It's a very standard construction in English. "Serverless GPU" means "GPU" that is "serverless". If you know what both words mean in the jargon, you know what they mean together. It's ok to not know what they mean, but arguing that it's "tortured" rings to me as misguided obtuseness.
It's a new buzzword coined by the marketing team at Amazon in 2014. Somewhat confusing here as you are renting time on a GPU server described as "serverless".
Also you're factually wrong. Amazon didn't invent this term and it was widely used before 2014. Here's a 2012 article using the term.
https://readwrite.com/why-the-future-of-software-and-apps-is...
I truly don't understand the bizarre and conspiracy-theory laden opposition to terms like this. It's a word, everyone who matters knows that it means.
I truly don't care and not opposed to the term. Just trying to explain why someone might get confused.
And sometimes sales people hijack a well-known, universally understood, industry term ("server") to make a product stand out ("serverless").
Serverless responding to http requests. Sure. I write some code. It gets fed data and returns data. I don't have to know how many cores the server has, or what microcode version the CPU is, or how many other things are running on the server, or if I'm writing an interpreted language (probably the case) even what architecture the CPU is.
But... I need to know all of that if I'm writing GPU code today.
I think it'll still be useful for plenty of people to choose which GPU runs their code, even if it's compatible with any GPU offered by the service. You might want to choose an older, cheaper GPU for basic parallel computation, but as supply catches up with demand for a newer and more energy-efficient model you'll then want to switch to that. There are only so many GeForce 4090s to go round :)
------
Recently, folks have begun to use the word "serverless" to mean something subtly different from its intended meaning in this document. Here are two possible definitions of "serverless":
Classic Serverless: The database engine runs within the same process, thread, and address space as the application. There is no message passing or network activity.
Neo-Serverless: The database engine runs in a separate namespace from the application, probably on a separate machine, but the database is provided as a turn-key service by the hosting provider, requires no management or administration by the application owners, and is so easy to use that the developers can think of the database as being serverless even if it really does use a server under the covers.
Cloud computing = someone's else's computer
Also, GPUs do other things than inferencing, right?
Perhaps "distributed scheduling and execution" is too straightforward than marketing wank.
(I can kind of understand it in case of CORBA - if you put the word "broker" in the name of your product, no one else will want to associate with it.)
GPUs don't contain servers. Serverlesses contain GPUs.