEMID.LOG file written that you can look at afterwards. Have done similar things in the past.

Any of the memory modules pressure sensitive (apply some good amount of force by pinching the memory front/back with your thumb and index finger) ?

Bingo! Log file is empty but I did redirect the output to mylog.txt. Here it is. There was nothing else after "Error ID 0VB001"

1R3MEMID version 1.07, (c) Copyright ATI Technologies Inc, 2003
2
3Log file generation enabled to .\R3MEMID.LOG ...
4Reference data file (RDF) loading disabled ...
5nbankbit = 2
6nrowbit  = 12
7ncolbit  = 8
8nchbit   = 2
9rankbit  = 0
10mc_cntl  = 00000022
11memsize  = 08000000
12calculated memsize = 08000000
13R3MEMID Test Program - version 1.07 - 9 tests
14==============================================================================
15[1  ] - Fill                            [6  ] - Macro + Micro8x2 tiling read   
16[2  ] - Data line toggle (GUI)          [7  ] - Byte swap macro+micro8x2 read  
17[3  ] - RW page (GUI)                   [8  ] - HOST_DATA/<f,b>/M2L            
18[4  ] - RW channel (GUI)                [9  ] - Line slopes/dir/P/solid        
19[5  ] - Data mask                       
20
21[* ] - Run all tests
22[p ] - previous page                    [s ] - start page
23[n ] - next page                        [e ] - end page
24[0 ] - Exit
25
26Enter selection: [1  ] Fill                                                : FAIL
27    Error ID 0VB001

This time I ran "*" All tests. log.txt is still empty but it generated a bunch of other files until my SDCard ran out of disk space.

fyi, I also have a copy of R6MEMID.

I had occasions where the log would not be written if I did not properly "quit" the program.

I'm just running it on a 9500 to give a full "silent" command that should work with writing to R3MEMID.LOG.

This worked well for me:

1R3MEMID.EXE -NOCFG -LOG -NW 1

I'm running it in FreeDOS with SafeMode. After that command has run, I get the output written to R3MEMID.LOG. In any case try to delete the R3MEMID.LOG file beforehand, just to make sure it re-creates it.

Do you see "things happening" when the test runs?

Yes things "happen" when it's running. I managed to get a log by freeing up disk space and running "*" again. Here is the result.

1R3MEMID version 1.07, (c) Copyright ATI Technologies Inc, 2003
2Log file generation enabled to .\R3MEMID.LOG ...
3Reference data file (RDF) loading disabled ...
4[1  ] Fill                                                : FAIL
5    Error ID 0VB001
6  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
7    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
8[2  ] Data line toggle (GUI)                              : FAIL
9    Error ID 0VB023
10  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
11    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
12[3  ] RW page (GUI)                                       : FAIL
13    Error ID 0VB025
14  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
15    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
16[4  ] RW channel (GUI)                                    : FAIL
17    Error ID 0VB028
18  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
19    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
20[5  ] Data mask                                           : FAIL
21    Error ID 0VB006
22  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
23    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
24[6  ] Macro + Micro8x2 tiling read                        : FAIL
25    Error ID 0VC006
26  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
27    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
28[7  ] Byte swap macro+micro8x2 read                       : FAIL
29    Error ID 0VC018
30  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
31    failing bit : MDA0 0 1 2 3 4 5 6 7 ... MDB0 0 1 2 3 4 5 6 7 ... MDC0 0 1 2 3 4 5 6 7 ... MDD0 0
32[8  ] HOST_DATA/<f,b>/M2L                                 : FAIL
33    Error ID 0TR014
34   640 x  480 -  8 bpp ( 75 Hz): TEST FAILURE
35     error : R051A.8 not found
36   640 x  480 - 15 bpp ( 75 Hz): TEST FAILURE
37     error : R052A.8 not found
38   640 x  480 - 16 bpp ( 75 Hz): TEST FAILURE
39     error : R053A.8 not found
40   640 x  480 - 32 bpp ( 75 Hz): TEST FAILURE
41     error : R055A.8 not found
42   800 x  600 -  8 bpp ( 75 Hz): TEST FAILURE
43     error : R061A.8 not found
44   800 x  600 - 15 bpp ( 75 Hz): TEST FAILURE
45     error : R062A.8 not found
46   800 x  600 - 16 bpp ( 75 Hz): TEST FAILURE
47     error : R063A.8 not found
48   800 x  600 - 32 bpp ( 75 Hz): TEST FAILURE
49     error : R065A.8 not found
50  1024 x  768 -  8 bpp ( 75 Hz): TEST FAILURE
51     error : R071A.8 not found
52  1024 x  768 - 15 bpp ( 75 Hz): TEST FAILURE
53     error : R072A.8 not found
54  1024 x  768 - 16 bpp ( 75 Hz): TEST FAILURE
55     error : R073A.8 not found
56  1024 x  768 - 32 bpp ( 75 Hz): TEST FAILURE
57     error : R075A.8 not found
58[9  ] Line slopes/dir/P/solid                             : FAIL
59    Error ID 0TR010
60   640 x  480 -  8 bpp ( 75 Hz): TEST FAILURE

