The subjecct is Huge and it could be a hard walk to discuss about what is and what is not USX / NOS. In my opinion, USX is a cool solution. It come from VDI land where storage was and is a a problem. At the beggining, it was designed to accelerate storage, manly by using host RAM / SSD to accelerate
So it's pure software too. Now it comes in a box (since 2014 as far as I know) as an hypervisor.
Nutanix come from Webscale and was designed at the very begining as a webscale platefrom for those who are not from the big five. It was designed as a simplifying box from Day 0 with top developpers from the big five inside. NDFS is really a great feature IMO and with Acropolis & the invisible infrastructure, it's not stopping improving.
I never go through benchmarking Atlantis, and digging on its features / code / design, so I can't really speak about and especially against Nutanix with Atlantis in mind.
It is well working on acceleration, is it webscale ? is it easy to use / migrate / administrate / like nutanix ? not so sure.
I would also encourage you to check out the USX pre-deployment checklist on the Atlantis website:
There's some interesting statements in here around data loss that would make me think twice about deploying Atlantis...
"Warning: No data is lost if you perform a graceful shutdown, but data will be lost if you power off or power reset the volume. "
"Warning: If you have multiple USX Managers and they are all down you must restart the "master" USX Manager first. Otherwise, all data that was in the grid is lost."
"Do not take snapshots of USX volume VMs. Doing so can cause its datastore to become inactive and its volume storage status to be fatal."
There's some other good things to know like...
"Do not change the IP addresses of any USX component after it is deployed."
Not very flexible
"The total capacity used by volumes cannot exceed the licensed capacity of the USX instance. If licensed capacity is exceeded, the volume will not be deployed or will not boot, and a notification is displayed."
Nice feature, especially if you aren't always looking at your available capacity.
"Do not deploy a USX volume on the same physical host where a USX Manager (console) is located. One USX Manager console is required, but you can deploy up to three consoles. Best practice is to deploy multiple consoles for redundancy."
Nutanix doesn't require seperate dedicated infrastructure for management. Prism is fully integrated into our distributed architecture, has an instance on every node for HA and therefore management scales automatically as the cluster grows.