Hewlett Packard Enterprise (HPE) has big ambitions for transforming the data center  — from composable infrastructure, which is almost ready to ship, to persistent memory, a flash/DRAM combination that the company plans to announce Thursday.

At a session in Palo Alto, Calif., this week, HPE executives stepped the media through an overview of the company's data center plans. Here's some of what we learned about composable infrastructure and persistent memory.

1. HPE Believes It's Got the Only Composable Infrastructure

Cisco has used the "composable" phrase as well, to introduce its UCS M-series servers in 2014. But HPE's Synergy is developed so that one API handles everything related to the infrastructure — not just the amount of compute and storage offered to an application, but also nitty-gritty details such as BIOS settings.

HPE doesn't believe competitors' composable offerings are tied together so smoothly. "A lot of them are cobbling bits and pieces and saying it's simple. Think of it as: They're testing it to be simple, not building it that way from the ground up," said Ric Lewis, HPE's senior vice president of converged data center infrastructure.

The single API is interesting because with composable infrastructure, the idea is that an application asks for the amount of computing and storage it needs. Synergy's API appears to take care of behind-the-scenes setup as well.

Composable infrastructure differs from hyperconverged infrastructure. Both put compute and storage under the control of one piece of software, but hyperconverged infrastructure is built to serve objects that don't change after being created. Virtual machines are a prime example, Lewis said.

Of course, HPE offers hyperconverged infrastructure as well, the latest example being the Hyper Converged 380, which is due to ship in April.

Another contrast between composable and hyperconverged is that the former operates at the scale of an entire data center, whereas the latter is built to handle relatively small groups of applications, Lewis said. HPE's hyperconverged offerings are "mostly being used at the small to midsized enterprises and the remote office or back office."

2. Composable Infrastructure Is Ready to Ship

HPE's composable infrastructure platform is available to order today, with beta units likely to ship in May, according to Lewis.

3. Synergy Will Vertically Integrate

HPE has been amassing software partners whose products could be offered as services on composable infrastructure. Orchestration could be one such service.

It's not a stretch to imagine those services being sold as part of the composable infrastructure package. "We're headed toward more and more vertical integration of that stack," said Lewis.

4. Synergy Will Use Persistent Memory

The platform will be outfitted to support persistent memory from Day One, Lewis said. That matters, because...

5. Persistent Memory Is a Software Story, Too

It's easy to cynically describe persistent memory as "DRAMs and flash sitting next to each other," but really, persistent memory has implications all the way up the software stack, said Tim Peters, vice president of server software and core enterprise technologies.

Persistent memory really is DRAM backed up by flash. The idea is to give applications the speed of DRAM — which uses a bus faster than that of long-term storage, such as disk drives — without risking the loss of all that data should the server suddenly reboot or crash. The concept goes hand-in-hand with the idea that data center servers are a commodity and don't need to be ultra-reliable.

Getting this to work requires some software help, though. The server has to be prepped to work with persistent memory. That's a key part of HPE's Thursday announcement: the launch of its ProLiant DL 360 and 380 servers, which feature a BIOS enabled for persistent memory. At Microsoft Build this week, HPE is also showing persistent memory running in a Windows Server 2012 environment.

HPE's first persistent memory cards will include 8 GB each of DRAM and flash, plus a microcontroller that handles the backing-up of data.

A more sophisticated type of persistent memory is 3D XPoint, being developed by Intel and Micron. HPE plans to use that memory as well, when it's available, Peters said.

6. Memory Will Be Composable Someday

Persistent memory doesn't imply composable memory, the pooled resource of memory that would be decoupled from the CPU. But HPE considers persistent memory to be a small step in the direction of composable memory.

HPE's first persistent memory products will keep the memory locked to a particular CPU. But the company does expect memory to evolve to the point where it's just "data with stuff around it," Lewis said — meaning, a clump of data storage that applications use at will.