No extension cable, it's on mobo.No PCI-E power extension cables?
I'm asking all these questions because i've literally burned, i mean BURNED, PCI-E extension cables and risers with my 3090 - they can move a LOT of power.
I've ran GPU-Z and logged everything without issues. Temps are looking good. I can do anything because it does not crash the whole system, just the game itself. There is no undervolt applied. Thermal pads were in terrible condition when I opened the card (pics attached). I've tried all 3 BIOS listed on Gigabyte's website, not sure if they differ from the ones on TPU. Cables are not daisy chained and power supply is one of the best in the market. I don't think baking would help with this anyways (maybe I'm wrong but I won't do it).Are you able to run GPU-Z and log the memory junction temperatures right until the point it crashes? If your card is running the VRAM at 108-110 range, expect it to be quite moody indeed. I've read reports of some Gigabyte cards having their factory thermal pads dry out because of the extreme heat stress. Any undervolting applied to it? Sometimes the GPU voltage you used cannot sustain the memory, even though otherwise it would be stable, i.e. I can do 1755 core +800 memory at 793 mV and 1890 core +1000 memory at 837 mV, but with 793 mV, it crashes almost instantly at 1755 core +1000 memory. You may also have a bad VBIOS installed on it, the Gigabyte Gaming OC's latest ReBAR 390W BIOS works great on my TUF OC, and should work great on your actual Gaming OC as well. Try a clean flash, downloading the right file from TPU.
FurMark is not useful for testing, not because it can kill cards (hardware power limiter and thermal throttling will kick in long before any real damage occurs), but because the graphics driver will detect it and further throttle the clocks. I've seen mine do low 1100s running it, it's not useful at all. Running Port Royal and Time Spy Extreme stability tests, as well as Unigine Superposition and the Final Fantasy XIV Endwalker/XV Windows Edition benchmarks work great for stability testing.
Like Mussels said, the RTX 3090 is an exceptionally hungry video card. Don't run risers, cheap custom cables, cable extenders and the sort and make absolutely sure your power supply can handle a strong current in its 12V rails with generous overprovisioning, another very common oversight - don't daisy chain and run one cable for both of your 8-pin connectors! Just don't.
And for the love of God, don't chuck your $2000 video card in the oven. That can and most definitely will place great stress on the entire PCB, and might cause damage where otherwise was fine.
I'll try and run the other benchmarks you said.
I've changed the pads, you can see attached pics of what it looked like after opening the card and seeing Gigabyte's pads...Normally with the 3090, Tjunction reports the hottest memory chip - its also possible here that the sensor is not in the location that's overheating and a teardown with new thermal pads is the logical next step
Oh yeah, i totally missed using the daisy chain or pigtail power cables: you need two seperate cables from the PSU, not one shared one.
I've tried underclocking both core and memory - does not help. Tried lowering power limit as well and it doesn't help.With the issue of Games crashing on launch.
I would like you try down-clocking your GPU core clocks and maybe VRAM clocks has well.
See if the games launch with these lower clock settings.
Try reference 3090 clocks or even lower
What's interesting is that I've never seen consumption of this card go above 240W (readings from GPU-Z) no matter what stress test I run.
Just confirmed that I'm already running ReBar 390W VBIOS - 94.02.42.40.A2