Dispatch in 3–7 working days · 1-year warranty on every part Call A Rep: +86 13537510010 | Request A Quote
Inspected & Load-Tested Before Dispatch Serial Recorded Dispatch in 3–7 working days
MZWL93T8HFLT · Samsung

Samsung PM1753 3.84TB PCIe 5.0 NVMe U.2 SSD (MZWL93T8HFLT)

3.84 TB 2.5-inch U.2 NVMe SSD, PCIe 5.0, ordering code MZWL93T8HFLT-00AW7

New Used
Pricing Request quote
MOQ 2 pc · volume pricing on request
Reply Within 24h
Volume Pricing
Worldwide Shipping
Secure Anti-Static Pack
Part Number
MZWL93T8HFLT
Brand
Samsung
Category
Enterprise & Data Center SSD
Accepted Payments
T/T Bank Transfer PayPal Western Union

Technical Specifications

Capacity 3.84 TB
Interface NVMe
Form Factor U.2 15mm
PCIe Version PCIe 5.0
NAND Type TLC
Endurance (DWPD) 1
TBW 7,008 TB
Sequential Read 14,000 MB/s
Sequential Write 9,000 MB/s
Random Read IOPS 3,200K IOPS
Random Write IOPS 440K IOPS
Power 17 W
DRAM Cache Yes
Power Loss Protection Yes
Encryption AES-XTS 256
Condition New · Used

Frequently Asked Questions

PM1753 unless the shelf is already at its power limit. It gives 3,200K random read against 1,700K, 440K random write against 250K and 9,000 MB/s write against 6,700, for 4.2 W more per drive. Endurance is identical at 7,008 TBW, so nothing but performance and power separates them.

Because the link was never the limit on that figure. 440K of 4 KB writes is 1.76 GB/s, well inside what four Gen4 lanes carry, so only the bandwidth-bound numbers fall when the host downgrades. Sequential read drops 46 percent and random read 47 percent; random write does not move at all.

The power supply is rarely the problem; the airflow budget usually is. 24 drives at 17 W is 408 W against 307 W for 12.8 W parts, which most shelves absorb. What changes is heat density in the bay, so check cross-flow through the 15 mm gap rather than the PSU rating.

Metadata and index lookups, virtual desktop boot storms, and the read side of a write-back cache. What those have in common is small reads against a working set that fits in a few terabytes, which is exactly the shape this capacity serves. Capacity tiers and backup targets do not use it.

The line does ship in E3.S, and the part number changes prefix from MZWL to MZ3L. We list the 15.36 TB E3.S part separately; for 3.84 TB we quote the U.2 carrier on this page. Ask if you need the EDSFF version and we will confirm stock rather than assume it.

At the 440K random write peak the drive absorbs 1.76 GB/s, so the whole budget goes in about 46 days. At the rated 1 DWPD average of 44.5 MB/s it lasts the full five years. Size by sustained write rate, not by the peak, and the number stops being alarming.

Samsung PM1753 3.84 TB, 2.5-inch U.2 at 15 mm, catalogue code MZWL93T8HFLT, ordering code MZWL93T8HFLT-00AW7. 14,000 MB/s sequential read and 9,000 MB/s write on PCIe 5.0 x4, 3,200K random read and 440K random write IOPS, 7,008 TBW at one drive write a day for five years. PLP, AES-XTS 256-bit, 17.0 W reading and 16.7 W writing. New and used both stocked.

3.84 TB is the first capacity in this line that is not held back by how much flash sits behind the controller. At 1.92 TB the drive writes at 4,800; here it writes at 9,000, and the largest parts in the same family quote 10,000. Random read settles one step earlier still: 3,200K here, and the same 3,200K at every capacity above it. Above this point the money buys capacity rather than throughput, which makes this the last rung where the two are still sold together.

Technical specifications

Product familySamsung PM1753 (PCIe 5.0 high-performance data centre NVMe SSD, 1 DWPD)
PNMZWL93T8HFLT (catalogue code)
Ordering codeMZWL93T8HFLT-00AW7
Capacity3.84 TB (unformatted; 1 GB = 1 billion bytes, formatted capacity is less)
Form factor2.5-inch U.2, 15 mm (100.2 × 69.85 × 15 mm, about 165 g)
InterfacePCIe 5.0 x4, NVMe 2.0, single port, 12 V ±10%
NANDSamsung V-NAND TLC
Sequential read14,000 MB/s (PCIe 5.0) / 7,500 MB/s (PCIe 4.0)
Sequential write9,000 MB/s (PCIe 5.0) / 7,100 MB/s (PCIe 4.0)
Random 4K read3,200K IOPS (PCIe 5.0) / 1,700K IOPS (PCIe 4.0)
Random 4K write440K IOPS in both modes
Latency65 µs random read, 10 µs random write (queue depth 1)
QoS 99%90 µs read, 20 µs write, identical on PCIe 4.0
Endurance (TBW)7,008 TB
Endurance (DWPD)1 drive write per day for five years
Power17.0 W typical read, 16.7 W typical write, 4 W idle
Power-loss protectionYes, capacitor-backed, with end-to-end data protection
EncryptionAES-XTS 256-bit engine, TCG Opal compliant
Sector size512 (default), 520, 4096, 4104, 4112 or 4160 bytes
MTBF / UBER2.5 million hours / 1 sector per 1017 bits read
Drive ready2 s or less from power-on to accepting commands
Data retention3 months unpowered, 0 to 70 °C operating

