It's often the most cost effective way to get redundant and A+A power supplies in servers that you need to protect beyond simple UPS methods.
Source: Loving the 2U and 4U blades with 2017-2020 era EPYC and Xeon processors with dirt-cheap Tesla P40 24GB VRAM GPUs in them for inference and all sorts of compute loads.
https://www.ibm.com/common/ssi/ShowDoc.wss?docURL=/common/ss...
Then by that definition, virtually every blade server I've ever seen was 2U :)
> Small servers like that are also very unlikely to be loaded up with high-watt GPUs for inference or training workloads.
True, though I've seen full-size GPUs get squeezed into form factors much smaller than the average 2U-high blade.
Huh?
Rack Us absolutely refer to a form factor not just a z-height. It would make no sense to refer to something in # of Us unless you either want to then mount it in a rack, or be deliberately annoying.
Most people (including, typically, the companies actually manufacturing and selling blades and enclosures) would also in this context reasonably infer "2U or 4U blade" to be referring to a (partially-or-fully-)filled enclosure, rather than pedantically assume that someone (who is implied to own and use such a server) doesn't know the difference between a blade and a pizza box - but given that we've already deviated from using terms the way anyone else uses them, what's one more deviation, right? :)
The entire point of blades is that they’re modular and a chassis can be filled with many different kinds of blades. Calling an enclosure a blade would be like calling a tire a car.
Source: I’ve specd out and ordered literally thousands of blade servers and hundreds of enclosures from every major vendor.
Edit: an even better simile would have been people who call a whole desktop computer a cpu.