Sandisk on Tuesday introduced one of the industry's first 256TB solid-state drives, designed for applications that benefit from maximum storage density. The new SSD relies on the all-new enterprise-grade UltraQLC platform that is specifically designed for drives that combine high capacity, high performance, and high reliability.

Sandisk's UltraQLC 256TB NVMe SSD comes in a U.2 form-factor and uses the company's custom new multi-core controller, custom firmware, as well as 2Tb BiCS8 3D QLC NAND memory. In addition to the 256TB model, the company also announced its SN670 SSD that relies on the same UltraQLC platform.

One of the key features of UltraQLC drives is its Direct Write QLC technology, which lets the controller write data directly to QLC memory, not to a pseudo-SLC buffer, thus enabling power-loss–safe writes on the first pass. This design simplifies the writing process and reduces latency. However, the performance impact of such an approach is something that remains to be seen (more on this later).

In addition, the UltraQLC 256TB NVMe SSD features Dynamic Frequency Scaling, which purportedly boosts performance by 10% at any given power level, though the company does not disclose how this technology works. It is likely that DFS optimizes controller frequencies and interface speed depending on the workload, though we are speculating.

Also, Sandisk's UltraQLC 256TB SSD features a Data Retention profile that can reduce retention-related recycling by up to 33%, which promises to improve reliability, resilience, and energy efficiency.

Sandisk positions its UltraQLC platform as a strategic solution for hyperscale cloud service providers (CSPs) and enterprises building AI data infrastructure and requiring storage density, efficiency, and performance.

Speaking of performance, the lack of pseudo-SLC caching certainly has an effect on the performance of UltraQLC drives. Normally, 3D QLC NAND is three to five times slower and has higher latency compared to pseudo-SLC, so the 256TB drives are not for everyone. There are some things to keep in mind here.

Native QLC programming latency is significantly higher compared to pseudo-SLC (~800–1200 µs vs. ~200–300 µs for SLC), so sustained sequential throughput is lower than pseudo-SLC-cached SSDs for short bursts. However, since there is no limited pseudo-SLC cache, performance is consistent over long writes, which may be beneficial for large AI data sets that require hours to write anyway. Also, Sandisk can mitigate slow writes with a large DRAM buffer and smart mapping, multi-plane and multi-die parallelism in the memory device and the controller, and over provisioning (keep in mind that we are speculating).

Sandisk's 128TB and 256TB SSDs will ship in the first half of 2026.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved