If you are saying that voltages to vcore and vram were way off before you probably have a bad chip now.

I purchased a RX 570 a while ago that was broken. The machine boots with the card in, drivers load properly, 2d video and video acceleration work fine, but if you try to do anything 3d it pukes all over the screen and gets hot. I took the cooler off and found a hole in the die where something in the 3d section of the chip blew up (and the 3d section is most of the chip die).

mats for dos fully compatible with NV40

you card is based on DDR1 ram which may fail very often. use mats to detect bad chip and reflow it.

if there is no memory errors underclock gpu to 300mhz and see if it solve the issue.

ps.
just swapped the core on another 6800gs to 6800gt because of failed gpu. but there was a full mess on the screen.

Mats version 4.173.1 works with NV40: Re: Broken GeForce FX5950 Ultra - chip transplant

Hi,

Thank you so much for your pieces of advice. I will try this mats version ASAP 👍

The die itself, at least the back side, is physically in good shape. No crack, nor burnt area.
Moreover, the defective MOSFETs were on the RAM voltage side. Nothing wrong on the buck converter for the GPU core. But for sure I can't exclude bad solder joints on GPU core...

From your experience, it's possible to have a fine 2D display on Windows while having garbage in d3d apps, only related to faulty RAM chip(s) ? Or is it likely to be related to the GPU core?

JB

memory fault in banks C or D may cause 2D work fine and 3D crappy

Hi,

Some news of this 6800GT...
I tried MATS, following the process detailed in the link provided by Karbist.
There are absolutely no issue with the RAM banks, no faults.
I used MATS with several memory sizes (10, 20, 60, 128MB), no issue with the RAM.
I also tested different algorithms (there are 8 of them), still no problem with the RAM banks.

Yet I noticed that on Windows XP, the card is well detected and identified, all seem to go the right way but after some minutes, big chunks appear on the screen and within a second it goes black with a system freeze.
Something tells me that it has to do with the GPU, some bad solder joint or so... Anyway, I will try to "hairdry" it, maybe the behaviour could change...

To be continued 😉

Check 2 capacitors in memory rail, they're 2x470mf aluminium caps.
replace with 2x1200(1500) solid 4v capacitors.

it is a well known issue, not enough capacity in memory rail causing issues with corrupted 3d or even lost voltage (black screen)

ps.

missed this is GT, no this caps. they're on NON-GT version.

ASUS has full VRM for GPU and MEM. so probably GPU needs to be replaced -(

you can try to declock memory and GPU using rivatuner and get what exactly is failing.

I'd give it a thorough wash and dry.
Dishwasher no soap, or sink, toothbrush and dishsoap.

Sun dry about 4 hours in full sun, or 3 days inside, or 15mins 90c oven.

Fixed a bunch of things with a simple wash.

Hi,

I started with a good water cleaning + dishsoap, rinced it with demineralized water and finally "baked" it 30min at 80°C.
The PCB is now nice and clean... and it's still working! But nothing better than before unfortunately: imposible to launch 3D stuff without freezing.

A guy told me to do the same thing as dm- : decklocking. And here we have interesting things...
I installed rivatuner and began to fiddle with core and memory frequencies.
Stock configuration is 350MHz for the core and 700MHz for the RAM (350MHz clock and 700MHz effective DDR frequency?)
I tried to reduce them by 50MHz but the rivatuner test failed. Minus 50MHz, failed again...
Finally 200MHz for the core and 450MHz for the memory was ok and with these settings, I'm now able to run 3D tests without freezes.
I refined the settings and finally found that the core frequency can be set at its max value, 350MHz without any problem. It can even be pushed a bit to 360MHz.
On the other hand, rivatuner test refuse to pass with RAM frequency above 470MHz (whereas it should be 700MHz!!).
In these conditions, 3D scores are a bit behind a vanilla 6800.
The good news is that the core seems to be in good shape!

RAM chips are Hynix HY5DU283222AF-22, which correspond to 2.2ns DDR, able to work at a clock frequency of 450MHz (900Mbps/pin) @ 2.8V according the datasheet.
I'm confused with frequency in rivatuner: is it the clock or the "effective" frequency ? I think it's the effective one because it reports 700MHz (i.e. 350MHz clock freq ?)
That would mean the actual clock frequency for which it works is 450/2 = 275MHz ?!! Whereas it should work fine at 450MHz clock frequency!!!

I also extracted the BIOS with nvflash and checked it with NiBiTor: it's perfectly fine and bit-to-bit equivalent to one of the two 5.40.02.15.00 found here: https://www.techpowerup.com/vgabios/?architec … memSize=&since=
So it shouldn't have been modified to overclock the RAM.
I'm not convinced by a thermal issue on the RAM chips beacause they stay cool and I remind you that MATS memory tests have always passed successfully.
Maybe a RAM VRM issue... From the beginning I'm using an old (but still good) low ESR electrolythic cap to replace the bulgy one, at the RAM VRM output...

NB : I'm using 91.31 drivers but same things happen with older 71.89 drivers.

What do you think ?

JB

so this is DDR1 GT version? show me the picture.
if it is a DDR1 version then the caps in memory rail must be replaced.

This one :

V9999GT/TD/P/128M/A

2 brown capacitors, one is already bulged
replace with 1200 or 1500 4v solid state caps

Actually, I've already replaced these caps (this is an old picture). At this moment, there is only one electrolytic with low ESR, but this is the next step to replace both by new ones.
One of the two transistors on this synchronous buck stage (RAM VRM) was burnt. Both have already been replaced.

What do you mean by ''solid state'' capacitor? tantalum?

https://eu.mouser.com/ProductDetail/KYOCERA-A … ex6aEhMKg%3D%3D

for example