Intel’s 3D XPoint memory has attracted a lot of interest after the recent Intel press event, so today we’ll be going over what it is, why it’s important, and what to expect in the future from Intel’s new DIMMs.
While there are still several hurdles for Intel to overcome, the newest in storage class memory promises some powerful server hardware upgrades across several use cases. But before we get into that, let’s go over what it actually is:
What is Intel 3D XPoint/ Optane Persistent Memory/ NVDIMM / Cross Point Memory?
Originally codenamed Apache Pass, Intel’s 3D Xpoint Optane DIMMs developed with Micron (who brands it as QuantX), are the newest development in the field of non-volatile DIMM (NVDIMM). Like traditional storage, NVDIMM memory is “persistent,” and doesn’t lose its data when powered off, which is why it’s commonly referred to as storage class memory.
This is not the first iteration of non-volatile memory, with Viking Technology, AgigA, SMART Modular, and others providing a more contrived method. Essentially they make the DIMM’s persistent by keeping it in DRAM and then copying it over via a microcontroller to a NAND chip once the system powers down. It relies on a backup power supply of supercapacitors and buffers between the NVDIMM and the server’s memory bus channels.
3D XPoint Optane on the other hand utilizes a wholly new architecture with persistent data as a built in feature, non-dependant on supporting hardware.
Why is Cross Point Memory Important?
3D Xpoint is quite possibly the biggest advancement in storage mediums since NAND over two decades ago.
Initially, Intel and Micron made some overzealous 2015 marketing claims of:
- 10x the storage memory of DRAM
- 1000 times better endurance and speed than NAND
These stats had many excited to sell all of their memory and upgrade as soon as the modules came on market. However, Intel later backpedaled in a following presentation to significantly lower numbers:
- 3x the endurance & 10x the speed of NAND
- 4x the capacity of DRAM
While the regression from promised performance is disappointing, the technology still does present enticing possiblities for I/O improvements by removing the bottlenecks present in most architectures.
How does 3D Xpoint Work?
Cross Point Memory Structure
3D XPoint DIMMs, also known as cross point memory, owe their properties to a unique architecture. XPoint does not rely on phase change like other architectures according to Intel (though some have disputed this). Instead, each cell’s binary state is determined by the electrical resistance of the bulk material according to Intel’s Krzanich.
Cells are housed on a stacked array of perpendicular wires -hence the name crosspoint memory – and varying voltages can be sent through the array to selectors to write – and change resistance levels through bulk property changes – or read the status of the cell.
The density of XPoint is about 4x higher because transistors are not necessary for individual data cells, freeing up quite a bit of space in the grid.
The Optane NVDIMMs will also require specialized cooling if preliminary findings of heat production are any indication.
Additionally, According to a Lenovo Thermal Architect, Vinod Kamath, “the Xpoint memory can consume about 3x the power of a standard 16GB DDR4 DIMM…. could range from 15-18W.”
While intel did promise that Optane DIMMs would have comparable lifetimes to other DIMMs, we still don’t know what the official metrics are for endurance with varying workloads.
Another consideration is that, like SSDs, XPoint isn’t infinitely persistent. Higher endurance data center SSDs only retain data for about 3 months, while some client SSDs can hold data as long as one year. The data on this has not been made available yet.
The 3D XPoint modules are not compatible with AMD CPU based systems, or any current gen Xeons. They essentially only work with the as of yet not available Cascade Lake chips.
Persistent memory also can’t be mixed and matched with DRAM. Intel’s responses, and much of the talk within the industry, imply that you’ll need to use dedicated channels for XPoint to avoid write failures. While it’s possible that Intel and Micron could solve this issue by launch, word on the street is that the issue is built into the hardware, and insiders don’t predict a solution in the near future.
What to Expect from 3D Xpoint Memory
When Can I Buy Intel 3D Xpoint Memory?
The short answer? 2019 at the earliest. They claim they’ll be shipping in 2018, but that will be more for more qualifying purposes, considering that Cascade Lake chips aren’t even available till 2019. With the XPoint release already having been delayed several times since 2015, and most recently due to (speculated) endurance issues, it’s very possible that widespread adoption could come as late as 2021, so don’t worry about selling off your servers for NVDIMM compatible systems any time soon.
That being said, Intel is offering test access to a cloud environment utilizing this memory, allowing developers to test out their software on the new tech. The DIMMs can utilize the same programming/software framework of the NVDIMMs, which should help adoption once the modules do come to market.
You can apply for test access here. Applications are only being accepted until June 13, when the testing begins.
What Does Storage Class Memory Mean for the Future of Memory Storage?
The first thing to note is that Intel and Micron’s NVDIMM brainchild will not have enterprises everywhere selling all of their SSDs or HDDs. With more storage dollar for dollar, they will still fill their niche for more cold storage purposes.
It doesn’t make sense to fill every DIMM slot with Optane NVDIMMs unless the speed of memory you need is conservative, given that DRAM still is significantly faster. As a result, 3D Xpoint memory will likely find its place between DRAM and the other types of longer term storage, keeping workloads out of long term storage to minimize latencies.
With the ubiquitousness of I/O bottlenecks in everything from virtualization to in-memory databases, cross point memory’s adoption wouldn’t be isolated to just one or two use-cases.
If Intel can work out all of the kinks with it, it could mean a serious bump in market share to help them recover from the recent Intel chip vulnerability fiasco. Any enterprises that want to make use of the memory will be forced into using Intel’s next-gen Cascade Lake Xeon processors, Xeon Bronze, Xeon Silver, Xeon Gold, and Xeon Platinum, at least until other competitors can adapt. If you are going to be upgrading processors and getting rid of old ones, always make sure to look into if you can sell your old processors for money so they don’t go to waste.
While there are still quite a few unknowns, the potential upside to this technology is too significant to ignore.
Have something to add? Let us know your thoughts in the comments below!