New additions to Intel's memory and storage product roadmap include Optane's fastest solid-state drive yet and dense 144-layer QLC NAND SSDs that top out at 30.72 TB.
Intel claimed its new P5800X would be the "world's fastest data center SSD" and launched dense 144-layer QLC NAND SSDs offering up to 30.72 TB of capacity as part of the latest update to its memory and storage roadmap.
Large customers are now qualifying the Optane P5800X that is the first Intel SSD to use second-generation 3D XPoint memory technology. Intel said the new PCIe 4.0-based Optane SSD would boost random-read IOPS and sequential read-write bandwidth by 3x, improve quality of service (QoS) by 40%, and offer 67% higher endurance than its PCIe 3.0-supporting P4800X predecessor built on first-generation 3D XPoint.
Optane is the brand name that Intel uses for products built on 3D XPoint technology, a type of non-volatile phase change memory that it developed with Micron Technology to fill the performance gap between more expensive DRAM and less costly NAND flash. Intel and Micron ended their development partnership with the completion of the second-generation 3D XPoint, and Micron bought out Intel's interest in their IM Flash Technologies manufacturing facility in Utah. Intel subsequently struck a deal to buy 3D XPoint wafers from Micron through a contractual agreement.
A year ago, Micron claimed to have the world's fastest SSD with the unveiling of its X100 full-height, half-length SSD. Micron said at the time that the X100 could deliver 2.5 million IOPS for data reads, more than 9 GBps of read-write throughput, and latency of 8 microseconds (μs) or less in a test scenario.
Intel upped the ante today with its claims that the Optane P5800X in a U.2 form factor could hit 4.6 million random-read IOPS, at a block size of 512 bytes, and latency of less than 6 μs based on a random 4 kilobyte transfer size in a test environment. Sequential read and write throughput was 7.2 GBps and 6.2 GBps, respectively, in Intel's test case. P5800X capacity options range from 400 GB to 3.2 TB, and the Optane SSDs support 100 drive writes per day.
Optane latency advantage
"The big advantage to Optane SSDs is their low latency," said Jim Handy, general director and semiconductor analyst at Objective Analysis. "Some applications really need this, and they will benefit greatly. Other applications would experience a marginal advantage that might not be worth the added cost."
Target workloads for Optane SSDs include financial services applications, such as fraud detection and analytics, online transaction processing, high-performance computing, write caching, booting systems and logging. David Tuhy, vice president and general manager of Intel's data center Optane storage division, claimed at a press briefing yesterday that about 40% of the Fortune 500 use Optane SSDs, typically as a complement to NAND flash SSDs, not as a replacement. He cited examples of Alibaba Cloud accelerating databases, Telefonica using Optane SSDs with flash drives to meet service-level agreements, and Baidu fronting tape with Optane SSDs for endurance purposes.
As a single-port SSD, the Optane P5800X will likely find its main use cases in servers. Storage OEMs generally prefer dual-port drives for redundancy and failover in enterprise arrays. A first-generation dual-port Optane SSD is currently available, but Intel has said in the past that second-generation dual-port Optane SSDs would not ship until the second half of 2021.
Next-gen Optane PMem
Also on the Intel roadmap are next-generation Intel Optane Persistent Memory (PMem) modules, also known as dual in-line memory modules (DIMMs). Intel was sparse on details about the upcoming Optane PMem, saying only that it would be code-named Crow Pass and run on future Xeon Scalable processors code-named Sapphire Rapids.
Workloads that can benefit from Optane PMem, situated in closer proximity to server processors than SSDs, include SAP HANA and other in-memory databases, virtual infrastructure, high performance computing, AI and analytics. Customers can use Optane PMem in Memory Mode, where Optane expands the available pool of volatile memory, with DRAM, or App Direct Mode, where applications can access Optane as a tier of non-volatile memory to store data permanently.
Alper Ilkbahar, vice president of Intel's data platforms group and general manager of the Optane group, said about 60% of customers currently use Optane PMem in Memory Mode, but he expects the mix to shift more heavily to App Direct mode over time.
One bit of information that surfaced on existing Optane PMem 200 series modules built for third-generation Xeon Scalable processors code-named Barlow Pass is their use of enhanced Asynchronous DRAM Refresh (eADR) technology. The eADR feature can eliminate application wait time for cache flushes to speed performance, Intel said.
On the client front, Intel announced Optane Memory H20 with solid-state storage for thin notebooks and other space-constrained devices. H20 combines Optane memory and quad-level cell (QLC) 3D NAND flash to speed PC performance, application launches and data transfer. The Optane Memory H20 with SSD product, due in the second quarter of 2021, upgrades the 3D XPoint media and uses new NAND and Optane controllers over its H10 predecessor. The PCIe 3.0-based M.2 SSD combines 32 GB of Optane memory and either 512 GB or 1 TB of flash storage. Target use cases include gaming and content creation.
144-layer NAND flash
Intel's shift to 144-layer QLC 3D NAND flash will pave the way for denser SSDs such as the new D5-P5316, at capacity points of 16.36 TB and 30.72 TB in U.2 and E1.L form factors. Intel claimed the new SSDs also improve random read performance by more than 30% and improve endurance by 4x over its 96-layer QLC D5-P4326 predecessor. The D5-P5316 is due for general release in the first half of 2021.
"The 5316 is aimed at users that need to squeeze as much flash as possible into a small amount of rack space," Handy said. "Naturally, this application will use whichever chip provides the absolute highest available density, although the bandwidth and endurance improvements will help guide it into other applications."
There's this perception in the industry that basically QLC is not as reliable as TLC, and that's just false.
Jonmichael HandsSenior strategic planner and product manager, Intel
Jonmichael Hands, a senior strategic planner and product manager at Intel, said many customers and even some analysts are confused about QLC in comparison to other types of NAND flash such as triple-level cell (TLC).
"There's this perception in the industry that basically QLC is not as reliable as TLC, and that's just false," Hands asserted. "The cell takes more time to program, and there's a tradeoff. You get better cost at a lower endurance and higher program time. But the reliability is measured in mean time to failure, annual failure rate, temperature range, uncritical bit error, retention and shelf life. All that is equivalent to TLC.
"We use the same SSD controller, and it has good enough [error correction code] ECC to cover both TLC and QLC," Hands added. "So, we actually have the same quality metrics as TLC on the QLC."
In addition to the new QLC data center SSD, Intel announced a new 670p PCIe 3.0 client SSD that uses 144-layer QLC 3D NAND SSD and targets mainstream gaming and content creation. The 670p is due to ship in the first quarter of 2021.
By year's end, Intel plans to ship a TLC-based D7-5510 SSD tuned for cloud data center workloads. The D7-5510 SSD improves performance and lowers latency over the prior model and supports a dynamic multiple namespace to offer customers greater flexibility in multi-tenant and virtualized environments. Capacity options for the U.2 SSD are 3.84 TB and 7.68 TB.
Intel recently sold its NAND flash business to SK Hynix for $9 billion to focus on Optane and long-term priorities in AI, 5G networking and edge computing. The transaction is not expected to close until late 2021 after government approvals.
