Ozzuneoj wrote on 2025-09-08, 20:49:

Anyone know why a GTX 770 would simply not work in an X79 motherboard in Windows 10? […]
Show full quote

Anyone know why a GTX 770 would simply not work in an X79 motherboard in Windows 10?

I have two different EVGA 770s and I have tried different driver packages (475.14 is the latest), tried DVI and DisplayPort...

With the default Windows 10 drivers from 2020 I was just getting a black screen during startup (not no signal, an actual black screen). Switching to 475.14 (the latest with 770 support) it will crash during the driver install with a garbled screen, and when I reboot the system it will hang with a garbled screen or slowly load a sad face blue screen, where it then hangs indefinitely.

Very odd issue though since it happens with two different cards. I am not familiar with the X79 platform and I haven't used a Kepler based card in Windows 10 in years, so I don't know if there are some hoops I have to jump through to make these work together happily. Any ideas?

BTW, I know this is probably not retro to a lot of people, but none of this 12+ year old hardware is supported by Windows or the manufacturers anymore. If I ask anywhere but VOGONS I'm going to get harassed... 😀

EDIT: Hmm... seems to work if I switch the PCI-E slots to Gen2 in the BIOS? Man... that'd put a damper on my plan to run SLI on this thing...

I am running a 4930K so it should work. Also, a Quadro K2200 (Maxwell based... basically a 4GB GTX 750) works fine without touching PCI-E gen settings.

Googling it I'm not coming up with any reason why an X79 system with a 4930K would not work with a GTX 770 in a Gen3 slot. The motherboard BIOS is fully up to date as well. The board is an ASUS P9X79-E WS.

On a similar note, I'm still trying to understand why I had this issue with a pair of GeForce FX 5500s.

Two FX 5500s but only one works in my 440EX

X79/GTX770/Win10 : should work.
Although I don't have an X79 machine, the GTX770 was fine in X99.
Currently it runs in AM4.

DarthSun wrote on 2025-09-09, 22:23:

X79/GTX770/Win10 : should work.
Although I don't have an X79 machine, the GTX770 was fine in X99.
Currently it runs in AM4.

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS.

I thought maybe a CPU socket pin was damaged, but a Quadro K2200 works fine and it also uses PCI-E 3.0 x16, so I don't believe it's that. I already had the cooler off to repaste and it's a pain in the butt to remove and reinstall, so I will save a socket check as a last resort. 😮

Anyway... I searched on Google and the AI slop answer said this:

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often s […]
Show full quote

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often stable in PCIe 2.0 (Gen2) mode
. This problem stemmed from the early and non-native nature of PCIe 3.0 support on these motherboards.
Reasons for X79 instability with PCIe 3.0

Non-native Gen3 support: The original Intel X79 chipset only natively supported PCIe 2.0. To enable PCIe 3.0, motherboard manufacturers had to implement workarounds using different clock generators or signal re-drivers. This "hack" was inconsistent across motherboard brands and CPUs, and was never officially certified by Intel.
Variable signal timing: NVIDIA noted in 2021 that on X79 platforms running at Gen3 speeds, they observed "significant variation in signal timing across different motherboards and CPUs". This variation could cause instability, especially with early PCIe 3.0 graphics cards like the GTX 600 series.
CPU and motherboard "hit or miss": Whether a system could achieve stable PCIe 3.0 performance was often a gamble, even with compatible Ivy Bridge-E CPUs, as stability depended on a combination of the specific CPU and motherboard.
Overclocking issues: Users who overclocked the Base Clock (BCLK) on X79 systems found that doing so also overclocked the PCIe bus, further increasing instability and forcing them to revert to Gen2.

How to resolve the instability
If you are running an X79 system, you can enforce PCIe 2.0 speeds for stability.

Check your BIOS: The most reliable method is to enter your motherboard's BIOS settings. Look for the PCIe configuration options under "Advanced," "Peripherals," or "OC Tweaker."
Manually set the PCIe speed: In the BIOS, find the setting for "PCIe Link Speed" or a similar option and manually set it to "Gen2," "PCIe 2.0," or "5 GT/s".
Use a software patch (not recommended): Early on, some users attempted to force PCIe 3.0 using software tools and registry hacks, particularly for NVIDIA cards. This method was notoriously unreliable and could create more problems.