The capacity where the write rating stops climbing

Between 1.92 TB and 3.84 TB the sequential write figure doubles, from 4,800 to 9,000, and one more step takes it to 10,000 at 7.68 TB. Above that it does not move again: the 15.36 TB and 30.72 TB parts quote the same 10,000. Write bandwidth is set by how many dies the controller can program at once, and by this capacity there are already enough of them to keep the programming pipeline full; the dies added above this point mostly add address space.

Sequential read does not move at all, and it cannot. Four Gen5 lanes carry 128 GT/s, which is 128 Gb/s or 16,000 MB/s of raw signalling; the 128b/130b line encoding takes that to about 15,750 MB/s, and packet overhead leaves roughly 14,000 MB/s of usable payload. This drive quotes 14,000. There is no headroom left for a larger capacity to claim, which is why the whole line sits at the same sequential read number.

Random read is the column worth reading twice. It is 2,200K at 1.92 TB and 3,200K at every capacity from here to the 30.72 TB part. So the read-IOPS ceiling is reached at this capacity and then stays flat across four times the flash. Anyone who needs 3,200K and reaches for a bigger drive pays for capacity and gets no additional queue depth for it.

Two Gen5 drives at 3.84 TB, 4.2 watts apart

ModelCapacityForm factorSeq read / writeRandom 4 KBTBWPower
PM9D3a3.84 TBU.2 2.5-inch, 15 mm12,000 / 6,700 MB/s1,700K / 250K7,008 TB12.8 W
PM17533.84 TBU.2 2.5-inch, 15 mm14,000 / 9,000 MB/s3,200K / 440K7,008 TB17 W
PM175315.36 TBU.2 2.5-inch, 15 mm14,000 / 10,000 MB/s3,200K / 620K28,032 TB20.8 W

The first two rows are the same capacity, the same shell, the same 1 DWPD rating and the same 7,008 TBW. What separates them is a controller and 4.2 W. Random read is 88 percent higher, random write 76 percent, sequential write 34 percent and sequential read 17 percent. Across a 24-bay shelf that is 101 extra watts for a cache tier that answers nearly twice as many 4 KB reads per second.

The third row is there to show where the money stops working. Four times the flash buys the same 3,200K random read as this drive and 620K random write rather than 440K, for 3.8 W more per drive. Capacity per watt improves by three times; read IOPS per watt does not improve at all. Two different reasons to buy two different drives, and this one is the second.

What a PCIe 4.0 host takes away, and what it leaves

On a Gen4 link the ratings are 7,500 MB/s read, 7,100 MB/s write, 1,700K random read and 440K random write. Sequential read loses 46 percent, sequential write 21 percent, random read 47 percent, and random write nothing.

That last one is worth understanding rather than skimming. 440K of 4 KB writes is 1.76 GB/s, comfortably below what four Gen4 lanes can carry, so the link was never the constraint on that figure. A drive used as a commit target, a log volume or a write-back cache on an older platform keeps its full random write advantage over the PM9D3a's 250K, and keeps it without spending on a platform refresh.

The comparison that decides whether an older host is worth upgrading is this one: on PCIe 4.0 this drive writes at 7,100 MB/s, which is faster than the PM9D3a manages on PCIe 5.0 at the same capacity, 6,700. Its Gen4 random read of 1,700K is exactly the PM9D3a's Gen5 figure. A Gen4 chassis therefore gives up the sequential read ceiling and nothing else that matters.

Latency is unaffected in both modes: 65 µs random read and 10 µs random write at queue depth one, 90 and 20 µs at the 99th percentile. Those are properties of the controller and the flash, not of the number of lanes, which is why a narrower link degrades bandwidth without touching tail behaviour.

Seventeen watts, measured three ways

17.0 W reading, 16.7 W writing, 4 W idle. Read above write looks wrong until the drive is understood as a device that spends its read power on the link and its write power on the flash; every capacity in this family behaves the same way. Twenty-four drives hold 92 TB and draw 408 W, or 4.43 W per terabyte. The same shelf of 15.36 TB parts holds 368 TB for 499 W, or 1.36 W per terabyte, so capacity per watt is more than three times better three steps up the range.

