Intel just put out two new lines of data center oriented drives in the S3500 family, one M.2 and one 2.5 inch. While both are similar in function, the form factor differences point to very different use cases.
First off is the vanilla one, a 2.5″ form factor SATA device aimed at data center workloads. If you have been paying attention, Intel has drives in this line and form factor already, the new bit is higher capacities of 1.2 and 1.6TB. The flash they use is 20nm Intel and the controller is also a new Intel part but as of now we have not been told its name. It replaces a similarly unnamed Intel controller found in the lower capacity S3500 drives.
Update November 13, 2014 @ 1:00PM: The new controller is the Intel PC29AS21CB0 and the old one was the PC29AS21CA0. The differences in functionality should be clear from the names. :P
S3500 line is the lowest end that Intel makes and is specced for only .35 writes a day vs 3x for the S3600 and 10x for the S3700 parts. Officially the 1.2TB model has a write life of 660TB and doing the math puts that at a bit over 4.3 years of steady use or longer than most servers live for. It should be good enough if you run the correct workloads because the warranty is set at 5 years, the spec sheet is probably quite conservative. More important are the inclusion of ultracaps to make sure writes complete in a total power loss.
How well does it perform? 500MBps read and 450MBps write is claimed, something that any decent modern drive should be capable of, nothing special there. For IOPS it will do 65K 4K reads and 15.5K 4K writes, decent enough for a low-end device aimed at read intensive workloads. Power is a claimed 5W for typical active workloads, .7W at idle. For the technically pedantic, the Identify Device Command Data returned word 95 is a fixed value of 0h for the Stream Minimum Request Size. Not sure why it matters but we looked it up and felt compelled to tell you.
Yes it looks like an M.2 drive
More interesting is the M.2 form factor S3500 which shares much of the same component base as the 2.5″ form factor version. Since it is an 80mm version the physical space limits capacities to 80, 120, and 340GB but there is still room for backup ultracaps. The write endurance is the same as the 2.5″ version, .35 a day for ~5 years or 45, 70, and 180TB respectively.
The form factor and thermal constraints mean lower power use at ~4W typical, .5W idle. Lower draw from the same components typically translates to lower performance and that mostly holds true here. Sequential bandwidth is listed at 420/370MBps R/W and 67K/14.5K 4K IOPS R/W, just a bit of a hit and not bad for the size difference. The M.2 S3500 should be more than enough for it’s intended purpose.
That brings up the most important question, why would anyone want an M.2 device for a server? The answer is as a boot device for VMMs and other sub-OS level management code that used to be pulled from an SD card, USB stick, or across the net. Most of this code is small, never changes, and stays memory resident after a single load. The M.2 SSD not only replaces these form factors but they offer massive upgrades to capacity, reliability, and performance. Basically they do what is needed to bootstrap and manage a server without taking a precious drive bay from the main storage arrays. Average users won’t care but this capability is golden for the datacenter set.
In the end neither drive brings anything massively new to the table but both are solid advances from their predecessors. An M.2 form factor device with the full enterprise feature set will be very welcome in some markets, completely ignored in most others, and is very welcome to have as an option. And if you are still paying attention, the M.2 form factor also returns 0h as word 95 on an Identify Device Command Data request, and the world is still roundish.S|A
Latest posts by Charlie Demerjian (see all)
- Intel shows off 10nm 112Gbps SerDes - Mar 12, 2019
- Intel releases Compute Express Link spec - Mar 11, 2019
- Qualcomm rolls out a second gen 5G modem called X55 - Feb 19, 2019
- What is Intel’s Foveros tech and what isn’t it? - Feb 11, 2019
- Why SemiAccurate called 10nm wrong - Jan 25, 2019