Note: For most single-GPU configurations, the performance difference between PCIe 2.0 and PCIe 3.0 is negligible and not noticeable in real-world use. Sticking with PCIe 2.0 on an X79 platform is a practical solution that ensures maximum stability.

I have no idea if any of that is true or not. I feel like using an Ivy Bridge CPU in such a high end board with the latest BIOS should be fine, but according to this, early Gen3 cards may have issues with early Gen3 boards. That might explain why the Maxwell-based K2200 works...

Though this system came with a 770. Maybe the newer BIOS version I put on it broke that compatibility...

Bleh... anyway. I have no idea what I'm even going to do with this monstrosity, so I should probably just put a note on it about GTX 770 compatibility and not worry about it for now.

If I came across three or four 980 Ti cards it'd be funny to turn this system into a space heater that trades blows with modern mid-range systems (as long as 6GB of VRAM is enough).

Ozzuneoj wrote on 2025-09-10, 00:07:

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS. […]
Show full quote

DarthSun wrote on 2025-09-09, 22:23:

X79/GTX770/Win10 : should work.
Although I don't have an X79 machine, the GTX770 was fine in X99.
Currently it runs in AM4.

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS.

I thought maybe a CPU socket pin was damaged, but a Quadro K2200 works fine and it also uses PCI-E 3.0 x16, so I don't believe it's that. I already had the cooler off to repaste and it's a pain in the butt to remove and reinstall, so I will save a socket check as a last resort. 😮

Anyway... I searched on Google and the AI slop answer said this:

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often s […]
Show full quote

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often stable in PCIe 2.0 (Gen2) mode
. This problem stemmed from the early and non-native nature of PCIe 3.0 support on these motherboards.
Reasons for X79 instability with PCIe 3.0

Non-native Gen3 support: The original Intel X79 chipset only natively supported PCIe 2.0. To enable PCIe 3.0, motherboard manufacturers had to implement workarounds using different clock generators or signal re-drivers. This "hack" was inconsistent across motherboard brands and CPUs, and was never officially certified by Intel.
Variable signal timing: NVIDIA noted in 2021 that on X79 platforms running at Gen3 speeds, they observed "significant variation in signal timing across different motherboards and CPUs". This variation could cause instability, especially with early PCIe 3.0 graphics cards like the GTX 600 series.
CPU and motherboard "hit or miss": Whether a system could achieve stable PCIe 3.0 performance was often a gamble, even with compatible Ivy Bridge-E CPUs, as stability depended on a combination of the specific CPU and motherboard.
Overclocking issues: Users who overclocked the Base Clock (BCLK) on X79 systems found that doing so also overclocked the PCIe bus, further increasing instability and forcing them to revert to Gen2.

How to resolve the instability
If you are running an X79 system, you can enforce PCIe 2.0 speeds for stability.

Check your BIOS: The most reliable method is to enter your motherboard's BIOS settings. Look for the PCIe configuration options under "Advanced," "Peripherals," or "OC Tweaker."
Manually set the PCIe speed: In the BIOS, find the setting for "PCIe Link Speed" or a similar option and manually set it to "Gen2," "PCIe 2.0," or "5 GT/s".
Use a software patch (not recommended): Early on, some users attempted to force PCIe 3.0 using software tools and registry hacks, particularly for NVIDIA cards. This method was notoriously unreliable and could create more problems.

Note: For most single-GPU configurations, the performance difference between PCIe 2.0 and PCIe 3.0 is negligible and not noticeable in real-world use. Sticking with PCIe 2.0 on an X79 platform is a practical solution that ensures maximum stability.

I have no idea if any of that is true or not. I feel like using an Ivy Bridge CPU in such a high end board with the latest BIOS should be fine, but according to this, early Gen3 cards may have issues with early Gen3 boards. That might explain why the Maxwell-based K2200 works...

Though this system came with a 770. Maybe the newer BIOS version I put on it broke that compatibility...

Bleh... anyway. I have no idea what I'm even going to do with this monstrosity, so I should probably just put a note on it about GTX 770 compatibility and not worry about it for now.

If I came across three or four 980 Ti cards it'd be funny to turn this system into a space heater that trades blows with modern mid-range systems (as long as 6GB of VRAM is enough).

On X99/AM4 I didn't have to configure the cards either, GenX was the same in the UEFI.
In the case of X299, however, I had to configure: 7900GTX/7800GTX - Gen1, HD6870 - Gen2, in this case it is really the only stable one. Luckily, on the Taichi board, you can configure per slot, because there is now a GTX1650 in it in addition to the 7800GTX.