Random read per watt runs the other direction. 3,200K over 17.0 W is 188K IOPS per watt; the 1.92 TB part works out at 169K and the 15.36 TB part at 154K. On that basis this is the strongest capacity in the line, which is not the order the model numbers suggest. A shelf that is short of read IOPS and short of power at the same time should be filled with this part rather than with a larger one.

The idle figure belongs in the same conversation for a less obvious reason. At 4 W, 92 TB of cache powered and not being read costs 96 W across 24 bays. That number never appears on a datasheet comparison and it shows up every month on a power bill.

Seven petabytes that do not distinguish one drive from another

7,008 TBW is 3.84 TB of host writes a day for 1,825 days, or 44.5 MB/s sustained. Every 3.84 TB drive rated at one drive write a day carries the same figure, including the PM9D3a in the table above, so endurance cannot be used to choose between them and should not appear as a reason in a quotation.

The useful version of the number is how quickly it can be spent. At the 440K random write peak the drive absorbs 1.76 GB/s, and 7,008 TB lasts 46 days at that rate. That is not an argument against the drive; it is an argument for sizing by sustained write rate rather than by the peak a datasheet leads with, and for putting host bytes written into the monitoring before it goes into the purchase justification.

Retention is quoted at three months with the drive unpowered, which is the JEDEC convention rather than a limitation of this part. A spare that sits on a shelf for a year is outside that window, so power it up and let it run a full read before it goes into a live slot. It costs an afternoon and it is the cheapest failure prevention in a spares cage.

Ordering codes: MZWL93T8 and the -00AW7 suffix

CodeTypeCapacityConditionQuote
MZWL93T8HFLTCatalogue code3.84 TBNew / usedEnquire
MZWL93T8HFLT-00AW7Ordering code3.84 TBNew / usedEnquire

The suffix records the distribution block, and -00AW7 and -00BW7 both appear on this capacity for the same physical drive. The character that decides what arrives is the fifth one: MZWL9 is PM1753, MZWLO is PM1743, MZWL6 is PM9D3a. Three product lines, one character, three different controllers.

Checks worth running on delivery:

Warranty, lead time and purchasing

Both conditions are covered by one year warranty, applied to the drive itself rather than to the chassis it came out of, and a used unit is tested and wiped before packing. Stock leaves the warehouse in three to seven working days and the tracking number is issued at dispatch. Two pieces is the minimum on a line, capacities in this series can ride one order together, and a quotation comes back the same working day once the part number, condition and quantity are clear.

Get Pricing For This Part

Tell us who you are — we reply within one working day with a firm quote, stock status and lead time. Volume discounts available.

Prefer email? Write to sales@oceanserverparts.com or WhatsApp +86 13537510010.
Please tell us your name.
A valid email is required.
Need a full RFQ with multiple parts? Use The Full Form →

Related Products

Samsung PM9A3 15.36TB E1.L NVMe SSD (MZEL215THBLA)
Samsung · Enterprise & Data Center SSD
Samsung PM9A3 15.36TB E1.L NVMe SSD (MZEL215THBLA)

15.36 TB E1.L 1U long ruler NVMe SSD, PCIe 4.0, ordering code MZEL215THBLA-00AMV

New Used
Request quote
MOQ 2 pc
View
Samsung PM9A3 7.68TB E1.S NVMe SSD (MZCL27T6HBLA)
Samsung · Enterprise & Data Center SSD
Samsung PM9A3 7.68TB E1.S NVMe SSD (MZCL27T6HBLA)

7.68 TB E1.S 1T NVMe SSD, PCIe 4.0, ordering codes MZCL27T6HBLA-00AMV / MZTL27T6HBLA-00A07

New Used
Request quote
MOQ 2 pc
View
Samsung PM9A3 3.84TB E1.S NVMe SSD (MZCL23T8HCLS)
Samsung · Enterprise & Data Center SSD
Samsung PM9A3 3.84TB E1.S NVMe SSD (MZCL23T8HCLS)

3.84 TB E1.S 1T NVMe SSD, PCIe 4.0, ordering codes MZCL23T8HCLS-00AMV / MZTL23T8HCLS-00A07

New Used
Request quote
MOQ 2 pc
View
Samsung PM9A3 1.92TB E1.S NVMe SSD (MZCL21T9HCJR)
Samsung · Enterprise & Data Center SSD
Samsung PM9A3 1.92TB E1.S NVMe SSD (MZCL21T9HCJR)

1.92 TB E1.S 1T (9.5 mm) NVMe SSD, PCIe 4.0, ordering code MZCL21T9HCJR-00AMV

New Used
Request quote
MOQ 2 pc
View