61     error : R051A.9 not found
62   640 x  480 - 15 bpp ( 75 Hz): TEST FAILURE
63     error : R052A.9 not found
64   640 x  480 - 16 bpp ( 75 Hz): TEST FAILURE
65     error : R053A.9 not found
66   640 x  480 - 32 bpp ( 75 Hz): TEST FAILURE
67     error : R055A.9 not found
68   800 x  600 -  8 bpp ( 75 Hz): TEST FAILURE
69     error : R061A.9 not found
70   800 x  600 - 15 bpp ( 75 Hz): TEST FAILURE
71     error : R062A.9 not found
72   800 x  600 - 16 bpp ( 75 Hz): TEST FAILURE
73     error : R063A.9 not found
74   800 x  600 - 32 bpp ( 75 Hz): TEST FAILURE
75     error : R065A.9 not found
76  1024 x  768 -  8 bpp ( 75 Hz): TEST FAILURE
77     error : R071A.9 not found
78  1024 x  768 - 15 bpp ( 75 Hz): TEST FAILURE
79     error : R072A.9 not found
80  1024 x  768 - 16 bpp ( 75 Hz): TEST FAILURE
81     error : R073A.9 not found
82  1024 x  768 - 32 bpp ( 75 Hz): TEST FAILURE
83     error : R075A.9 not found

Wow. Memory issues on *every* channel? Consistently starting with bit 0, that's brutal. I guess it's time to check voltages?

tehsiggi wrote on 2025-09-23, 08:18:

Wow. Memory issues on *every* channel? Consistently starting with bit 0, that's brutal. I guess it's time to check voltages?

I’ll report back with voltages but is it possible the previous owner reversed the polarity on the 4 pin connector thus blowing all the memory chips?

You'd probably have a hard short on the MVDDQ line then. I'll get you some reference values for voltages from a non-Pro and drop them here.
Also, the regulator should just work fine with 12V on the input as well. There are many more parts before the memory ICs that'd die. The input caps for example would go bang very quickly if anything like that happened.

Measuring time.

Here are the values of my non-Pro (Infineon HYB25D128323C-3.3):

Yellow/Orange (VTT): 1.245V
Green (MVDDC): 2.850V
Blue (MVDDQ): 2.483V

I also measured my Hercules 9700Pro (has the same memory as yours, Samsung K4D26323RA-GC2A):

Yellow/Orange (VTT): 1.395V
Green (MVDDC): 2.922V
Blue (MVDDQ): 2.793V

VTT should always be roughly half of MVDDQ, though the reference layout does not configure it as a /2 follower.

I attached the reference schematic as well.

tehsiggi wrote on 2025-09-23, 08:22:

You'd probably have a hard short on the MVDDQ line then. I'll get you some reference values for voltages from a non-Pro and drop […]
Show full quote

You'd probably have a hard short on the MVDDQ line then. I'll get you some reference values for voltages from a non-Pro and drop them here.
Also, the regulator should just work fine with 12V on the input as well. There are many more parts before the memory ICs that'd die. The input caps for example would go bang very quickly if anything like that happened.

Measuring time.

The attachment 9700ref.jpg is no longer available

Here are the values of my non-Pro (Infineon HYB25D128323C-3.3):

Yellow/Orange (VTT): 1.245V
Green (MVDDC): 2.850V
Blue (MVDDQ): 2.483V

I also measured my Hercules 9700Pro (has the same memory as yours, Samsung K4D26323RA-GC2A):

Yellow/Orange (VTT): 1.395V
Green (MVDDC): 2.922V
Blue (MVDDQ): 2.793V

VTT should always be roughly half of MVDDQ, though the reference layout does not configure it as a /2 follower.

I attached the reference schematic as well.

MVDDQ doesn't look happy. Only 1.13v

1.13V does not sound very happy. Do you have an oscilloscope handy? With that you can check if the ISL6522 is in current protection mode by measuring the voltage on Pin 3 (SS) of it. If it's ramping up and down, we're in current limiting mode. The ISL6522 for VDDQ is under the small heatsink, together with the SC1175.

One statement to clear up from earlier: 5V_EXT is used for the GPU, 12V_EXT for the Memory.

I did a measurement from GND to VDDQ in resistance mode and it's around 70Ohms on my R97Pro. Any of the memory chips getting hotter than the others?
It's also worth mentioning that VDDQ also goes to the GPU and might be loaded by it.

tehsiggi wrote on 2025-09-23, 09:59:

1.13V does not sound very happy. Do you have an oscilloscope handy? With that you can check if the ISL6522 is in current protect […]
Show full quote

1.13V does not sound very happy. Do you have an oscilloscope handy? With that you can check if the ISL6522 is in current protection mode by measuring the voltage on Pin 3 (SS) of it. If it's ramping up and down, we're in current limiting mode. The ISL6522 for VDDQ is under the small heatsink, together with the SC1175.

One statement to clear up from earlier: 5V_EXT is used for the GPU, 12V_EXT for the Memory.

I did a measurement from GND to VDDQ in resistance mode and it's around 70Ohms on my R97Pro. Any of the memory chips getting hotter than the others?
It's also worth mentioning that VDDQ also goes to the GPU and might be loaded by it.

I managed to get the heat sink off, 3x ICs were glued to it. Do you know which chip it is?

It’s the larger chip! I’ll get the oscilloscope.

It's U51:

Pin 3 is SS (Soft-Start):

Here is what it looks like in theory:

I have a real scope screenshot down in the basement for reference, if needed.

Here's the link to the ISL6522 datasheet:
https://www.renesas.com/en/document/dst/isl6522-datasheet

I’ve just noticed there is an IC that got ripped off it’s pads from the glue and is stuck under the heat sink. I’ll continue tomorrow with getting that IC soldered back on 😆 I think next time I’ll put it in the freezer before pulling that heat sink off.

Really appreciate your help so far. There might be hope for this card.

Yep, you just ripped of it's dual FET (Q9) 😁 No more switching there, that's for sure.

While it's gone (lol) feel free to measure the resistance between MVDDQ and GND.

Here's btw. a picture of the SS pin from ISL6522 with over-current protection active:

Yellow is the waveform of interest. I think blue was one of the Gates, either high or low-side. In this case the FET for the low side had a disconnect on it's gate. Leading to the voltage not being generated correctly.