Ozzuneoj wrote on 2025-09-10, 00:07:

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS. […]
Show full quote

DarthSun wrote on 2025-09-09, 22:23:

X79/GTX770/Win10 : should work.
Although I don't have an X79 machine, the GTX770 was fine in X99.
Currently it runs in AM4.

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS.

I thought maybe a CPU socket pin was damaged, but a Quadro K2200 works fine and it also uses PCI-E 3.0 x16, so I don't believe it's that. I already had the cooler off to repaste and it's a pain in the butt to remove and reinstall, so I will save a socket check as a last resort. 😮

Anyway... I searched on Google and the AI slop answer said this:

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often s […]
Show full quote

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often stable in PCIe 2.0 (Gen2) mode
. This problem stemmed from the early and non-native nature of PCIe 3.0 support on these motherboards.
Reasons for X79 instability with PCIe 3.0

Non-native Gen3 support: The original Intel X79 chipset only natively supported PCIe 2.0. To enable PCIe 3.0, motherboard manufacturers had to implement workarounds using different clock generators or signal re-drivers. This "hack" was inconsistent across motherboard brands and CPUs, and was never officially certified by Intel.
Variable signal timing: NVIDIA noted in 2021 that on X79 platforms running at Gen3 speeds, they observed "significant variation in signal timing across different motherboards and CPUs". This variation could cause instability, especially with early PCIe 3.0 graphics cards like the GTX 600 series.
CPU and motherboard "hit or miss": Whether a system could achieve stable PCIe 3.0 performance was often a gamble, even with compatible Ivy Bridge-E CPUs, as stability depended on a combination of the specific CPU and motherboard.
Overclocking issues: Users who overclocked the Base Clock (BCLK) on X79 systems found that doing so also overclocked the PCIe bus, further increasing instability and forcing them to revert to Gen2.

How to resolve the instability
If you are running an X79 system, you can enforce PCIe 2.0 speeds for stability.

Check your BIOS: The most reliable method is to enter your motherboard's BIOS settings. Look for the PCIe configuration options under "Advanced," "Peripherals," or "OC Tweaker."
Manually set the PCIe speed: In the BIOS, find the setting for "PCIe Link Speed" or a similar option and manually set it to "Gen2," "PCIe 2.0," or "5 GT/s".
Use a software patch (not recommended): Early on, some users attempted to force PCIe 3.0 using software tools and registry hacks, particularly for NVIDIA cards. This method was notoriously unreliable and could create more problems.

Note: For most single-GPU configurations, the performance difference between PCIe 2.0 and PCIe 3.0 is negligible and not noticeable in real-world use. Sticking with PCIe 2.0 on an X79 platform is a practical solution that ensures maximum stability.

I have no idea if any of that is true or not. I feel like using an Ivy Bridge CPU in such a high end board with the latest BIOS should be fine, but according to this, early Gen3 cards may have issues with early Gen3 boards. That might explain why the Maxwell-based K2200 works...

Though this system came with a 770. Maybe the newer BIOS version I put on it broke that compatibility...

Bleh... anyway. I have no idea what I'm even going to do with this monstrosity, so I should probably just put a note on it about GTX 770 compatibility and not worry about it for now.

If I came across three or four 980 Ti cards it'd be funny to turn this system into a space heater that trades blows with modern mid-range systems (as long as 6GB of VRAM is enough).

If it works on 2. Forget about 3. It makes next to zero difference.
I have a gen 4 ssd that, with the dodgy aliexpress special wifi m2 1:4 adapter I use wasn't stable and also very slow in gen 3.
I had to mod the bios to unlock the speed settings.
Set it to 2 and its fine. It's 4x faster too.

16 lanes of gen 2 has about 8 gigabytes per second. I'm assuming without looking that a 770 is probably a 2gb card.
So you can fill the board 4x per second.

That's plenty, mate.

myne wrote on 2025-09-10, 03:17:

If it works on 2. Forget about 3. It makes next to zero difference. I have a gen 4 ssd that, with the dodgy aliexpress special w […]
Show full quote

Ozzuneoj wrote on 2025-09-10, 00:07:

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS. […]
Show full quote

DarthSun wrote on 2025-09-09, 22:23:

X79/GTX770/Win10 : should work.
Although I don't have an X79 machine, the GTX770 was fine in X99.
Currently it runs in AM4.

