Since Intel announced its 3D Xpoint storage (Optane), the company claims that the new storage class would mean a major step forward for the industry as a whole. Proof of these claims has been relatively slow – it is difficult to replace existing storage technologies, and the existing storage stack (spinning disks, NAND and DRAM) covers a wide range of price points, power consumption, reliability and capacity.
Intel launches a new class of Optane called Optane DC Persistent Memory. Its capabilities are designed to bridge the gap between DRAM and nonvolatile memory while expanding the available memory per CPU socket to 3TB.
The Optane DC Persistent Memory is Anandtech pin-compatible with DDR4 and comes in packages of up to 512GB per Stick offered (six RAM slots = 3 TB of addressable RAM per socket). Systems will be able to deploy large Optane caches alongside smaller DRAM pools; One of Intel's demos showed a Cassandra database running on 256 GB DDR4 RAM and 1 TB Optane DC PM, as opposed to 1 TB DRAM. Intel's main focus on Optane DC PM is on the memory performance consistency . The use of DRAM + NVMe-attached memory may be limiting in certain scenarios, with performance degraded by memory write-backs. An optane cache avoids this problem.
Intel is also working on a capability it calls Persistent Memory Over Fabric (PMoF), a low-data-replication method with direct load / memory access that has a performance of 287,000 Ops / second compared to an initial performance of 3,164 Ops / second for a conventional DRAM + memory system (possible consistency for the DRAM + memory system, Intel notes, is similar to the Optane-Rig.) But Optane's low latencies are also in others Contexts useful, such as restarting a database: Intel reports a restart time of 2,100 seconds ds for a conventional configuration compared to 17 seconds for Optane.
Much of the work to achieve this was done on the software side. Optimizations and file system abstractions are needed to put the "Persistent Memory" into the Intel started Optane DC drives. To help companies optimize their databases and software for Optane DC PM, Intel has developed a new Persistent Memory Development Kit (PMDK) with a collection of libraries, APIs and other software tools. PM is supported on both Windows and Linux, and Intel has added support for the performance analysis software Vtune.
Details about the SDK and how it interacts with software are currently in short supply. But the implication here is that the heavy lifting to work Persistent Memory is handled outside of the applications themselves. In other words, if you run Application X, X does not necessarily have to be optimized by the original developer to take advantage of persistent storage. That is the task of the PMDK. The implication here is that app developer optimizations are useful and productive, but are not a literal requirement – and that's an important feature in talking about a capability with such a potential for reshaping storage hierarchies.
Other tidbits from the Q & A: These new capabilities will be linked to new, upcoming CPUs from Intel, with sales to selected customers in 2018 and broad availability in 2019. Intel also expects QLC NAND Drives such as Micron will ship in the second half of the year and by 2019. DIMM opt – in performance, power consumption, or pricing data are not currently released, but Optane DIMMs are expected to have standard DDR4 clock rates. V3.espacenet.com/textdoc?
Overall, this shift towards optane and a focus on fabric performance is a critical component of Intel's own transformation from a personal CPU-focused business to a more connected to data center and cloud computing. Intel sees new storage technologies as the key to its long-term addressable marketplaces and aligns its initiatives in these areas, possibly to tie companies closer to their CPU products. Because if Optane offers significant performance benefits and works with Intel (or only with Intel) Best With, then Intel has a clean method to make itself even more relevant in the data center market.
This, in turn, may also reflect the need to optimize other facets of computing beyond pure CPU performance, possibly as part of the pursuit of exascale computing, where power consumption of DRAMs is an important limiting factor. With Qualcomm leaving the server business, Intel is not facing the heat of ARM in its data centers, and AMD's Epyc ramp, while posing a competitive threat over the long term, should not shake Intel's server dominance (AMD hopes to reach 4). -6 percent of the server market this year.