Intel conducted yet another demonstration of its jointly developed Intel/Micron 3D XPoint technology during its keynote at Computex 2016. A few more details also emerged about how Intel plans to implement 3D XPoint on the Kaby Lake platform.
Intel has long touted the new product, which is the first new memory to be productized in more than 25 years, as a game changer. Unfortunately, both Intel and Micron have been vague about the technology behind the new storage medium, and both companies simply refer back to the now-tired mantra that it is "1000x the performance and endurance of NAND with 10x the density of DRAM."
We know that 3D XPoint is fast, but exactly how fast is up for debate. Intel has demonstrated working 3D XPoint on several occasions, but the demonstrations do not reveal the actual hard performance data, endurance metrics or pricing information (which are the key considerations that will dictate its success). Intel also clearly designs the presentations to obfuscate the underlying performance characteristics, such as using it as a backup device over a Thunderbolt 3 connection (a performance bottleneck).
The Computex demo was more of the same, as the company measured the render time of a fluid dynamics video. The scale of the rendering was impressive. It consisted of 1.1 billion particles and presented itself as a swirl of water that lasted only seven seconds. Surprisingly, Intel indicated that it required 35 hours to render the short clip with a computer sporting "the world's fastest PCIe SSD," which one would assume is an Intel DC P3608.
Intel presented a time-lapse of the rendering process, and the competing 3D XPoint-infused computer was able to render the same video in a mere 9 hours, which is an impressive achievement. However, once again, the presentation did not reveal any concrete performance specifications.
Intel has publicly stated that its upcoming chipsets will support 3D XPoint, and according to motherboard industry sources at Computex, Intel is requiring manufacturers to add 3D XPoint-supporting RAM slots to some Kaby Lake motherboards.
3D XPoint can be used as a normal block-addressable storage device with NVDIMM, PCIe, SAS or SATA devices, but users can only address it as memory (bit-addressable) if it rides the memory bus. 3D XPoint can be used as a block storage device when it is on the memory slot as well, which provides it the advantages of a lower latency connection, thus possibly offering unparalleled storage performance.
Intel will utilize JEDEC-compliant DIMMs, but motherboard vendors have to add BIOS support for NVDIMM form factor devices. The industry has already adopted a standardized NVDIMM specification, and Micron indicated during its NVDIMM launch that it would use the same architecture for future 3D XPoint-based devices. We have also spotted Intel NVDIMMs in the field.
Using 3D XPoint as memory will also require special software architectures and system optimizations for the applications to utilize the somewhat slower 3D XPoint tier, so it is far from plug-and-play. There are system level optimizations that can abstract the complexity of the dual-tiered memory configuration away from the application, but the emerging software solutions are immature and difficult to implement. However, using 3D XPoint as memory will offer incredible density, up to 10x more capacity per RAM slot, and its lower cost will actually allow us to pursue such lavish accommodations.
Intel's admission that Kaby Lake is 3D XPoint compatible is telling; technically 3D XPoint is compatible with any interface, such as PCIe, SAS/SATA (or even USB 1.0), because the device merely presents itself to the host using standardized interfaces (when used as a block device). 3D Xpoint will require the special considerations that merit the "3D XPoint" compatibility tag only when it is used in a RAM slot.
Kaby Lake is coming later this year, and there is no doubt that 3D XPoint is going to play a big part in the future platform. Using 3D Xpoint-powered Optane SSDs may change our perception of what fast storage means, but utilizing it as a persistent memory tier has the power to change the entire computing paradigm.