It's very odd that it works perfectly if I force PCI-E Gen2 in the BIOS.

I thought maybe a CPU socket pin was damaged, but a Quadro K2200 works fine and it also uses PCI-E 3.0 x16, so I don't believe it's that. I already had the cooler off to repaste and it's a pain in the butt to remove and reinstall, so I will save a socket check as a last resort. 😮

Anyway... I searched on Google and the AI slop answer said this:

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often s […]
Show full quote

The Intel X79 platform is known to have compatibility issues with PCIe 3.0 (Gen3), making it unstable in this mode while often stable in PCIe 2.0 (Gen2) mode
. This problem stemmed from the early and non-native nature of PCIe 3.0 support on these motherboards.
Reasons for X79 instability with PCIe 3.0

Non-native Gen3 support: The original Intel X79 chipset only natively supported PCIe 2.0. To enable PCIe 3.0, motherboard manufacturers had to implement workarounds using different clock generators or signal re-drivers. This "hack" was inconsistent across motherboard brands and CPUs, and was never officially certified by Intel.
Variable signal timing: NVIDIA noted in 2021 that on X79 platforms running at Gen3 speeds, they observed "significant variation in signal timing across different motherboards and CPUs". This variation could cause instability, especially with early PCIe 3.0 graphics cards like the GTX 600 series.
CPU and motherboard "hit or miss": Whether a system could achieve stable PCIe 3.0 performance was often a gamble, even with compatible Ivy Bridge-E CPUs, as stability depended on a combination of the specific CPU and motherboard.
Overclocking issues: Users who overclocked the Base Clock (BCLK) on X79 systems found that doing so also overclocked the PCIe bus, further increasing instability and forcing them to revert to Gen2.

How to resolve the instability
If you are running an X79 system, you can enforce PCIe 2.0 speeds for stability.

Check your BIOS: The most reliable method is to enter your motherboard's BIOS settings. Look for the PCIe configuration options under "Advanced," "Peripherals," or "OC Tweaker."
Manually set the PCIe speed: In the BIOS, find the setting for "PCIe Link Speed" or a similar option and manually set it to "Gen2," "PCIe 2.0," or "5 GT/s".
Use a software patch (not recommended): Early on, some users attempted to force PCIe 3.0 using software tools and registry hacks, particularly for NVIDIA cards. This method was notoriously unreliable and could create more problems.

Note: For most single-GPU configurations, the performance difference between PCIe 2.0 and PCIe 3.0 is negligible and not noticeable in real-world use. Sticking with PCIe 2.0 on an X79 platform is a practical solution that ensures maximum stability.

I have no idea if any of that is true or not. I feel like using an Ivy Bridge CPU in such a high end board with the latest BIOS should be fine, but according to this, early Gen3 cards may have issues with early Gen3 boards. That might explain why the Maxwell-based K2200 works...

Though this system came with a 770. Maybe the newer BIOS version I put on it broke that compatibility...

Bleh... anyway. I have no idea what I'm even going to do with this monstrosity, so I should probably just put a note on it about GTX 770 compatibility and not worry about it for now.

If I came across three or four 980 Ti cards it'd be funny to turn this system into a space heater that trades blows with modern mid-range systems (as long as 6GB of VRAM is enough).

If it works on 2. Forget about 3. It makes next to zero difference.
I have a gen 4 ssd that, with the dodgy aliexpress special wifi m2 1:4 adapter I use wasn't stable and also very slow in gen 3.
I had to mod the bios to unlock the speed settings.
Set it to 2 and its fine. It's 4x faster too.

16 lanes of gen 2 has about 8 gigabytes per second. I'm assuming without looking that a 770 is probably a 2gb card.
So you can fill the board 4x per second.

That's plenty, mate.

Right, for a single GTX 770 there is definitely no difference. My hope was to use the board to dabble in SLI since that's how the system was originally used. I have basically no experience with SLI (beyond the PTSD I suffer from the 7950GX2), so I assumed that having the additional PCI bandwidth was a big deal... but maybe it isn't? I've Googled it a bit, and the few answers I'm finding seem to indicate it won't make much of a difference, assuming the board can provide at least x8 lanes to each slot, which this one should for up to four GPUs.

