Carefully remove the heatsink from the video card and inspect the condition of the parts and tracks in the specified area on both sides. The last two contacts (one on each side) are responsible for selecting the AGP voltage.

The fact that it won't even POST means it's not a driver issue...

I would suspect marginal connections- "marginal" because other cards
work. Closely inspect the card edge and AGP on-board, look for anything even
slightly corroded or otherwise marginal.

The other thing that comes to mine is some sort of weird BIOS clock
settings... (but most BIOS will POST at safe settings so you can enter to fix
"weird things") - I assume you've tried BIOS factory reset.

[just re-read .. for some reason I was thinking it was one board combination
that doesn't work - chances of marginal connections on multiple configs
are not very likely at all]

shevalier wrote on 2025-03-14, 11:35:

Carefully remove the heatsink from the video card and inspect the condition of the parts and tracks in the specified area on both sides. The last two contacts (one on each side) are responsible for selecting the AGP voltage.

Is there any way to test this to see if it's correct? e.g. is it just a certain resistance. Probably not but just curious if it can be tested to see if it has the correct "value".

Thanks for the suggestion, I'll take a look for anything suspicious.

DaveDDS wrote on 2025-03-14, 11:53:

The fact that it won't even POST means it's not a driver issue... […]
Show full quote

The fact that it won't even POST means it's not a driver issue...

I would suspect marginal connections- "marginal" because other cards
work. Closely inspect the card edge and AGP on-board, look for anything even
slightly corroded or otherwise marginal.

The other thing that comes to mine is some sort of weird BIOS clock
settings... (but most BIOS will POST at safe settings so you can enter to fix
"weird things") - I assume you've tried BIOS factory reset.

[just re-read .. for some reason I was thinking it was one board combination
that doesn't work - chances of marginal connections on multiple configs
are not very likely at all]

Good suggestion, I hadn't reset some of the BIOS. But that doesn't seem to make a difference. Still no boot on most systems.

What's odd too is sometimes I don't even get POST beeps, it just sits doing nothing.

I have no idea why it seems so happy on the one system, 🤣.

Tried another board I had lying around that supports AGP 4x/8x
MSI K8T Neo https://theretroweb.com/motherboards/s/msi-k8t-neo#docs
Same results at P3 systems. The "good" 8x board works fine, the problem child pictured above (apparent 4x) does not POST. No beeps or anything.

So....only seems happy on the one system, this is very odd.

I have a Ti4600 that’s very fussy. It will only post on certain motherboards. What did help is using an old PSU, 50amps on the 5v line (I was using a modern PSU with 20amps on the 5 line). Old PSU with certain motherboards worked better. I also had motherboards where the modern PSU worked fine with the Ti4600. So maybe try changing the PSU and see if that makes a difference.

Guld wrote on 2025-03-14, 16:55:

... the problem child pictured above (apparent 4x) does not POST. No beeps or anything ...
So....only seems happy on the one system, this is very odd.

Agreed, it is very odd - "no beeps" kinda says something ... Most BIOS do some very basic tests
at the very beginning, and if they don't "think" enough of the system is able to start, they will
produce "beep codes".

So as far as I can tell, it's gotta be something like:

1. Something fairly fundamental is getting so heavily loaded (clock of some kind?, address bus?)
that the system can't even get as far as those basic tests.

- or -

2. It's passing the basic tests and something else is preventing it from getting to full POST
(although in my experience most BIOS will produce one beep very soon after the basic tests)

I'd probably figure out the pinout for the BIOS ROM and try to scope it and see
if it looks like code fetches are happening out of reset. Since you can boot the system
with another video card, you can capture the BIOS code and then see if those initial code
fetches are reading the instructions from ROM correctly.

(This can be painful - but doing stuff like this is one of the main reasons I got a 4-channel scope)

zuldan wrote on 2025-03-14, 20:45:

.... What did help is using an old PSU, 50amps on the 5v line ...

I hadn't though of this - presumable POST wouldn't put the card into a "high current demand"
state - and it works find with other cards ....

Noise?
Worth scoping the power rails during startup to see if there's anything looking odd.

When I did some investigation as to why the Ti4600 worked on some systems… I found some motherboards with my modern PSU, reported the 5v line being 4.5v (out of spec). When I used the old PSU on the same motherboard the 5v line went to 4.89 (in spec). If I plugged the same modern PSU into another motherboard, the 5v line would report as 4.95v. So there is some weird compatibility with modern/older PSUs and certain motherboards. It’s like some old voltage regulators like older PSUs. I’m no electronics engineer. Just reporting what I’ve seen.

Guld wrote on 2025-03-14, 16:42:

shevalier wrote on 2025-03-14, 11:35:

Carefully remove the heatsink from the video card and inspect the condition of the parts and tracks in the specified area on both sides. The last two contacts (one on each side) are responsible for selecting the AGP voltage.

