Retronerd878 wrote on 2026-05-21, 20:06:

I measured the IC's on the front of the card. Anything suspicious?

Everything's fine
And, if a donor is available, replace the logic ICs (circled in yellow).

Checked the other regulators on the back and they seem to be fine.

Guess I need to get to transplanting

One of the logic ICs receives a RESET signal from the AGP bus.
The other is a buffer for the monitor’s synchronisation signals.
However, given that the configuration strap-resistors are connected to these signals, if this chip fails, it is likely to affect them.

And I’ve run out of ideas, sorry.

I measured the oscillator and the frequency is correct, but isn't that wonky waveform bad?

Retronerd878 wrote on 2026-05-30, 20:51:

I measured the oscillator and the frequency is correct, but isn't that wonky waveform bad?

50/50
It seems more likely that the probe is set to x10, but the oscilloscope isn't configured to scale to that range.
Also, the trimmer capacitor (which adjusts only the x10 range) on the probe (near the BNC connector) hasn't been adjusted.
If everything is set up correctly, then either the oscillator is dead, or the GPU pin is clipping the signal.
At 0.5V peak-to-peak, the 3.3V logic levels won’t switch.

Such signals should only be measured using a probe set to the x10 range (which ‘bypasses’ the coaxial cable’s capacitance) fitted with a spring clip, not a crocodile clip.

The probe was set to 10x.
The oscilloscope was indeed set to 1x and changed to 10x.

Same result. Changing the Coupling mode from DC to AC also yields the same results.

This isn't the 9800 circuit, but it's very similar to that implementation.
Page 4.

The oscillator is powered by 3.3V, so the peak-to-peak voltage cannot exceed 3.3 V.
Apart from narrow spikes on the fronts, which may exceed the 0–3.3 V range due to inductance.
The fact that the waveform doesn’t match the “textbook” is because that’s how it always looks in reality, especially if it isn’t tested with a ground spring and probe isn’t tuned.
If you’re absolutely certain about the measurements, then you’ve got a donor.
Just swap the oscillator.

P.S. It’s possible that the 9800 model uses a 5 V generator, but this is easy to check using a multimeter on its terminals.
But I have serious doubts about that.
The oscillator on the Radeon 9800 (9700, PCB-style) is definitely connected to the +3.3V supply; I’ve checked.

PPS Try using the same oscilloscope settings to measure either a CR2032 battery or the +3.3V power supply voltage.

Mounted the faulty card and measured the oscillator.
The curve looks better but still not a perfect sinewave

I also have a working card and measure the oscillator. It looks the same as the one under test. So, I'm guessing the oscillator is good and the measurement is not in perfect conditions.

It could be either a sine wave or a meander. It depends on the oscillator itself.
However, if you have 27 MHz at the V_supply/2 level, then that is most likely not the problem.

I replaced the IC's you marked in yellow and no improvement.
Booted to windows XP with a PCI card and the Radeon is detected in HW Info, but what's interesting is that the video memory quantity is not detected and the bus width is only 64-bit.

Retronerd878 wrote on Today, 09:42:

I replaced the IC's you marked in yellow and no improvement.
Booted to windows XP with a PCI card and the Radeon is detected in HW Info, but what's interesting is that the video memory quantity is not detected and the bus width is only 64-bit.

TC7SZ08FU
https://toshiba.semicon-storage.com/info/TC7S … dName=TC7SZ08FU

I would check the four logic ICs with an oscilloscope.
They receive the AGP_reset signal from the 74ACT08, which buffers it from the AGP port.
The CKE signal is applied to the second input.
The logic is as follows: whilst AGP_reset is low, CKE on the VRAM is also low, regardless of what the GPU is outputting.
When Reset is released (pin 2), the output of this IC (4) equls the input (pin 1).

If that doesn’t work, run RMID from DOS and test the memory.
Either using a bat-file in blind mode with logging, or via a second graphics card.
Re: Hercules Radeon 9700 Pro: weird output

shevalier wrote on Today, 16:14:

TC7SZ08FU https://toshiba.semicon-storage.com/info/TC7S … dName=TC7SZ08FU […]
Show full quote

Retronerd878 wrote on Today, 09:42:

I replaced the IC's you marked in yellow and no improvement.
Booted to windows XP with a PCI card and the Radeon is detected in HW Info, but what's interesting is that the video memory quantity is not detected and the bus width is only 64-bit.

TC7SZ08FU
https://toshiba.semicon-storage.com/info/TC7S … dName=TC7SZ08FU

I would check the four logic ICs with an oscilloscope.
They receive the AGP_reset signal from the 74ACT08, which buffers it from the AGP port.
The CKE signal is applied to the second input.
The logic is as follows: whilst AGP_reset is low, CKE on the VRAM is also low, regardless of what the GPU is outputting.
When Reset is released (pin 2), the output of this IC (4) equls the input (pin 1).

If that doesn’t work, run RMID from DOS and test the memory.
Either using a bat-file in blind mode with logging, or via a second graphics card.
Re: Hercules Radeon 9700 Pro: weird output

I had defective Radeon 9700s with the exact error and the card was not listed as 64bit (which would mean 3 out of 4 are bad).
R3MEMID appears not to work with the card as a secondary adapter. So unless the system boots and just not displays an image, there is no way to leverage it 😒