At this point, I don't even have two identical GPUs from this time period on hand to mess with SLI so this is just a future project.

Maybe I could run four way 8800 GTX SLI to see how far that'd stretch the performance of nearly 20 year old GPUs... I'm pretty sure I actually have those on hand, though I haven't tested them yet.

I doubt there's any real difference in sli.

tehsiggi wrote on 2025-09-09, 13:32:

The Hercules Radeon 9600 Pro arrived.. it either does not post.. or if I apply pressure on certain memory IC:

The pattern changes, so that's good. No time for it today.. but it'll live again!

As for my X800GTO that arrived.. well not sure, it plays HL2 just fine 😁

So I had a minute today to get it booting into FreeDOS and run R3MEMID:

1TEST RESULT SUMMARY:
2====================
3Rv350 (0x4150) detected.
4128M video memory.
5Test suite ran 1 of 1 times.
6Checking test status array ...
7[1  ] Fill                                                : FAIL
8    Error ID 0VB001
9  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
10    failing bit : MDA0 45 47 48 49 50 51 52 53 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
11
12[2  ] Data line toggle (GUI)                              : FAIL
13    Error ID 0VB023
14  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
15    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
16
17[3  ] RW page (GUI)                                       : FAIL
18    Error ID 0VB025
19  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
20    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
21
22[4  ] RW channel (GUI)                                    : FAIL
23    Error ID 0VB028
24  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
25    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
26
27[5  ] Data mask                                           : FAIL
28    Error ID 0VB006
29  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
30    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
31
32[6  ] Macro + Micro8x2 tiling read                        : FAIL
33    Error ID 0VC006
34  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
35    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
36
37[7  ] Byte swap macro+micro8x2 read                       : FAIL
38    Error ID 0VC018
39  1024 x  768 - 32 bpp ( 60 Hz): TEST FAILURE
40    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 MDA1 32 33 34 35 36 37 38 39 ... 
41
42[8  ] HOST_DATA/<f,b>/M2L                                 : FAIL
43    Error ID 0TR014
44   640 x  480 -  8 bpp ( 75 Hz): TEST FAILURE
45    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
46
47   640 x  480 - 15 bpp ( 75 Hz): TEST FAILURE
48    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
49
50   640 x  480 - 16 bpp ( 75 Hz): TEST FAILURE
51    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
52
53   640 x  480 - 32 bpp ( 75 Hz): TEST FAILURE
54    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 
55
56   800 x  600 -  8 bpp ( 75 Hz): TEST FAILURE
57    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 
58
59   800 x  600 - 15 bpp ( 75 Hz): TEST FAILURE
60    failing bit : MDA0 34 36 37 38 39 40 41 42 ... MDB0 11 

61
62   800 x  600 - 16 bpp ( 75 Hz): TEST FAILURE
63    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 
64
65   800 x  600 - 32 bpp ( 75 Hz): TEST FAILURE
66    failing bit : MDA0 33 34 35 36 39 40 41 42 ... MDB0 11 
67
68  1024 x  768 -  8 bpp ( 75 Hz): TEST FAILURE
69    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
70
71  1024 x  768 - 15 bpp ( 75 Hz): TEST FAILURE
72    failing bit : MDA0 33 34 35 36 37 38 39 40 ... MDB0 11 
73
74  1024 x  768 - 16 bpp ( 75 Hz): TEST FAILURE
75    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
76
77  1024 x  768 - 32 bpp ( 75 Hz): TEST FAILURE
78    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
79
80[9  ] Line slopes/dir/P/solid                             : FAIL
81    Error ID 0TR010
82   640 x  480 -  8 bpp ( 75 Hz): TEST FAILURE
83    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
84
85   640 x  480 - 15 bpp ( 75 Hz): TEST FAILURE
86    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
87
88   640 x  480 - 16 bpp ( 75 Hz): TEST FAILURE
89    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
90
91   640 x  480 - 32 bpp ( 75 Hz): TEST FAILURE
92    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
93
94   800 x  600 -  8 bpp ( 75 Hz): TEST FAILURE
95    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
96
97   800 x  600 - 15 bpp ( 75 Hz): TEST FAILURE
98    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
99
100   800 x  600 - 16 bpp ( 75 Hz): TEST FAILURE
101    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
102
103   800 x  600 - 32 bpp ( 75 Hz): TEST FAILURE
104    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
105
106  1024 x  768 -  8 bpp ( 75 Hz): TEST FAILURE
107    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
108
109  1024 x  768 - 15 bpp ( 75 Hz): TEST FAILURE
110    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
111
112  1024 x  768 - 16 bpp ( 75 Hz): TEST FAILURE
113    failing bit : MDA0 48 49 50 51 52 53 54 55 MDB0 11 
114
115  1024 x  768 - 32 bpp ( 75 Hz): TEST FAILURE
116    failing bit : MDA0 32 33 34 35 36 37 38 39 ... MDB0 11 
117
118Failure detected.

