Intel launches its first Optane SSD, the DC P4800X

Computing cj Times

Roughly 18 months ago, Intel and Micron announced they had developed a new type of memory they dubbed 3DXPoint. While details have always been scarce on how the NAND flash competitor would perform in the real world, Intel made impressive claims about its reliability, power consumption, and performance.

Today, Intel is launching its first Optane SSD, the DC P4800X. This new card has a 375GB capacity, with much lower latencies than NAND.

OptaneSpecs

Intel is still playing its cards close to its chest. It doesn’t want to talk about how, precisely, Optane works or what it’s made of. Still, we now know a great deal more about where Intel sees this technology fitting into the data center, as well as how much it will cost.

In the old days, data was kept in-cache as much as possible, in main memory if not, and on a hard drive if there was absolutely nowhere else to put it. Hard drives were always the biggest problem with this approach; their latencies are measured in milliseconds and are orders of magnitude higher than DRAM, which is significantly worse than CPU cache. NAND flash latencies are vastly lower than hard drives, but there’s still a significant gap between NAND and DRAM. Intel’s Optane memory is meant to sit between DRAM and NAND in much the same fashion as NAND might sit between DRAM and a conventional hard drive in a modern system.

BareOptane

Intel’s DC P4800X without its cover.

Another advantage of Optane is that it offers the same performance at multiple queue depths. SSDs typically run faster the higher the queue depth, and so does Optane, but the gap isn’t nearly so large.

IntelPerfGraph

Intel sees two main use-cases for the Optane DC P4800X, both shown below:

OptaneSSDUse

It can be used as a fast storage and cache drive, where its primary purpose is to buffer NAND and offer higher performance, or it can be deployed as a memory replacement / extension. We’ve seen some of these approaches before that actually utilized NAND flash — and while it may seem counterintuitive to deploy NAND (which is still vastly slower than DRAM), there are workloads that simply don’t care much about access speeds. Being able to hold colossal amounts of information in memory for significantly less power can be more important than raw bandwidth.

lIKE ()orShare