Why Intel eradicated its Optane reminiscence enterprise • The Register

Evaluation Intel CEO Pat Gelsinger has confirmed that Intel will exit its Optane enterprise, ending its try to create and promote a stage of reminiscence that’s barely slower than RAM however has the virtues of persistence and a excessive IOPS.

Nonetheless, the information mustn’t come as a shock. The division has been on life help for a while following Micron’s choice in 2018 to finish its three way partnership with Intel, promoting the manufacturing unit through which the 3D XPoint chips present in Optane drives and modules had been made. Whereas Intel has signaled that it’s open to utilizing third-party foundries, with out the means to make its personal Optane silicon, the writing was on the wall.

As our sister website, Blocks and Information, reported in Could, the sale solely got here after Micron charged Intel with extra 3D XPoint reminiscence modules, greater than the chipmaker may promote. Estimates put Intel inventories at about two years of provide.

In its poor second-quarter earnings report, Intel mentioned ditching Optane will end in a $559 million stock write-down. In different phrases, the corporate provides up on the challenge and writes off the stock as a loss.

The deal additionally marks the tip of Intel’s SSD enterprise. Two years in the past, Intel offered its NAND flash enterprise and manufacturing plans to SK hynix to focus its efforts on the Optane enterprise.

Introduced in 2015, 3D XPoint reminiscence arrived within the type of Intel’s Optane SSD two years later. Nonetheless, not like rival SSDs, Optane SSDs could not compete on capability or pace. As a substitute, the gadgets supplied a few of the strongest I/O efficiency in the marketplace, a top quality that made them notably engaging in latency-sensitive purposes the place uncooked IOPS had been extra necessary than throughput. Intel claimed that its PCIe 4.0-based P5800X SSDs may obtain as much as 1.6 million IOPS

Intel additionally used 3D XPoint in its Optane persistent reminiscence DIMMs, notably across the launch of its second and third era Xeon Scalable processors.

From a distance, Intel’s Optane DIMMs did not look any completely different than your run-of-the-mill DDR4, aside from maybe as a warmth sink. Nonetheless, on nearer inspection, DIMMs may have a lot larger capacities than is feasible with DDR4 reminiscence immediately. Capacities of 512 GB per DIMM weren’t unusual.

DIMMs had been positioned alongside customary DDR4 and enabled quite a few novel use circumstances, together with a tiered reminiscence structure that was basically clear to working system software program. When carried out this fashion, DDR reminiscence was handled like a big stage 4 cache, and Optane reminiscence behaved like system reminiscence.

Whereas not providing the efficiency of DRAM, the method enabled the implementation of very massive, memory-intensive workloads, comparable to databases, at a fraction of the price of an equal quantity of DDR4, with out the necessity for software program customization. That was the thought, anyway.

Optane DIMMS is also configured to behave as a excessive efficiency storage system or a mix of storage and reminiscence.

Now what?

Whereas DDR5 guarantees to deal with a few of the capability challenges that Optane persistent reminiscence solved, with deliberate DIMM capacities of 512GB, it is unlikely to be competitively priced.

DDR is not getting cheaper, no less than not rapidly, however NAND flash reminiscence costs are plummeting as provide outstrips demand. In the meantime, SSDs are getting sooner in a rush.

This week, Micron started quantity manufacturing of a 232-layer module that may push client SSDs into 10+ GB/sec territory. That is nonetheless not quick sufficient or low latency sufficient to exchange Optane for big in-memory workloads, analysts say. Registernevertheless it’s getting very near the 17 GB/sec supplied by a single channel of low-end DDR4.

So if NAND is not the reply, then what? Nicely, there’s really a substitute for Optane reminiscence on the horizon. It is known as Compute Specific Hyperlink (CXL), and Intel has already invested closely within the expertise. Launched in 2019, CXL defines a cache-consistent interface for connecting CPUs, reminiscence, accelerators, and different peripherals.

Attributable to ship alongside Intel’s long-delayed Sapphire Rapids Xeon Scalable processors and AMD’s fourth-generation Eypc Genoa and Bergamo processors later this yr, CXL 1.1 permits reminiscence to be immediately connected to the CPU via the PCIe 5.0 hyperlink.

Distributors together with Samsung and Marvell are already planning reminiscence growth modules that slot into PCIe like GPUs and supply a big pool of further capability for memory-intensive workloads.

Marvell’s acquisition of Tanzanite this spring will permit the seller to additionally supply Optane-like tiered reminiscence performance.

Additionally, as a result of the reminiscence is managed by a CXL controller on the growth card, older, cheaper DDR4 and even DDR3 modules can be utilized alongside trendy DDR5 DIMMs. On this regard, CXL-based reminiscence tiering may be superior, as it isn’t primarily based on a specialised reminiscence structure like 3D XPoint.

VMware is contemplating software-defined reminiscence that shares a server’s reminiscence with different packing containers, an effort that might be far more highly effective if it makes use of a regular like CXL.

Nonetheless, emulation of some points of Intel’s Optane persistent reminiscence might have to attend till the primary CXL 2.0-compliant CPUs hit the market, which can add help for pooling and reminiscence switching. It additionally stays to be seen how the software program interacts with CXL reminiscence modules in tiered reminiscence purposes. ®

Leave a Comment