Since all tests show MDA0 32 onwards.. I'll check that first. The Hercules uses BGA memory so we have dual rank memory setup. I'll check U67 tomorrow, as it is bits 32 onwards on MDA Rank 0. And then we'll see.. I feel MDA 1 is just a side product.

tehsiggi wrote on 2025-09-10, 13:02:

Since all tests show MDA0 32 onwards.. I'll check that first. The Hercules uses BGA memory so we have dual rank memory setup. I'll check U67 tomorrow, as it is bits 32 onwards on MDA Rank 0. And then we'll see.. I feel MDA 1 is just a side product.

I'm curious to see how that goes. I've got that Radeon 9800 with display issues I need to address. I'm still hoping that it's a power delivery issue, but in my heart of hearts I feel like it's memory.

I wanted an AGP 4x board so I went to the closet and pulled out 2001-2003 from the box in the back.

Cleaned and recapped. Which is saying something because I smoked and had cats back then. I've got a waste can of brown cotton swabs and shredded paper towels.

Soyo Sy7isa+
ECS K7S5A
MSI K7T Turbo V3

ASUS A7V133 w/ AGP Pro is on deck ( Never found and AGP Pro video card to put in it ) along with a grimy looking Shuttle an35n ultra v1.1 that was my pride and joy once upon a time.

douglar wrote on Today, 00:21:

I'm curious to see how that goes. I've got that Radeon 9800 with display issues I need to address. I'm still hoping that it's a power delivery issue, but in my heart of hearts I feel like it's memory.

Well, I made some progress yesterday.
I was lazy, so I re-placed U67 since I was not in the mood to reball the existing memory. All of them are the usual K4D26323RA-GC2A from Samsung. Hercules used a strong strong strong adhesive to mount these memory coolers. I had to really go to town in terms of warming them up to get them removed. And even now the chips are not clean.

Before changing that memory chip, the card would usually get stuck on post code 25 and the system will not continue.

Now we get consistently trough with a beep that the GPU init isn't working properly, however the rest of the system boots. Furthermore, If I apply pressure on the neighboring IC (U66) the card gives a picture, albeit with some graphical issues.

So I assume the "bridge" cooler on that memory at some point go some mechanical stress, which causes BGA joints of both memory ICs to suffer. Next step is to de-solder, re-ball and re-solder U66, since the issue is pressure related.

I'll get to that next week, but I'm confident to get that card going again. Will be a nice addition to the repair report thread, as it's not as "straight forward" as previous repairs.

I also noticed that this card suffered from a bad life before. There's a lot of nicotine residue on the card. Nasty.

tehsiggi wrote on Today, 04:34:

Well, I made some progress yesterday. I was lazy, so I re-placed U67 since I was not in the mood to reball the existing memory. […]
Show full quote

douglar wrote on Today, 00:21:

I'm curious to see how that goes. I've got that Radeon 9800 with display issues I need to address. I'm still hoping that it's a power delivery issue, but in my heart of hearts I feel like it's memory.

Well, I made some progress yesterday.
I was lazy, so I re-placed U67 since I was not in the mood to reball the existing memory. All of them are the usual K4D26323RA-GC2A from Samsung. Hercules used a strong strong strong adhesive to mount these memory coolers. I had to really go to town in terms of warming them up to get them removed. And even now the chips are not clean.

The attachment WhatsApp Image 2025-09-12 at 06.25.41.jpeg is no longer available

Before changing that memory chip, the card would usually get stuck on post code 25 and the system will not continue.

The attachment WhatsApp Image 2025-09-12 at 06.25.40.jpeg is no longer available

Now we get consistently trough with a beep that the GPU init isn't working properly, however the rest of the system boots. Furthermore, If I apply pressure on the neighboring IC (U66) the card gives a picture, albeit with some graphical issues.