Is there any way to test this to see if it's correct? e.g. is it just a certain resistance. Probably not but just curious if it can be tested to see if it has the correct "value".

Thanks for the suggestion, I'll take a look for anything suspicious.

When removing the cooler, SMD parts often break off. And it is in this place that very important resistor dividers are located, which are responsible for the AGP levels - 3.3 V (2x), 1.5 V (4x) or 0.75 V (8x)

https://elektrotanya.com/nvidia_geforce_2mx_n … f/download.html
page 4
signal
-"typedet" - A2
- AGPVREFGC - A66
-AGPVREFCG- B66

zuldan wrote on 2025-03-14, 20:45:

I have a Ti4600 that’s very fussy. It will only post on certain motherboards. What did help is using an old PSU, 50amps on the 5v line (I was using a modern PSU with 20amps on the 5 line). Old PSU with certain motherboards worked better. I also had motherboards where the modern PSU worked fine with the Ti4600. So maybe try changing the PSU and see if that makes a difference.

So I pulled out an older power supply (as you said, I was mostly using modern +5/20A PSUs). This one is +5/36A.
And slight change in behavior.

+5/36A PSU:
P3 system + AGP 4x Ti4200 now gives me the POST error beeps for no video, seems to do it regularly.

+5/20A PSU:
P3 system + AGP 4x Ti4200 rarely gives me POST error beeps. Also not seeing any error codes on my diag card...works with other video cards...hmmmm.
I disconnected the floppy drives I had connected to the power supply, and then it gives me the regular error code beeps and seems much more consistent about it (not sure if 100% of the time, but possibly).

So....might be a power issue. Not sure why the combo-Z does better with it. Maybe because the CPU uses the CPU power connector and the +12 on that and leaves more of the +5 for other devices?
Doesn't explain why my other motherboard with a similar CPU connector also does not work though...Maybe just different design.

DaveDDS wrote on 2025-03-14, 20:46:

I'd probably figure out the pinout for the BIOS ROM and try to scope it and see if it looks like code fetches are happening out […]
Show full quote

I'd probably figure out the pinout for the BIOS ROM and try to scope it and see
if it looks like code fetches are happening out of reset. Since you can boot the system
with another video card, you can capture the BIOS code and then see if those initial code
fetches are reading the instructions from ROM correctly.

(This can be painful - but doing stuff like this is one of the main reasons I got a 4-channel scope)

I do have a 4 channel scope and even have some chip clips, but I have no idea how to go about looking at what you are suggesting. Is this the same as the POST code displayed on my diag card? Would appreciate any more info.

shevalier wrote on 2025-03-15, 05:02:

When removing the cooler, SMD parts often break off. And it is in this place that very important resistor dividers are located, […]
Show full quote

Guld wrote on 2025-03-14, 16:42:

shevalier wrote on 2025-03-14, 11:35:

Carefully remove the heatsink from the video card and inspect the condition of the parts and tracks in the specified area on both sides. The last two contacts (one on each side) are responsible for selecting the AGP voltage.

Is there any way to test this to see if it's correct? e.g. is it just a certain resistance. Probably not but just curious if it can be tested to see if it has the correct "value".

Thanks for the suggestion, I'll take a look for anything suspicious.

When removing the cooler, SMD parts often break off. And it is in this place that very important resistor dividers are located, which are responsible for the AGP levels - 3.3 V (2x), 1.5 V (4x) or 0.75 V (8x)

https://elektrotanya.com/nvidia_geforce_2mx_n … f/download.html
page 4
signal
-"typedet" - A2
- AGPVREFGC - A66
-AGPVREFCG- B66

Thanks! Will take a look. I do find it somewhat odd that the two boards look nearly identical and are both reportedly 128 MB Ti4200, and both even have the same memory manufacturer and part number; and yet one seems to be AGP 4x and the other AGP 8x. Could be unrelated though, maybe just binned and sold cheaper?

ADDED:
Okay, I used HWINFO to get more info on the two cards. The one I am testing is an NV25 which only goes up to AGP 4x, and the one that works better is an NV28 which is AGP 8x. So they are very similar but different chips. So the reporting of the one board as AGP 4x seems correct.

There shouldn't be any issues with the KT7-RAID as I have one and it works with just about everything I tried in it.
I have a Leadtek ti4200 64mb that's AGP 4x only and never had a problem.
It even worked with a Radeon HD3650.

Guld wrote on 2025-03-16, 17:35:

I do have a 4 channel scope and even have some chip clips, but I have no idea how to go about looking at what you are suggesting. Is this the same as the POST code displayed on my diag card? Would appreciate any more info.

No, you want to actually see if the CPU is fetching instructions from the ROM.

