zuldan wrote on Today, 10:53:

I’ve managed to find the identical 4ns chips. 2 weeks away. I’m also having a custom stencil cut for the NV15 (an off the shelf […]
Show full quote

Trashbytes wrote on Yesterday, 05:01:

I had a look myself, seems like Alibaba China might be one of the few places with these specific chips available in the 4ns version...assuming you want to take that gamble.

I’ve managed to find the identical 4ns chips. 2 weeks away. I’m also having a custom stencil cut for the NV15 (an off the shelf one doesn’t exist) in case I have to reball the existing chip. I’ve also ordered some NV15 GF2 Ti chips in case the original NV15 GF2 Ultra is buggered. They should be drop in replacements. I’m determined to get this card working 😂

Trashbytes wrote on Yesterday, 05:01:

If you can find a Datasheet for these chips and you have a Oscilloscope you could check the data lines on each IC, my guess is the dead ones will have issues on the data lines which could help you find the bad ones.

I think the data would be so random on those lines it would be difficult to see something “odd”. I’m going to remove 1 chip as see what Mats reports.

That's the idea, itll either not be there or garbled, you shouldn't have any issue seeing a normal data pattern from the bad ones.

Trashbytes wrote on Today, 10:58:

That's the idea, itll either not be there or garbled, you shouldn't have any issue seeing a normal data pattern from the bad ones.

I could probe a few data lines but I got no idea how they work. Does a data line do read and write or only certain ones do writes. To probe them properly without shorting something I would need a AGP riser cable. The one I have isn’t great. Most AGP cards don’t work with it. If you know of any, let me know.

zuldan wrote on Today, 10:53:

I’ve managed to find the identical 4ns chips. 2 weeks away. I’m also having a custom stencil cut for the NV15 (an off the shelf […]
Show full quote

Trashbytes wrote on Yesterday, 05:01:

I had a look myself, seems like Alibaba China might be one of the few places with these specific chips available in the 4ns version...assuming you want to take that gamble.

I’ve managed to find the identical 4ns chips. 2 weeks away. I’m also having a custom stencil cut for the NV15 (an off the shelf one doesn’t exist) in case I have to reball the existing chip. I’ve also ordered some NV15 GF2 Ti chips in case the original NV15 GF2 Ultra is buggered. They should be drop in replacements. I’m determined to get this card working 😂

Trashbytes wrote on Yesterday, 05:01:

If you can find a Datasheet for these chips and you have a Oscilloscope you could check the data lines on each IC, my guess is the dead ones will have issues on the data lines which could help you find the bad ones.

I think the data would be so random on those lines it would be difficult to see something “odd”. I’m going to remove 1 chip as see what Mats reports.

This is not as unhelpful as it might sound. On my last 9500 pro repair i could identify the data lines not getting data. It's usually all or nothing. So launching the card in an OS with constant output and then probing the data lanes can already tell you a bit. You don't need to understand the data, but check if the pattern is similar and if there are some that look off.

Instead of removing a whole memory chip, go and remove a 4pack series termination resistor network. It should translate 1to1 with 4 bits failing.

tehsiggi wrote on Today, 11:24:

This is not as unhelpful as it might sound. On my last 9500 pro repair i could identify the data lines not getting data. It's usually all or nothing. So launching the card in an OS with constant output and then probing the data lanes can already tell you a bit. You don't need to understand the data, but check if the pattern is similar and if there are some that look off.

In that case it shouldn’t be hard to determine the bad chips (good idea Trashbytes!). How are you positioning the card when testing? I don’t trust my hands to be that steady when the card is plugged directly into the motherboard. My AGP riser cable is perfect. Just need one that works without degrading the voltage.

tehsiggi wrote on Today, 11:24:

Instead of removing a whole memory chip, go and remove a 4pack series termination resistor network. It should translate 1to1 with 4 bits failing.

