AN INDUSTRY FIRST
On 21 May, Micron began shipping and selling the industry's first QLC SSD, delivering the next evolution in storage that delivers fast capacity for less. Micron's QLC technology provides the features, capacity, performance and value to challenge HDDs in environments where performance is critical and the cost of SSDs has been too high. With QLC SSDs, you can quickly access and analyse the read-intensive, performance-sensitive workloads that drive your business - such as real-time analytics, machine learning, artificial intelligence, Big Data, content delivery, user authentication and more.
THE THREE MOST IMPORTANT ADVANTAGES OF QLC SSDS
1. LOWER TCO
While performance-driven, read-oriented workloads rely on massive HDD arrays to deliver the results users demand, QLC storage gets by with fewer drives - lowering your performance-driven total cost of ownership (TCO)*.
2. MORE CAPACITY
QLC stores more bits per cell than its predecessors, enabling immense increases at system, rack and data centre levels that are more cost-effective.
3. LESS SPACE REQUIRED
QLC technology's ability to increase density per server reduces rack space requirements by up to 7.7 times compared to 2TB HDDs, saving valuable and expensive data centre space.
*All comparisons based on a 3.5-inch 8TB 7200RPM HDD. TCO statement based on common read-centric workload of 128K 95% Read, estimates for similar performance (total capacity varies); 3X lower power comparisons based on public power consumption data for HDDs and Micron 5210 Enterprise QLC SSD; 41% less rack space based on 42U rack (fully populated), 50PB data set, 2. 5-inch 7.68TB Micron 5210 Enterprise QLC SSDs, 8TB 3.5-inch Enterprise SATA HDDs and high-density 2U servers that can support 32X 2.5-inch or 18X 3.5-inch form factor devices per enclosure, RAID 0
IDEAL APPLICATIONS FOR QLC ENTERPRISE SSDS
Give these applications a performance boost,
more cost-effectively by using high-capacity, lightning-fast QLC-based SSDs.
- Real-time analytics and Big Data: Use Big Data applications such as Hadoop Distributed File Systems.
- Business Intelligence and Decision Support Systems: Sift through massive data sets with faster and deeper queries for real-time business insights and decision making.
- Active Archives and Large Block Storage: Transform scalable active archives into a strategic asset and effortlessly deliver massive large-block data streams.
- Read-intensive artificial intelligence: Get the speed AI algorithms depend on to quickly identify patterns in large datasets.
- Machine and deep learning: Crunch immense data sets in seconds instead of minutes - because machines can only learn as fast as they can read and analyse data.
- Content Delivery, Video on Demand, Media Streaming: Deliver more content to more users, consistently with support for massive, parallel requests and streams.
- NoSQL databases: Bring to life data-driven workloads such as content classification and tagging, and user profile acceleration.
- User authentication: Perform fast authentication with fast storage.
Workload | Micron 9300 (TLC) | Micron 5210 (QLC) |
|---|---|---|
AI & Machine Learning | Directly connected memory in the node (intermediate storage of training data) | Central storage for the over 50 TB data lakes that feed model training and development |
NoSQL databases (Cassandra, MongoDB) | Accelerate write-intensive and mixed-use workloads | HDD shift for read-intensive workloads |
Big Data & Real-Time Analytics (Hadoop) | YARN caching | HDFS HDD shift for faster time-to-insight |
Block and object storage (Ceph) | Small random block memory | Object Storage & Media Streaming |
vSAN | Caching layer | Caching layer |
SQL databases | OLTP | BI/DSS (business intelligence) |
High-frequency trading | Comparative analytics, acquisition of transient data sets, complexity reduction (systems, drives, support) | HDD shift for backtests, regulatory reports and to store records that need to be retrieved quickly |
To the product | MICRON 9300 (TLC) |
HOW MUCH IS YOUR TIME WORTH?
The following table should give you an indication of how time-efficient it is to replace conventional HDDs with QLC SSDs.
ML Workflow # Number of images Size of the data set | 7.2K HDD (8TB) Completion time | 5210 SSD (7.68TB) Completion time | Savings Time saved Efficiency increase | ||
|---|---|---|---|---|---|
1000 | 23GB | 10 Minutes | 1 Minute | 9 Minutes | 10x |
10.000 | 230GB | 90 Minutes | 11 Minutes | 1 hours 19 Minutes | 9x |
50.000 | 1.150GB | 454 Minutes | 54 Minutes | 6 hours 40 Minutes | 8x |
100.000 | 2.300GB | 910 Minutes | 113 Minutes | 13 hours 17 Minutes | 8x |
5 INTRODUCTIONS
- The more ML you do, the more time you can save
- The larger your data set, the more time you can save
- Memory is often a bottleneck in ML workloads, as the size of the dataset often dwarfs the installed memory capacity
- The long-term cost of low-power GPUs/CPUs often far exceeds the cost of QLC SSDs
- The Micron 5210 is 3x more energy efficient, so you save on power and cooling too