Karbist wrote on 2024-03-10, 12:00:
So I put the card on the pre-heater, applied hot air and chip was floating right at 190C.
ran furmark for few minutes and card reached 80C (probably need a better paste) and it was stable, no artifact .
Nice, great to hear you got it revived! 😀
Better paste / thermal compound won't help you lower the temperatures that much though - maybe 4-5C max, which is still too much for these cards, especially one that has failed already.
Again, my recommendation is to keep temperatures below 60C (preferrably in the mid-50's max)... which won't be possible with the stock cooler.
Replacing it with something like the Zalman VF-1000 is a good start... though those were (and still are) rather pricey... not to mention even difficult to find nowadays (unless you look for a broken video card on eBay that has this cooler - some MSI GeForce 8800 GT cards did, among others.) Or taking a cooler off of another card and adapting it for the GF 6800 is another option if you're handy / don't mind DIY. Just needs to be a cooler from a card rated for 120W TDP or more.
Socket3 wrote on 2024-03-10, 13:50:
I'll give it a shot when I get a pre-heater - they're not expensive.
Pre-heaters are somewhat of a waste of money, IMO. You can easily make one from the components of a heat gun or an old toaster. I was going to go that route, until I found that the kitchen (gas) stove is actually better than any pre-heater I could buy - lots of power (heat) and a lot of room for adjustment, allowing me to do both small and big PCBs. I've also done it on an electric stove, but it can be a bit more tricky depending on what you have. "Modern" IR-based electric stoves with touch-sensitive buttons are a bit of PITA to use. But anything of the older kind with an "exposed" coiled heating element (like one found in the oven) and a proper knob for variable power makes for a pretty good pre-heater.
Just place the PCB on blocks of wood or other object to suspend the PCB a few inches (maybe 3-7 cm) above the stove, and that's it. Need to have a type-k thermocouple meter, though, to monitor PCB temperatures. With that and hot air station, you should be able to do quite a bit.
shamino wrote on 2024-03-09, 01:20:
All it takes is a bad wire. How much current will cause a problem depends on how bad it is.
Whether you'd see a melting connector also depends where the problem is.
If it's a bad wire, either the card won't start (detects molex connector is not present or simply drops out due to disconnection of power for GPU V_core) or will drop out intermittently (again, for the same reasons.) Whichever the case, this type of failure is usually very unpredictable in nature... whereas in the case of O/P's card, the failure seems to be much more predicable - crashing only under high GPU/mem. load and after the GPU has ran for a while / reached a certain temperature.
Bad solder joints on the PCB are out of the question too, as this is a multi-layer PCB and the power connector is soldered to multiple layers with vias connecting all of the power traces together.
So that only leaves weak contacts on the connector... which might loose a few 100 mV at most (along through the wires), but hardly more. At 4-5 Amps current draw (about 60W for these 6800 GPUs), any more would start producing a lot of heat around the connector. At 1V drop, that's 4-5W of power. But even if that was the case, 11V on the 12V rail won't crash these 6800 cards - they will happily run down to even 10V. If the connector was dropping 2V, you'd see it.