Much easier than removing a chip. Going to remove this guy first.

zuldan wrote on Today, 11:56:

In that case it shouldn’t be hard to determine the bad chips (good idea Trashbytes!). How are you positioning the card when test […]
Show full quote

tehsiggi wrote on Today, 11:24:

This is not as unhelpful as it might sound. On my last 9500 pro repair i could identify the data lines not getting data. It's usually all or nothing. So launching the card in an OS with constant output and then probing the data lanes can already tell you a bit. You don't need to understand the data, but check if the pattern is similar and if there are some that look off.

In that case it shouldn’t be hard to determine the bad chips (good idea Trashbytes!). How are you positioning the card when testing? I don’t trust my hands to be that steady when the card is plugged directly into the motherboard. My AGP riser cable is perfect. Just need one that works without degrading the voltage.

tehsiggi wrote on Today, 11:24:

Instead of removing a whole memory chip, go and remove a 4pack series termination resistor network. It should translate 1to1 with 4 bits failing.

Much easier than removing a chip. Going to remove this guy first.

The attachment IMG_0497.jpeg is no longer available

There is life in the old girl, the fact she posts and can be tested is a good sign she can be repaired, gently goes and you'll have a fully working card and plenty of new knowledge for the next card that needs saving !

Trashbytes wrote on Today, 12:02:

There is life in the old girl, the fact she posts and can be tested is a good sign she can be repaired, gently goes and you'll have a fully working card and plenty of new knowledge for the next card that needs saving !

Yep I’m sure I can get her going again. I won’t have another opportunity to get this sort of card again.

For reference, here is a video showing the memory issue in dos https://www.youtube.com/watch?v=QTts1uvix_E

You mentioned that you reflowed all memory IC pins and that, by doing so, "the error count has now been halved".

This sounds like an important clue to me. It's very hard to imagine that there were two issues to begin with (a few bad solder joints + a dead memory IC). Are you 100% sure that the reflow was successful and, more importantly, that you did not leave any solder bridges behind?

bloodem wrote on Today, 12:16:

You mentioned that you reflowed all memory IC pins and that, by doing so, "the error count has now been halved".

This sounds like an important clue to me. It's very hard to imagine that there were two issues to begin with (a few bad solder joints + a dead memory IC). Are you 100% sure that the reflow was successful and, more importantly, that you did not leave any solder bridges behind?

Very good point about solder bridges .. all it takes is one bridge for a memory IC to fail testing.

bloodem wrote on Today, 12:16:

You mentioned that you reflowed all memory IC pins and that, by doing so, "the error count has now been halved".

This sounds like an important clue to me. It's very hard to imagine that there were two issues to begin with (a few bad solder joints + a dead memory IC). Are you 100% sure that the reflow was successful and, more importantly, that you did not leave any solder bridges behind?

I'm 200% sure the reflow was successful and there are no solder bridges. I have a stereo microscope and actually spent a few days before the reflow "binning" DDR chips so it's not my first time (Re: Socket A: Aiming for the Stars!!!).

I agree with you. It's strange the errors reduced. Bad solder joints plus bad chips is pretty rare. Before and after logs below.

zuldan wrote on 48 minutes ago:

I'm 200% sure the reflow was successful and there are no solder bridges. I have a stereo microscope and actually spent a few day […]
Show full quote

bloodem wrote on Today, 12:16:

You mentioned that you reflowed all memory IC pins and that, by doing so, "the error count has now been halved".

This sounds like an important clue to me. It's very hard to imagine that there were two issues to begin with (a few bad solder joints + a dead memory IC). Are you 100% sure that the reflow was successful and, more importantly, that you did not leave any solder bridges behind?

I'm 200% sure the reflow was successful and there are no solder bridges. I have a stereo microscope and actually spent a few days before the reflow "binning" DDR chips so it's not my first time (Re: Socket A: Aiming for the Stars!!!).

I agree with you. It's strange the errors reduced. Bad solder joints plus bad chips is pretty rare. Before and after logs below.

The attachment Before memory reflow.TXT is no longer available

The attachment After memory reflow.TXT is no longer available

I wouldn't give too much about the change in the error amount. The affected bits are still the same.