G80 is not worth repairing, at least if you plan to use it daily in retro setup. You can make it functional for brief benchmarks, but that's it.

Out of all Tesla family of chips, only something that has low TDP can survive for long and that's with a cooler that won't let GPU heat up beyond 70C. Or alternatively something that can be manipulated with direct undervolting, like original GTX260/280 and, most ironically, GTX295 Revision A.

hmmm god to know , so the G80 suffers from the Nvidia wafer fiasco also.... *I used hot iron on my M8700gt to temp-fix it, not seem to work on old G80... besides these 2 GPU's I have are old volt-modded than broken & de modded versions , so they got something burned from oc (that's how they ended up to me).

ediflorianUS wrote on 2023-05-15, 12:00:

Second option would be to attempt a repair on 8800GTXs I have , one use to boot w/artefact's , but today did not.

A legendary performance card with a legendary idiotic heat spreader.
Look for a new chip, remove the heat spreader from it, solder it to the board, adjust the cooling system for thermal contact.
Suicidal mission.

I find the Quadro 2000 to be good for Windows XP and very cheap these days. About the same power level as the 8800, but only like 62w. Still works well enough in Windows 10 also if that's your thing.

The precision(I'm on win7 64bit) use to have a quadro FX1500 series , and I sold that one to get the hd7750 , but If I Find one on the cheap(quadro 2000 , not listed on ebay so no shipping is involved) ,I will upgrade the FX550 ... (this is flaky... may have a memory chip with improper contacts).

BTW Nice photo of the G80 , I never knew it had a heat spreader on the chip , I always remembered it was the chip itself (gray).

ediflorianUS wrote on 2023-05-15, 18:25:

I always remembered it was the chip itself (gray).

Gray and fried. Alas.
These chips are all on thermal paste, not solder.

These images seem to sugest tiny solder balls (BGA) .... not paste .... oh well who knows.

ediflorianUS wrote on 2023-05-16, 07:55:

These images seem to sugest tiny solder balls (BGA) .... not paste .... oh well who knows.

Thermalpaste, not solder paste 😉
For example, some P IV
Low-melting solder (gallium-based)

"Finaly found a upgrade GPU for the old Dell Precision 490"

update , I buyed sunday a (partially)broken Quadro K4200 , 4gb gddr5 ,however, it gave a code 43 on the gpu under windows , and some horiz lines(after driver install) ....
Cleaned the whole thing , after dissasambly , then hot ironed evrything (gpu+ram chips) but had no effect , no progress was made.
Eventually I managed to find the bios and Kepler bin editor , so I used that and just keep lowering the frequencies till it was stable enough to not give code43. (rewrote the firmware/gpu/bios/bin 4-5 times)....
I got it to a point I can load CS:S (HL2 engine) , at 640x480 or 1024x768 ... but in other 3d it just gives me , blank screen for a moment, and sometimes it starts most of the times it goes out of the 3dx app...
- I am stuck , not sure what else to do , voltage is back to stock (in .bin file , and I gave a uper 0.025v room in case it needed)
How do I check if it's a ram module? or the GPU ? causing evrything? (to narrow down the problem , it also has some laptop style VRM/MOSFET's - would one of them be acting-up?

ediflorianUS wrote on 2023-11-02, 15:58:

How do I check if it's a ram module? or the GPU ?

https://github.com/ihaque/memtestCL
Works great on Radeons. At nVidia - it should be too, besides there is a CUDA version.

1Test iteration 5 on 1500 MiB of memory on device 0 (Pitcairn): 0 errors so far
2        Moving Inversions (ones and zeros): 0 errors (156 ms)
3        Moving Inversions (random): 0 errors (156 ms)
4        Memtest86 Walking 8-bit: 0 errors (1266 ms)
5        True Walking zeros (8-bit): 0 errors (625 ms)
6        True Walking ones (8-bit): 0 errors (625 ms)
7        Memtest86 Walking zeros (32-bit): 0 errors (2531 ms)
8        Memtest86 Walking ones (32-bit): 0 errors (2516 ms)
9        Random blocks: 0 errors (203 ms)
10        Memtest86 Modulo-20: 0 errors (12031 ms)
11        Logic (one iteration): 0 errors (78 ms)
12        Logic (4 iterations): 0 errors (78 ms)
13        Logic (local memory, one iteration): 0 errors (78 ms)
14        Logic (local memory, 4 iterations): 0 errors (79 ms)
15
16Test summary:
17-----------------------------------------
185 iterations over 1500 MiB of memory on device Pitcairn
19      Moving inversions (ones and zeros): 0 failed iterations
20                                         (0 total incorrect bits)
21                 Memtest86 walking 8-bit: 0 failed iterations
22                                         (0 total incorrect bits)
23              True walking zeros (8-bit): 0 failed iterations
24                                         (0 total incorrect bits)
25               True walking ones (8-bit): 0 failed iterations
26                                         (0 total incorrect bits)
27              Moving inversions (random): 0 failed iterations
28                                         (0 total incorrect bits)
29             True walking zeros (32-bit): 0 failed iterations
30                                         (0 total incorrect bits)
31              True walking ones (32-bit): 0 failed iterations
32                                         (0 total incorrect bits)
33                           Random blocks: 0 failed iterations
34                                         (0 total incorrect bits)
35                     Memtest86 Modulo-20: 0 failed iterations
36                                         (0 total incorrect bits)
37                           Integer logic: 0 failed iterations
38                                         (0 total incorrect bits)
39                 Integer logic (4 loops): 0 failed iterations
40                                         (0 total incorrect bits)
41            Integer logic (local memory): 0 failed iterations
42                                         (0 total incorrect bits)
43   Integer logic (4 loops, local memory): 0 failed iterations
44                                         (0 total incorrect bits)
45Final error count: 0 errors
46
47Press <enter> to quit.

shevalier wrote on 2023-11-02, 16:32:

ediflorianUS wrote on 2023-11-02, 15:58:

How do I check if it's a ram module? or the GPU ?

https://github.com/ihaque/memtestCL
Works great on Radeons. At nVidia - it should be too, besides there is a CUDA version.

tnx I wil try-it