X
Innovation

Intel claims fastest data centre storage title with Optane-charged P5800X

The Optane P5800X can do 1.5 million IOPS of random 4k reads and writes, and up to 1.8 million in mixed workloads, the chip giant claims.
Written by Chris Duckett, Contributor
intel-optane.png
Image: Intel

Armed with PCIe 4.0, Intel is laying claim to creating the fastest SSD for data centres on the planet, as it unveiled its Optane-powered P5800X storage on Wednesday.

According to the spec sheet, the P5800X is capable of hitting 7.2GB per second on sequential reads, 6.2GB per second on sequential writes, up to 1.5 million IOPS of random 4k reads and writes, which can increase to 1.8 million IOPS for mixed workloads of 70% reads and 30% writes.

Compared to its previous generation of Optane storage, the PCIe 3-powered P4800X which was unveiled in 2017, the new product delivers three times better performance in mixed workloads, 67% more writes per day, 40% quality of service, Intel said.

"The fundamental thing that Optane does as an SSD is fix performance bottlenecks in other slower media ... most people think of hard drives as being a slower media, but believe it or not, both SATA SSDs and NVMe NAND SSDs are getting slower on a key metric, which is 70/30 read/write random performance," David Tuhy, vice president and general manager of Intel data center optane storage division, said.

"Basically when you're using an SSD to hold hot data, active data, where you're doing a lot of manipulation all of our NAND SSDs, the entire industry's NAND SSDs are getting exponentially slower."

The sort of workloads where Optane is especially useful, according to Tuhy, is allowing cloud vendors to handle many virtual machines on chips with many cores, without the storage part of the equation holding the system back.

intel-p5800x-kiops.png
Image: Intel

In testing against its recently launched D7-P5600 SSD on Aerospike, Intel SSD produced 480,000 transactions per second while the P5800X hit 2 million transactions per second, and had not one single failure.

"Nano SSDs could do okay on some transactions per second, but they have a lot of jitter and they have a lot of failures," Tuhy said.

"The fundamental problem with NAND is every once in a while you have to do something called garbage collection, or you have to ... move data different blocks so you can age the SSDs better, and those situations cause lots and lots of jitter, lots and lots of quality of service actions."

On the boost in performance on mixed workloads, Tuhy put this down to the bidirectional line rate in PCIe and every pin being able to do full reads and writes at the same time. In this way, he said the Optane storage was able to run quicker than the theoretical limit of the specification.

"That's very stunning ... never seen that before," he said.

"Now in the reason that line goes down when you get to 0% reads, which is now 100% writes are just because we're not using both channels of the bus anymore."

The P5800X will be available in 400GB, 800GB, 1.6TB, and 3.2GB volumes, and use the U.2 form factor. Intel said the drive is in production and it has begun to ship its generic version, while working on a version for OEMs.

Intel also announced the successor to its H10 SSD that combined Optane and NAND storage for the thin laptop market, the H20 that uses the M.2 form factor.

Using PCIe 3.0, the H20 will be available in 512GB or 1TB of 144-layer QLC NAND storage attached to 32GB of Optane memory. The H20 features the new Optane 3D Xpoint media, new NAND and Optane controllers, and Pyrite 2 security. Support for the chip will be limited to 11th generation Intel Core U processors. The H20 will be available in the second quarter of 2021.

A cheaper, more mainstream option of the H20 was also announced, with Intel unveiling its 670p SSD that also uses 144-layer QLC NAND and features updated controllers, Pyrite security, and power loss notification. The 670p is pencilled in to be available in the first half of 2021 and will be offered in 512GB, 1TB, and 2GB options.

Taking 144-layer NAND into the data centre will be the D7-P5510 that uses TLC 3D NAND, and the D5-P5316 that uses QLC NAND.

The D7-P5510 is available today and offered in 3.84TB and 7.68TB capacities via the U.2 form factor, while the D5-P5316 will be targeting cloud storage with its capacity as either a 15TB or 30.72TB drive, which would allow for 1PB of data to be stored in a 1U rack.

The D5-P5316 will cater for U.2 and E1.L form factors when it arrives in the first half of 2021.

Speaking to journalists, Intel senior strategic planner and product manager Jonmichael Hands said the company had worked on a new trim architecture for the D7-P5510.

"Before trim took a long time and blocked I/Os and many people in Linux and other software developers actually disabled it by default because of the performance impact," he said.

"But this is very bad, you want to use trim because it enables better SSD longevity, higher SSD endurance, and higher SSD performance over time, so with the new trim architecture we can trim the entire capacity of the drive, 7.68TB in under a hundred milliseconds -- on the previous generation product this took up to 30 seconds, so this is over a 100x improvement and we are seeing large benefits in cloud storage databases."

The PCIe 4-powered D7-P5510 will also feature dynamic multiple namespaces, and be able to support up to three different power states.

"This is very good for multi-tenant applications -- for containers for virtualised environments -- to split up the NVMe SSD into up to 128 namespaces and each namespace can be configurable with its own key with opal 2.0, and have an encryption key per namespace ... and not accessible to any other namespace without the key," Hands said.

Intel said its third generation of Optane persistent memory is code-named Crow Pass, and will run on future Sapphire Rapids Xeon Scalable processors. The 10nm server chips are expected to land during the second half of 2021.

The company confirmed it intends to eventually replace TLC storage with Optane plus QLC.

Related Coverage

Editorial standards