Does 3D XPoint Spell the End for Flash Storage?

Thursday Apr 21st 2016 by Paul Rubens

Intel and Micron's new storage technology is faster than flash and cheaper than DRAM. The big question is how it will be adopted.

If your storage systems are due a refresh and you are planning future purchases, then there's a huge elephant in the room that needs addressing: Intel and Micron's high performance 3D XPoint technology.

Touted by Intel and Micron as the first new memory tier to be launched since the introduction of NAND flash memory in 1989, its launch could have a significant effect on the storage market and your purchasing decisions.

So here's the problem: There's a good chance you've been looking at flash SSD-based systems to replace some of your conventional storage capacity, but not much is known about 3D XPoint yet—even though it should start appearing in significant quantities in the next 18-24 months.

Do you really want to be buying an expensive flash SSD system offering higher performance than an HDD-based array just as a new tier of even higher performance storage hits the market?

Before answering that question, let's do a quick recap about what we know about 3D XPoint. Intel and Micron have been fairly coy about the precise nature of the technology that underlies it, but we do know that it is non-volatile — so it's more like NAND flash (or even HDD storage) than volatile DRAM.

But according to Intel, 3D XPoint is up to 1,000 times faster than flash, with 1,000 times more endurance (or tolerance to write cycles.) It's also up to ten times denser than DRAM, which is important as that means there's the potential for manufacturing costs to be low. And its latency is about ten times lower than an enterprise class SSD connected over NVM Express (NVMe.)

It's still slower than DRAM, but it will also be less expensive.

This means that in the storage hierarchy, 3D XPoint slots in between DRAM and NAND: cheaper but slower than volatile system memory, but much faster though more expensive than non-volatile SSDs or system-mounted flash cache modules.

By its nature 3D XPoint is more like fast but expensive flash than slow but cheaper DRAM, because it's a non-volatile storage medium. And Intel promises that SSDs filled with 3D XPoint storage rather than conventional NAND will offer up to ten times the performance.

But due to its durability and its high density (compared to DRAM,) Intel says it can also serve as a single combined pool of high-speed system and storage memory.

So the question is really this: how will 3D XPoint be used: in system memory as a replacement for volatile DRAM, or as a high-speed replacement for NAND in conventional solid state storage systems and flash caches?

Jim Handy, solid state storage expert and semiconductor analyst at Objective Analysis, believes that it's the former use case that will be the most popular. "The point is to achieve better performance for less cost, and that is (best) done by reducing DRAM use through the addition of 3D XPoint," he says.

"Any use of 3D XPoint to reduce SSD or HDD use would be to replace a cheaper technology with a more costly one with little, if any, speed gain," he adds, "I do not expect for 3D XPoint to impact SSD sales at all."

That may sound counterintuitive, but if you have a fixed budget, then spending $100 on 3D XPoint instead of DRAM will often lead to a performance boost.

That's because even though 3D XPoint is a little slower than DRAM you can buy more 3D XPoint than DRAM for your money, thus capturing more data or instructions that would otherwise require a much slower SSD or HDD access, Handy explains. "So adding 3D XPoint instead of DRAM will make your system faster," he says.

As for the idea of 3D XPoint replacing SSDs, Handy points out that although many vendors are touting all-flash arrays, it rarely makes sense to move all your corporate data into fast-but-expensive flash storage.

"When it comes to it, hard drives still have an important role for data that may only be accessed once a week," he says. "The same argument applies to 3D XPoint. Some things it will make sense to store on it, but a large volume of stuff will stay in NAND."

In terms of where 3D XPoint will replace DRAM, Handy is convinced that in the short to medium term this will be in the server market — both in corporate data centers and in the cloud —rather than in end-user machines.

"I expect to see 3D XPoint for the first five years in servers as data centers have a concrete way of calculating return on dollars spent," he says. "So when they compare a server with XPoint or DRAM they will see that the performance is about the same, and the cost is lower. In a PC the cost benefit is not as well defined, and I don't see XPoint catching in in the mobile community."

In terms of specific workloads that will particularly benefit from 3D XPoint, Micron suggests that initially these will be:

  • Data analysis/data mining
  • Data warehousing
  • Online transaction processing
  • Virtualized infrastructure
  • Graph analytics
  • VDI

One particularly interesting use of 3D XPoint may be in the area of vast in-memory databases which benefit from fast but expensive DRAM, says Handy. 3D XPoint will make systems that can run these databases much lower cost to build because of the lower cost of 3D XPoint and its higher density.

It will also solve another problem: what happens to these applications when something goes wrong, he says. The fact that 3D-XPoint is non-volatile, unlike DRAM, is very important here. "A key to the idea of < ahref="">in-memory databasesis that if things go wrong, the data is in one place. 3D-XPoint can be that place," says Handy.

Micron is expecting significant take-up of the new technology in the server market, according figures released at a 2016 Analyst Conference in February. It expects about 5 percent of mainstream servers to be using some 3D XPoint within two years, rising to almost 30 percent by 2022. And it expects over 4 billion gigabytes of 3D XPoint memory to be in use at that point.

If it turns out that 3D XPoint is used principally as an alternative for DRAM rather than as a replacement for NAND, that is potentially good news for Intel and Micron. That's because the global DRAM market is worth around $45 billion every year, dwarfing the $31 billion market for flash, according to Gartner estimates.

But Intel believes that there will be a market — eventually — for high-performance SSDs based on 3D XPoint. It's also banking that there will be a market for high-end laptops and desktops filled with 3D XPoint for applications such as gaming that require very low latency.

In addition, the company says the technology may be adopted by mobile device manufacturers, leading to improved battery life, although the power requirements of 3D XPoint-based devices are not yet known.

The earliest that 3D XPoint will be available in significant quantities is likely to be next year, but in the meantime NAND manufacturers like Micron are concentrating on the production of cheaper but slower 3D NAND which offers far higher storage density and 25 percent lower cost per gigabyte than high-density 16nm planar NAND.

The company is also working to release its second-generation 3D NAND this year , which will offer a 30 percent cost reduction over the first-generation 3D NAND. Micron is planning to ramp up production of 3D NAND to the extent that 3D NAND will make up over half of its fabrication plant output by fall 2016.

How should all of this affect your storage purchasing plans? While the introduction of a new tier of storage is certainly significant, the likelihood is that it will have little impact on the storage array market until the next storage refresh cycle. And the good news is that in the meantime NAND-based storage costs will continue to fall rapidly.

Photo courtesy of Shutterstock.

Mobile Site | Full Site
Copyright 2017 © QuinStreet Inc. All Rights Reserved