I agree with you man but I’m really broke right now and the prices are all over the place... Getting the same performance out of another card will cost me upwards of 150€... People are selling 580s for 220€ and 1080tis for 450€... Can you recommend a memory test application? Also by vrm shoulders you mean the actual chips? I can do that no problem, like I said I have a bga rework station...
Well, in my opinion, time and effort would be significantly more than the numbers you mentioned.
For hardware diagnostics, I would
- check resistances on all power buses (pcie, pcie controller, memory, gpu, etc..) and see if something catches your eye.
- Check individual shoulders (involves removing coils from every other shoulder's output as they are interconnected) looking for a dead horse. If you have a thermal imaging device (flir for example) you can check and see which phases are working and which are not without taking stuff off, just by heat distribution.
- On individual phases check for proper impulses and levels with scope.
- Check the controllers and couplers for proper signals
For memory tests, there is specialized software but it is not available to consumers. Instead, you can try Kombustor with artifact scanner (for example), that will test some memory.
Besides memory and power delivery issues, since those boards run hot, there might be partial IC separation (either memory or the GPU or the GPU die... those can be solved by reballing. Also might be a few caps degraded in the decoupler circuitry, causing power starvation or dumps that cause crashes... and that will take a lot of time to find, even when having a boardview for this exact board version and revision (pm-d you a link for the one I have access to).
Good luck.