- Sandsk’s 256TB SSD completely leaves a cache, which raises concerns about short burst work load performance
- Public standards or high -speed claims are unauthorized without the IOPS Performance Number
- Directly writing QLC can sacrifice speed for high reliability and data integrity
Sandsk has announced 256TB SSD, Ultra QLCSN 670, which is ready to ship in the first half of 2026.
This model represents the largest SSD ever by the company, which identifies a bold step towards a high -density storage solution for AI and Hypers Scale Infrastructure.
Although the company plans to release the 128TB version to the testers within weeks, the full commercial availability is at a distance of months.
You can like
A architecture that is made for measurement, not speed
In its basic part, the SN670 218 layer is built on the 3D 3D nand architecture and has 2TB dye properties (is directly connected to the CMOs).
It connects PCIE Gen5 through the NVME interface and is part of the Sandesk’s new Ultra QLC platform.
Unlike traditional SSD, which buffer data through the Sedo-SLC catches, this model uses the “directly written QLC” approach.
This simplifies the writing process and causes maximum power loss to the drive, but it also introduces the trade office, especially when it comes to performance under heavy or short burst loads.
Without SLC cache, SN670 can gradually suffer from writing short burst writing, conflicting performance under the load, and increasing the demand for controller, which makes it less accountable during extremely or unexpected workloads.
However, the Sandesk claims that the SN670 provides 68 % faster reading and 55 % randomly written compared to a leading 128 TB general 5 QLCSD.
The sequence reading speed is better than 7 %, while the internal comparison is better than 27 %.
Sandesk has emphasized the benefits such as dynamic frequency scaling, which is said to improve performance at the same strength level by 10 %.
It also claims that data retention profile can reduce recycling wear by more than 33 %.
The purpose of both features is to increase longevity and reduce energy consumption.
However, none of these claims are supported by the performance data such as reading/writing speed or endurance data.
Internally, the ultra -QLCSN 670 is supported by the customs controller and firmware, which Sandesk says enables better delays and bandwidth, but without the original benchmark or IOPS comparisons, these statements remain marketing estimates.
It is worth considering the first repetitions of the Sandesk enterprise drives using QLC Nand as compared to TLC -based models.
In this case, local QLC programming can reach 800-1200 micro -seconds, which is many times slower than SLC -based designs.
Sandesk can rely on improvements such as large dramas buffers or advanced dye parallel, but such architectural details are yet to be confirmed.
The final product will initially reach the U.2 format, which will later expected more variations in 2026.
For now, Sandesk’s 256TB drive is a symbolic jump toward future data infrastructure, not a realistic option for mainstream users.
Through blocks and files
