Best 15TB SSDs: Choosing a 15.36TB NVMe or SAS Enterprise Drive

A 15TB SSD in an enterprise catalogue is almost always a 15.36TB drive, the next step above 7.68TB in the read-intensive capacity series. It is the size at which flash starts replacing disk on capacity, not just speed, because one 2.5-inch bay now holds several times what a 2.5-inch hard drive can. The best 15TB SSD for you comes down to three questions: does your server have NVMe or SAS bays, is it PCIe 4.0 or 5.0, and is the data read-heavy enough for QLC? Below are the 15.36TB drives Nexus Compute supplies, grouped the way buyers actually choose between them.

What are the best 15TB SSDs available on the market?

• PCIe 5.0 NVMe, U.3: the Kioxia CM7-R, rated up to 14,000 MB/s read and 6,750 MB/s write, and the Samsung PM1743. For Gen5 servers where local flash feeds GPUs or high-throughput analytics.
• PCIe 4.0 NVMe, TLC: the Micron 9400 PRO (176-layer TLC) and Kioxia CD6-R on U.3; the Samsung PM1733 and Solidigm D7-P5520 (144-layer TLC) on U.2. The mainstream choice for Gen4 servers.
• PCIe 4.0 NVMe, QLC: the Solidigm D5-P5430 and D5-P5336, both 192-layer QLC on U.2 and listed as read-intensive.
• Dell PowerEdge: the Dell 15.36TB NVMe Read Intensive drive (PCIe 4.0, U.2 hot-plug, Dell carrier included), for PowerEdge estates that standardize on Dell-branded drives.
• 24Gb/s SAS, dual port: the Samsung PM1653 and Kioxia PM7-R, rated up to 4,300 MB/s read.
• 12Gb/s SAS, dual port: the Samsung PM1643a, Seagate Nytro 3350 (3D eTLC) and HGST Ultrastar DC SS530 (3D TLC), all rated up to 2,200 MB/s read.

NVMe or SAS at 15.36TB: which should you choose?

Choose NVMe when the server has NVMe bays and the drives are a performance tier. A 15.36TB drive puts a lot of data behind one interface, and NVMe gives that data a direct PCIe x4 path with no SAS controller in between. PCIe 5.0 drives such as the CM7-R double the per-lane bandwidth of Gen4, which pays off when the drives feed GPU training or large scans. Our article on PCIe Gen5 NVMe performance for AI storage covers when that extra bandwidth is actually used.

Choose SAS when the drives go into a dual-controller array, a JBOD shared by two hosts, or behind a hardware RAID controller you intend to keep. Every SAS drive listed here is dual-port, which is the basis of controller failover. SAS also scales to large drive counts through expanders, whereas NVMe needs PCIe lanes or a PCIe switch for every drive. The trade is a lower throughput ceiling per drive, which matters less when many drives work in parallel. The 12Gb/s drives from Samsung, Seagate and HGST remain a sensible way to add capacity to an existing 12G array without replacing its controllers, while the 24Gb/s PM1653 and PM7-R suit new SAS deployments and 24G controller upgrades.

U.2, U.3 or E3.S: which 15.36TB form factor fits your server?

U.2 and U.3 are both 2.5-inch NVMe drives on the SFF-8639 connector. U.3 drives are designed for tri-mode backplanes, where a single bay accepts SAS, SATA or NVMe through a tri-mode controller, and they generally also work in U.2 bays. A U.2 drive usually will not run in a U.3 tri-mode bay. Of the drives above, the Micron 9400 PRO, Kioxia CD6-R and CM7-R and Samsung PM1743 are U.3; the Samsung PM1733, the three Solidigm drives and the Dell drive are U.2.

E3.S is the EDSFF form factor that newer Gen5 server designs use in place of 2.5-inch bays. It is a different mechanical format with a different connector, so an E3.S bay needs an E3.S drive. None of the 15.36TB drives in this guide are E3.S. If your platform uses E3.S bays, send the server model and we will check what is available for it rather than guess.

When does a 15TB-class SSD make sense?

The case for 15.36TB is density. A 24-bay 2U server holds about 369TB raw with 15.36TB drives, against about 184TB with 7.68TB drives. For a fixed capacity target that halves the servers, the bays and, in many cases, the software licenses counted per node. It also leaves free bays for growth instead of forcing a new chassis, and fewer servers means fewer power supplies, switch ports and management endpoints to run. It is the right size when:

• The server is bay-limited, and capacity rather than IOPS per terabyte is the constraint.
• The data is read-heavy: analytics, file and object tiers, content libraries, AI training datasets and flash backup targets.
• You are consolidating several older servers, or a hard-drive tier, onto fewer flash nodes.

It is the wrong size when too much data has to share one drive's performance. A 15.36TB drive has the same x4 interface as the 7.68TB model in its family, so IOPS per terabyte roughly halves. For a busy transactional database, two 7.68TB drives of the same model will generally serve more I/O than one 15.36TB drive.

How should you plan RAID and rebuilds with 15.36TB drives?

Rebuild time scales with the data that has to be reconstructed, so a 15.36TB drive takes roughly twice as long to rebuild as a 7.68TB drive on the same controller and load. Flash rebuilds far faster than a hard drive of the same size, but the array still runs degraded for the duration. Plan for it:

• Use dual parity (RAID 6 or equivalent erasure coding) for large SSD groups, so a second failure during the rebuild does not cost data.
• Keep RAID groups narrower, or use distributed-spare layouts where your controller or storage software supports them.
• Hold a hot spare or shelf spare of the same model, so the rebuild starts at once.
• Remember that a rebuild writes a full drive to the replacement. At 1 DWPD, a 15.36TB drive is rated for about 15.36TB of writes per day, which is ample headroom for read-heavy tiers. Our guide to RAID configurations for large AI storage goes deeper on group sizing.

Is a QLC 15TB SSD good enough?

For read-heavy data, yes. The Solidigm D5-P5430 and D5-P5336 store four bits per cell and are rated read-intensive, with the same PCIe 4.0 x4 interface as the TLC drives. QLC is weaker on sustained and random writes and has lower write endurance, so it belongs on tiers that are written once and read many times. If the tier also takes steady writes such as logs, VM churn or ingest, choose TLC: the Solidigm D7-P5520, Micron 9400 PRO or Kioxia CD6-R.

Buy 15.36TB for density, protect it with dual parity, and keep QLC for data that is mostly read.

Nexus Compute supplies every drive in this guide new, through authorized distribution, with the full manufacturer warranty, and checks compatibility with your server before dispatch. Send the server model and quantity through request a quote for pricing within 48 business hours.