Intel just unleashed a new kind of computer memory it believes will fundamentally change the way the world builds computers. But it won't tell the world what's inside.
The company calls this new creation 3D XPoint---pronounced "three-dee cross-point"---and this week, after touting the stuff for a year-and-a-half, Intel finally pushed it into the market. You can think of the new technology as a computer building block that can serve more than one purpose---a single thing that can replace several others.
But the new tech also serves as a way for Intel to shift multiple markets in its favor. That's probably why it won't say what the memory is actually made of. Certainly, Intel doesn't want others duplicating the technology, which it developed alongside hardware maker Micron.
This chess game is worth watching because it could indeed change the way machines are built, most notably inside big internet companies like Google, Facebook, and Amazon. These companies now build machines on a massive scale. They're hardware markets unto themselves. And more than anyone else, they stand to benefit from Intel's new technology. But a built-in tension exists here as well. Because they operate on such a scale, Google and Facebook don't want their fate controlled by just one hardware supplier. They want choice, which brings efficiency by driving down prices. This dynamic is shaping the worldwide market for data center hardware. And Intel is pushing in the opposite direction.
Traditionally, computers stored data in two ways. They stored most of it on hard drives, which could hold large amounts of information for long periods of time, even as machines were powered on and off---and do it pretty cheaply. But computers also used separate memory systems called DRAM to store the data they needed right now. This memory was much faster, but it was also more expensive and held less data. 3D XPoint can replace all those pieces---or so Intel says. "This is truly transformational," Intel CEO Brian Krzanich tells WIRED. "It allows architects---both at the PC level and the data center level---to rethink how they build the system."
Krzanich sees this technology redefining gaming machines and other personal computers. But that's almost beside the point. A PC is a PC. Where this could really change things is inside the massive data centers operated by the internet's biggest companies. As their online empires continue to grow, these companies always need faster and cheaper ways of storing ever-larger amounts of data. Over the past several years, they've already moved beyond hard drives to significantly faster flash solid state drives, or SSDs. Since even these drives aren't fast enough in some cases, they've created a new breed of database that pushes massive amounts of data into memory. 3D XPoint could help in both those areas. According to Intel, it's about 1,000 times faster than flash and can store about 10 times more data than DRAM.
But right now, neither of those numbers really apply. On Sunday, Intel released the first 3D XPoint product, a storage device marketed under the name Optane that can replace hard drives or flash SSDs. But this device is only about five times faster than flash SSDs---not 1,000---because of the standard interface that connects the drive to the rest of the machine. Eventually, Intel will offer products that plug straight into the heart of the machine, removing this bottleneck, but that won't happen until at least the second half of the year.
When that happens, the likes of Google and Facebook really could build new kinds of systems---systems that can hold data in memory even when powered down, that merge memory and storage or mix and match them in new ways---all in an effort to create vast networks of machines that can more efficiently juggle information. This is no small undertaking. If they significantly reorganize their hardware, they must also rewrite their operating systems and other software to accommodate the changes. According Krzanich, Intel is already working with "almost all of the big cloud service providers" to explore these possibilities. These companies, he says, have been testing sample hardware since early last year.
Still, so many questions remain. It is unclear, for instance, how fast 3D XPoint will be relative to DRAM. For Jim Handy, an analyst with Objective Analysis who follows the memory and storage markets, the technology only makes sense as memory if it's significantly cheaper than DRAM. It's also unclear when Intel will allow this technology to plug straight into processors offered by competitors—though Krzanich indicates it will. "We will always have options that can fix with anybody's architecture," he says. "With something as ubiquitous as memory, you've got to have it work with everybody."
What Krzanich does make clear is that Intel will not sell raw 3D XPoint memory to anyone else. "We won't sell this in any other form but a finished product," he says. Today, Intel is one of so many companies that sell flash, and that means its opportunities are slim. It wants to build a market for 3D XPoint where other options are scarce.
Will the internet giants embrace that kind of arrangement? It's an important question. As Google, Amazon, and Microsoft expand their cloud computing businesses, offering services where others can set up their own online empires without buying their own hardware, these internet giants will account for more and more of the worldwide hardware market. Increasingly, these are the main customers for parts suppliers such as Intel. In 2012, Intel vice president Diane Bryant told WIRED that Google bought more server chips from Intel than all but about four other companies.
As Krzanich points out, Micron will also sell 3D XPoint hardware. So some competition will exist, eventually. And others are developing alternatives. It's possible competition will push down prices, and the internet's biggest companies will bite. But for now, the dance goes on to decide what really winds up inside.
Update: This story has been updated to clarify that it is still unclear how fast 3D XPoint relative to DRAM.