So I assume the "bridge" cooler on that memory at some point go some mechanical stress, which causes BGA joints of both memory ICs to suffer. Next step is to de-solder, re-ball and re-solder U66, since the issue is pressure related.

The attachment WhatsApp Image 2025-09-12 at 06.25.41(1).jpeg is no longer available

I'll get to that next week, but I'm confident to get that card going again. Will be a nice addition to the repair report thread, as it's not as "straight forward" as previous repairs.

I also noticed that this card suffered from a bad life before. There's a lot of nicotine residue on the card. Nasty.

Loving these updates. You should create a dedicated post for this repair.

What temp did you heat the top PCB to for removing the memory heat sinks?

@tehsiggi are you aware of a AGP version of a PCIE riser card?

I was messing with a fake FX5500 card and after some detective work, I found out chip is Geforce EMP salvaged from Audi J794 modules.
these chips still are availble in large quantity. chip itself has same spec as MX 4000.

zuldan wrote on Today, 12:07:

Loving these updates. You should create a dedicated post for this repair.

What temp did you heat the top PCB to for removing the memory heat sinks?

Once the card is finished, I'll add the report to the collection as soon as it's done.

zuldan wrote on Today, 12:13:

@tehsiggi are you aware of a AGP version of a PCIE riser card?

Long flexible riser?
No. There are some fixed PCB riser cards. But flexible with long connections? No. The issue is the architecture of AGP. Those Riser cards shown in your picture only use PCIe 1x, so it doesn't need to many connections. For AGP however, there is no smaller subset. So I guess this is not trivial.

The picture is a 1x riser, but full 16x risers exist, with a super wide ribbon as expected.

edit: but I should say that PCIe arranges things as impedance matched transmission lines, like it were twisted pair ethernet, but AGP won't, so that makes it not as easy.

Exactly. PCIe is way more forgiving imho than AGP

tehsiggi wrote on Today, 14:17:

Long flexible riser?
No. There are some fixed PCB riser cards. But flexible with long connections? No. The issue is the architecture of AGP. Those Riser cards shown in your picture only use PCIe 1x, so it doesn't need to many connections. For AGP however, there is no smaller subset. So I guess this is not trivial.

I actually just want to power the AGP card without having it connected to a motherboard so I can do some basic voltage checks. That’s a shame nothing exists.

zuldan wrote on Today, 17:45:

tehsiggi wrote on Today, 14:17:

Long flexible riser?
No. There are some fixed PCB riser cards. But flexible with long connections? No. The issue is the architecture of AGP. Those Riser cards shown in your picture only use PCIe 1x, so it doesn't need to many connections. For AGP however, there is no smaller subset. So I guess this is not trivial.

I actually just want to power the AGP card without having it connected to a motherboard so I can do some basic voltage checks. That’s a shame nothing exists.

My APM has breakout contacts that you can use for exactly that.

I use them for calibration of the current and voltage measurement during build and later on I leverage them to inject voltages without a motherboard. I thought about building a "dummy power" PCB that will just power a GPU via AGP and provide the voltages and current via USB, similar to what the APM does, but without the riser aspect of it. For now I'm good enough with the APM however.

tehsiggi wrote on Today, 17:55:

zuldan wrote on Today, 17:45:

tehsiggi wrote on Today, 14:17:

Long flexible riser?
No. There are some fixed PCB riser cards. But flexible with long connections? No. The issue is the architecture of AGP. Those Riser cards shown in your picture only use PCIe 1x, so it doesn't need to many connections. For AGP however, there is no smaller subset. So I guess this is not trivial.

I actually just want to power the AGP card without having it connected to a motherboard so I can do some basic voltage checks. That’s a shame nothing exists.

My APM has breakout contacts that you can use for exactly that.

How do you plug a AGP card into that? I’m confused 🫤

zuldan wrote on Today, 17:58:

tehsiggi wrote on Today, 17:55:

zuldan wrote on Today, 17:45:

I actually just want to power the AGP card without having it connected to a motherboard so I can do some basic voltage checks. That’s a shame nothing exists.

My APM has breakout contacts that you can use for exactly that.

How do you plug a AGP card into that? I’m confused 🫤

Well, what you see on that picture is the back, here's the front:

You can read more about it here: AGP Power monitor - diagnostic hardware tool