If you don't have a chip clip that fits the ROM (while in socket), you are
probably going to have to go 1 bit at a time (I'd probably start with just the
select, and add data bits once It's all working as expected.

So, I'd find the select pin on the ROM and scope that.
On a normally running PC - you should see LOTs of selects
to the BIOS ROM immediately after reset. This is where the
CPU starts, and will be all the code to initialize the system,
do POST, boot etc. (ie: >100,000 selects)

But we are only interested in the first few.

Once the scope is reliably triggering in selects, you can probe the data
pins to see what instructions it is fetching - you'll have to see each bit,
so be prepared to do this LOTs of times!

Depending on the storage depth of your scope you should be able
to figure out the first few instructions the CPU is fetching, and if you
have the BIOS data - see if this matches what's actually in the BIOS.

This will tell you if the CPU is getting as far as actually starting to execute
the BIOS, and if it's reading the BIOS ROM correctly.

Still a long way from solving the problem, but will help identify the most
fundamental problems and narrow down where you have to look.

Assuming it's actually executing the BIOS, you may be able to look at the
address lines/BIOS data to see where in the BIOS it's hanging...

DaveDDS wrote on 2025-03-17, 13:35:

No, you want to actually see if the CPU is fetching instructions from the ROM. […]
Show full quote

Guld wrote on 2025-03-16, 17:35:

I do have a 4 channel scope and even have some chip clips, but I have no idea how to go about looking at what you are suggesting. Is this the same as the POST code displayed on my diag card? Would appreciate any more info.

No, you want to actually see if the CPU is fetching instructions from the ROM.

If you don't have a chip clip that fits the ROM (while in socket), you are
probably going to have to go 1 bit at a time (I'd probably start with just the
select, and add data bits once It's all working as expected.

So, I'd find the select pin on the ROM and scope that.
On a normally running PC - you should see LOTs of selects
to the BIOS ROM immediately after reset. This is where the
CPU starts, and will be all the code to initialize the system,
do POST, boot etc. (ie: >100,000 selects)

But we are only interested in the first few.

Once the scope is reliably triggering in selects, you can probe the data
pins to see what instructions it is fetching - you'll have to see each bit,
so be prepared to do this LOTs of times!

Depending on the storage depth of your scope you should be able
to figure out the first few instructions the CPU is fetching, and if you
have the BIOS data - see if this matches what's actually in the BIOS.

This will tell you if the CPU is getting as far as actually starting to execute
the BIOS, and if it's reading the BIOS ROM correctly.

Still a long way from solving the problem, but will help identify the most
fundamental problems and narrow down where you have to look.

Assuming it's actually executing the BIOS, you may be able to look at the
address lines/BIOS data to see where in the BIOS it's hanging...

Interesting, thanks for explaining. I'm mostly familiar with my PCjr, but I assume the BIOS still starts at 0xF0000?

As noted above, when I use a power supply with a higher +5 V rating, I do now seem to get the beep codes from the system for no video card detected. (Updated some below, still spotty on different systems, seems to help, but not always).

Is that located at 0xC0000? And how would it behave if I have in a PCI card and an AGP card. I'm wondering if I can get it working enough to dump the AGP card VGA BIOS and see if it looks reasonable? Or maybe I'd have to write a quick .bat file to dump it to a disk at startup and work it that way.

Is there a good utility for dumping ROMs on later machines? I'm not sure if Brutman's utility I normally use would work on later machines or not. https://www.brutman.com/PCjr/pcjr_cart_dumping.html

shevalier wrote on 2025-03-15, 05:02:

When removing the cooler, SMD parts often break off. And it is in this place that very important resistor dividers are located, […]
Show full quote

Guld wrote on 2025-03-14, 16:42:

shevalier wrote on 2025-03-14, 11:35:

Carefully remove the heatsink from the video card and inspect the condition of the parts and tracks in the specified area on both sides. The last two contacts (one on each side) are responsible for selecting the AGP voltage.

Is there any way to test this to see if it's correct? e.g. is it just a certain resistance. Probably not but just curious if it can be tested to see if it has the correct "value".

Thanks for the suggestion, I'll take a look for anything suspicious.

When removing the cooler, SMD parts often break off. And it is in this place that very important resistor dividers are located, which are responsible for the AGP levels - 3.3 V (2x), 1.5 V (4x) or 0.75 V (8x)

https://elektrotanya.com/nvidia_geforce_2mx_n … f/download.html
page 4
signal
-"typedet" - A2
- AGPVREFGC - A66
-AGPVREFCG- B66

Okay, so I looked at my board and don't see anything that looks missing. I still need some time to look a little more. It does not match the circuit you referenced exactly, in that :

CORRECTED (See post below)
On the A side(A66) I found 4 resistors quickly, of the 4 1 goes to ground, the other to VDDQ.
In circuit the one between Vrefcg (B66) and VDDQ measured 1.5 kOhm (diagram says R111 1.5k)
In circuit the one between Vrefcg (B66) and ground measured 1.4 kOhm (diagram says R116 1.5k)
so...seems correct.

