Just thought it would be good to copy paste this
Card makers respond to RTX 30-series capacitor controversy; MSI quietly alters its RTX 3080 designs
A hot potato: The RTX 3080/3090 crashing issues that are likely related to the type of capacitors used on the cards have seen MSI stealthily change the design of its RTX 3080 products. Several vendors have given statements regarding the situation, though MSI has remained silent.
For those who don’t know, we’ve seen numerous reports of factory overclocked aftermarket RTX 3080 and RTX 3090 cards suffering from crashes and stability issues when they reach or exceed 2.0 GHz. It’s thought that the problem is due to the capacitors found on the rear of the PCB underneath the GPU.
https://www.techspot.com/news/86916-card-makers-respond-rtx-30-series-capacitor-controversy.html
Doesn’t appear to be the case…Igor’s hardware was probably wrong, although he never said he was 100% right in the problem’s diagnosis.
This follow up video from hardware unboxed pretty much sums it up it seems:
TLDR - appeared to be a driver optimisation issue. Even the cards with the preferred capacitor setup were still crashing. Interestingly, the 3080 cards weren’t crashing when used on Linux, which obviously removes it from being a hardware related issue. Nvidia’s updated driver fixes the issues with almost zero performance loss.
It’s really a non-story.
I’d be more inclined to say that major thermal problems from the 30xx cards is the issue. NVidia did a very poor job of cooling these cards.
In the case of 3080’s and 3090’s, the VRAM runs really hot. At 80C, it will cause throttling to the detriment of whatever it is you’re doing, especially GPU-intensive tasks such as rendering and crypto mining. You can literally fry an egg on the backplate.
And GPU temperature is not the same as the TJunction temperature. It’s commonly 20 degrees cooler than the junction. So if your GPU is at 80, your VRAM is frying itself at 100.
And if you overclock without knowing how to set fan speeds, undervoltage, memory cycles, etc., you will definitely crash your card (and void your warranty). Consider using HWINFO. It’s a good GPU monitoring program used to check the temperature at the VRAM Tjunction. (something that NVidia actually uses but neglects to show customers despite it being critical information).
The solution is to water cool or to add extra heat sinks/thermal pads at the GPU and backplate. The 17 w/mk Fujipoly 85x45 size pads are best. There are Youtube videos to guide you through this.
And if you spent the big bucks lately for one of these cards, you’d be well-advised to look into this.
