This is not correct. Solid-state parts wear out like mechanical parts, especially when you run them hot. The mechanism of this wear-out comes from things like electromigration of materials and high-energy electrons literally knocking atoms out of place.
That is not correct when wires are small and you run things hot. The physics of <10 nm transistor channels is very different than it is for 100 nm+ transistors.
Your assertions are at odds with modern computers - including Nvidia datacenter GPUs - still working fine after many, many years. If not for 1. improved power efficiency on new models and 2. Nvidia'a warranty coverage expiring, datacenters could continue running those GPUs for a long time.
Which GPUs have been running for decades that you're referring to? The A100s that are 4 years old? MTBFs for GPUs are about 5-10 years, and that's not about fans. AWS and the other clouds have a 5-8 year depreciation calendar for computers. That is not "decades."
You can keep a server running for 10-15 years, but usually you do that only when the server is in a good environment and has had a light load.
> Which GPUs have been running for decades that you're referring to? The A100s that are 4 years old? MTBFs for GPUs are about 5-10 years, and that's not about fans.
I said solid state components last decades. 10nm transistors have a thing for over 10 years now and other than manufacturer defect don't show any signs of wearing out from age.
> MTBFs for GPUs are about 5-10 years, and that's not about fans.
That sounds about the right time for a repaste.
> AWS and the other clouds have a 5-8 year depreciation calendar for computers.
Because the manufacturer warranties run out after that + it becomes cost efficient to upgrade to lower power technology. Not because the chips are physically broken.