On the B side (B66) has one resistor between it and ground and one between it and VDDQ.
In circuit the one between Vrefgc and VDDQ measured 128 kOhm (diagram says R112 301kK)
In circuit the one between Vrefgc and ground measured 127 kOhm (diagram says R115 221K)
So, values here seem off, even the ratio between the two is different. ???

typedet (A2)
Connected directly to ground (diagram says R103 0R)
So....assuming 0R means (0 resistance, 🤣)...seems right?

I can draw a diagram if it helps.

Also, back to the power supply front.

I tried the card with my KT-7 RAID with the higher current +5 power supply and still doesn't work....but:
1) I sometimes get POST beeps for no video, sometimes not.
2) When I did not have the VGA cable connected, I even got it to beep as if it were booting a few times, but by the time I got my floppy connected and tried to see if it would boot and was actually working it stopped doing this, 🤣. If I plug in the VGA cable I still don't see video, even if it sounded almost like it was working.
3) The board does have some old caps on it that need to be replaced, it could also be that the board itself really needs to be recapped

Tried MSI K8T Neo with the higher capacitor PSU and still no improvement there.
Sometimes get beep codes, sometimes not.

Maybe the NV25 might have a higher power requirement than the NV28?

Was hoping the higher capacitor PSU would resolve some of this but it doesn't appear to make much difference.

Still no idea why it won't boot on the P3 systems at all.
According to my notes it did work on one of my P3 systems long enough to do 2.5 hours running the video memory stress test (It also crashed a lot doing benchmark testing, hence the reason for doing the stress test, so no, unfortunately it was not working correctly before I did anything to the board). I had forgotten about this as I've been trying to diagnose this issue for some time and it was way up in my notes, 🤣. Plus tired and not much time to test things.
What's changed since then I'm sure is a great question to ask?
I replaced the surface mount electrolytic caps. Didn't seem to have any problems doing that, and wasn't particularly difficult other than heating up the pads enough for the bigger ones. The only ones I didn't replace are some Sanyo 510 uF caps (4 of them, 2 are next to an inductor and appear shorted to my meter but appears normal). I took them out and measured them to be good with low loss and low ESR, tried replacing with 470 uF but got unstable video output afterwards, so I put back in the original 510's. Seems to be very much tuned to that odd value.

Last test I had time for tonight. I need to re-organize my notes.

P3 450 MHz system seems to boot okay with the higher current PSU, other than lack of video and the POST beep code for video error. Was able to get it to boot from floppy even though I can't see anything. So curious if this is an issue with the ROM being recognized correctly? I don't know why only on some systems though.

If I install a PCI card, I still get the missing video POST beep code. If I go in the BIOS and change it to init PCI first, then I can get it to boot with video from the PCI card. I was hoping HWINFO might tell me something about a second installed video card but didn't see anything.

Would be curious to try a ROM dump if there's a tool I can use for that that will dump ROMs even if it doesn't think they have a valid header/CRC?

Just a silly note, have you cleaned the graphic card contacts with IPA or something similar? That's strange HWiNFO could not see it, although sometimes HWiNFO does have that issue. Maybe try Astra http://www.sysinfolab.com/files/astra.zip

Edited my post above with the readings for resistors an A66/B66. It appears that UP on the motherboard is side A and down on the motherboard is side B. I had originally taken it as component side is A, but looking over everything that seems backwards.

Also added into about typedet.

Reading above, resistors on B66 to VDDQ and ground are different than shown in the diagram.

Zuldan, yeah, I've tried cleaning the connector and the entire board for that matter.

Astra output:

Video Card Video Chip Vendor: Advanced Micro Devices, Inc. (AMD/ATI) Video Chip: […]
Show full quote

Video Card
Video Chip Vendor: Advanced Micro Devices, Inc. (AMD/ATI)
Video Chip: Mach64 VT4 [Video Xpression+]
Video Chip Type: Unknown ATI Mach64 (22102)
VESA OEM String: ATI MACH64
Bus Type: PCI
Video Memory Size: 2 MB
Video Card 2
Video Chip Vendor: NVIDIA Corporation
Video Chip: GeForce4 Ti 4200 [NV25]
Bus Type: AGP 2.0 1x 2x 4x
Video Memory Size: 4 GB DDR2

Thanks for the suggestion. So yes, Astra reports info on both cards. Although I'm not sure how it's accessing the card to get the info exactly. But yes, it does see it although it does report the wrong amount of memory? Should be 64 or 128 MB (I want to say this card is 128MB), certainly not 4GB. Not sure if that's any indication of an issue or just Astra reporting incorrectly?