Score:0

How does AWS charge for Elastic Inference accelerator?

ng flag

I am trying to understand how AWS charges for Elastic Inference accelerators. The relevant page says:

you pay only for the accelerator hours you use

But I can see two possible readings of that:

  1. The hours that an EC2 instance with an EI accelerator attached to it runs.
  2. The hours that the EC2 instance actually accesses/engages the EI accelerator.

The difference can be significant: the EC2 instance can do some other stuff apart from running EI, and only engage it from time to time (even though it remains attached/accessible all the time).

The linked page gives two examples where the prices for EC2 and EI are simply combined, which suggests that option #1 above probably applies. But, it could simply mean that in those particular examples the EC2 instances run the attached EI accelerators non-stop all the time i.e. there is no time when the instance runs but the accelerator is idling.

Can someone please clarify how the charges work?

Score:1
gp flag
Tim

You're right, the page is imprecise. Based on what other services do, I expect that you pay for the time the accelerator is attached to your EC2 instance, as you have reserved that hardware for your exclusive use.

Greendrake avatar
ng flag
I'm not too sure about "reserved hardware for exclusive use". EI is a network-attached device which, presumably, leverages a shared pool of GPUs, so, when it is not used (albeit attached), the GPUs in the pool should be available for other EI accelerators.
Tim avatar
gp flag
Tim
Yeah. You really need to ask AWS this question. Billing support is free.
I sit in a Tesla and translated this thread with Ai:

mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.