Quad-level cell (QLC) NAND is a relatively recent development in the world of flash storage. Micron pioneered the technology, selling the first QLC solid state drive (SSD) in May of 2018. QLC has become the flash memory of choice for a number of different use cases, including video streaming, real-time analytics, and big data applications. In particular, QLC provides massive benefits to machine learning applications, making the processing workflow faster and easier than ever before.
How does QLC NAND work?
The most obvious difference between QLC and older, triple-level cell (TLC) NAND is the number of bits: four bits per cell instead of three. This results in a 33% higher density of storage, allowing data centers to increase their overall capacity while conserving space. Compared to hard disk drives (HDDs), QLC storage requires 41% less rack space on average. QLC storage also consumes 3x less power, providing significant cost savings for large data centers.
Read-intensive workloads
Let’s examine why machine learning protocols are perfect for QLC NAND. Machine learning is essentially a system that can improve its decision-making capabilities and look for patterns in data without needing human intervention. A necessary component of machine learning is the analysis of data— massive amounts of data, in fact.
The read speed of QLC is leaps and bounds ahead of earlier generations of storage, making QLC the ideal storage choice for the read-intensive workload of a machine learning application. Compared to traditional data centers, which typically feature a 4:1 read-to-write ratio, AI and machine learning applications have an incredible 5000:1 read-to-write ratio.
Compare the performance of the Micron® 5210 QLC SSD compared to a series of HDD models from other manufacturers:
The Impact of HDD Throughput Limits & Reliablity Concerns
Drive |
Capacity |
Workload Rating (TB/Year) |
DWPD |
5210 Advantage |
|---|---|---|---|---|
| Micron® 5210 (QLC) | 7.68TB | 2,242* (and only limited on writes) |
0.80 | N/A |
| Vendor B 7.2K HDD | 8TB | 550 | 0.19* | 4x |
| Vendor B 7.2K HDD | 10TB | 550 | 0.15* | 5x |
| Vendor B 7.2K HDD | 12TB | 550 | 0.13* | 6x |
| Vendor C 7.2k HDD | 14TB | 550 | 0.11* | 7x |
*Numbers arent on datasheets, but can be calculated as folllows based on sequential transfers: Workload Rating: DWPD x capacity x 365 days per year DWPD: (Workload Rating / 365 days per year) / capacity |
||||
The 5210 boasts a workload rating of 2242TB per year, over quadruple the performance of the older HDDs. When processing terabytes of data every day, the increased speed of the QLC allows for faster breakthroughs and more agile product development. Instead of waiting days to complete a set of data analysis, QLC users only need hours.
The machine learning workflow
Let’s take a look at a real-life example of how the increased read speed of QLC can impact a machine learning workflow. Colfax International, a machine learning solution provider, tested the Micron 5210 ION against a Seagate 7200RPM HDD for the speed to access a TFRecord, a dataset for the popular end-to-end open source machine learning application TensorFlow. Developers use TFRecords to store massive amounts of data, such as a collection of images or videos, in a single file for faster reading and analysis.
For the test, the Colfax team used large high-resolution images to simulate the process of how applications would pre-process data in a deep learning training environment. The results were nothing short of astounding. As the number of processes scaled, the Micron 5210 continued to outperform the HDD. To run through one hundred processes, the HDD needed 413 seconds, compared to only 64 seconds for the Micron 5210.
“We found that a 2.3TB, 100,000 image dataset took 15.17 hours on our HDD platform while the Micron 5210 ION completed the same task in just 1.87 hours, resulting in 13 hours of savings on an everyday task. The more machine learning you do and the bigger your dataset, the more your time savings will compound,” commented Colfax CEO Gautam Shah.
The future of QLC in machine learning
With the rise of popularity in machine learning from everything ranging from facial recognition to how Netflix recommends movies, QLC will become the standard storage solution for developers looking to analyze large datasets. With lower NAND storage bit costs, reduced rack space, and higher energy efficiency, QLC will eventually replace slower, larger, costlier HDDs across the development ecosystem.
Micron 5210 QLC SSDs offer 175x faster random reads and 30x faster random writes, with half the floorspace of typical 3.5-inch 7200 RPM HDDs. To learn more about the Micron 5210, click here.
Tags: Storage, Kris Sharma
Get to Know the Author:
Kris Sharma is a content creator living in Boise, Idaho. He writes frequently on technology topics, including automation, machine learning, and data security. Feel free to hit him up on LinkedIn.
The opinions expressed in these articles are those of the individual authors and not Micron Technology, Inc., its subsidiaries or affiliates. Upgrading your systems and components can cause damage to the system or components, including potential data loss. Micron is not responsible for any damage or harm, including data loss or system interruptions, that may occur. All information is provided “AS-IS” and neither Micron nor the author make any representations or warranties with respect to the information provided. Neither Crucial nor Micron Technology, Inc. is responsible for omissions or errors in typography or photography. Micron products are warranted as provided for in the products when sold, applicable data sheets or specifications. Information, products, and/or specifications are subject to change without notice. Micron, the Micron logo, Crucial, and the Crucial logo are trademarks or registered trademarks of Micron Technology, Inc. Any names or trademarks of third parties are owned by those parties and any references herein do not imply any endorsement, sponsorship or affiliation with these parties.