TL;DR:
- DDN introduces an upgraded A1400 X2 ExaScaler array for AI and machine learning workloads.
- The array incorporates QLC SSDs and compression for higher capacity and performance.
- DDN’s A1400 X2 array delivers up to 10x more data storage compared to competitors.
- The system features 60TB QLC drives and offers 1.45PB storage capacity in a compact chassis.
- It achieves 3.5 million IOPS and 95GBps throughput with a multi-core RAID engine.
- Expansion trays can be added for up to 11.7PB effective QLC capacity.
- The array provides cost savings and faster performance than NFS arrays.
- DDN caters to data center AI storage budgets with options for different performance levels.
- The A1400 X2 QLC array is ideal for larger model datasets in AI training and has potential in gaming, drug discovery, and autonomous driving.
- DDN’s design eliminates the need for internal switches and networks, reducing complexity.
- Competition intensifies with VAST Data’s scale-out NAS system.
- QLC flash technology has become mainstream, with 60TB drives enabling increased capacity.
- DDN plans to ship A1400 X2 QLC systems between June and August, offering an efficient storage solution for growing AI workloads.
Main AI News:
DDN, a leading provider of data storage solutions, has recently introduced an enhanced version of its A1400 X2 ExaScaler array, specifically designed to cater to the demanding requirements of AI and machine learning workloads. This advanced storage solution incorporates QLC SSDs (Quad-Level Cell Solid-State Drives) and introduces a cutting-edge compression facility, enabling higher capacity and enhanced performance.
The ExaScaler array is built upon DDN’s highly regarded Lustre-based scale-out and parallel file system software. By utilizing QLC SSDs, which offer a 4bits/cell format, the array can store more data within each die compared to traditional TLC (Triple-Level Cell) SSDs, albeit at a slightly slower IO speed and shorter endurance. DDN’s compression feature has been meticulously optimized to deliver exceptional performance and efficiency specifically tailored for high-performance computing (HPC) and AI workloads.
Kurt Kuckein, DDN’s Vice President of Marketing, highlighted the significant rise in enterprise customer interest surrounding DDN’s systems over the past few years, largely driven by the growing adoption of AI algorithms. Kuckein emphasized the recent surge in interest in their AI solutions, particularly in conjunction with Nvidia SuperPOD systems, driven by the broad interest in generative AI and platforms like ChatGPT.
James Coomer, Senior Vice President of Products at DDN, revealed that the company currently supports approximately 48 A1400 X2 arrays, powering Nvidia’s largest SuperPODs. He further noted that most SuperPODs worldwide rely on multiples of the same unit, highlighting the widespread adoption of DDN’s solutions.
By incorporating QLC flash technology and compression capabilities into the A1400 X2 array, DDN aims to deliver superior performance and excellent flash capacity to its customers. The system boasts the ability to provide up to 10 times more data storage compared to competing solutions while also significantly reducing energy consumption. The array utilizes 60TB QLC drives, enabling an impressive 1.45PB of storage capacity within a compact 2RU x 24-slot chassis. This doubling of capacity per watt sets it apart from other suppliers’ offerings, which typically provide 30TB SSDs.
The A1400X2 QLC array is equipped with a standard A1400X2 controller housed in a 2RU chassis. It features 732TB of TLC SSD storage and boasts a multi-core, real-time RAID engine and controller combination capable of delivering an astounding 3.5 million IOPS and 95GBps of throughput. To further expand capacity, users have the option of adding two, four, or five SP2420 QLC SSD expansion trays seamlessly connected via NVMe/oF and Ethernet interfaces. Each expansion tray can accommodate up to 2.9PB of raw QLC storage capacity, resulting in 2.3PB of usable capacity, which can be effectively expanded to 4.7PB through compression.
In a fully configured system, the maximum effective QLC capacity reaches an impressive 11.7PB. DDN claims that this new array offers up to 80 percent cost savings compared to equivalent capacity TLC arrays while also enabling applications to run faster than with NFS (Network File System) arrays. The client-side compression feature utilizes the array’s CPU cycles, but due to the resulting smaller dataset, the overall read and write performance remains on par with non-compressed data.
The new array boasts an impressive price per flash TB compared to its existing TLC (Triple-Level Cell) counterpart while delivering enhanced IOPS performance of up to 70 million and superior throughput per rack. DDN offers a hybrid TLC flash/disk system for customers seeking an even lower price per TB. With its versatile A1400X2 QLC array, DDN caters to data center AI storage budgets across three tiers: optimized for sheer performance, price/performance, or lower costs.
The prevailing belief in the industry is that larger model datasets yield superior results in training machine learning models. This principle aligns perfectly with DDN’s A1400X2 QLC array, making it an attractive choice for AI applications. DDN also envisions the array’s potential in diverse fields, such as realistic 3D and immersive gaming environments, protein and molecule simulations for drug discovery, and autonomous driving.
Unlike scale-out NAS (Network-Attached Storage) systems that rely on internal switches and networks, DDN’s A1400X2 QLC system design eliminates the need for these components. By simplifying the architecture and removing these elements, the array reduces rackspace occupancy, cost, and management complexity.
James Coomer, Senior Vice President of Products at DDN, emphasized the limitations of current QLC scale-out NAS systems, citing their inefficiency in terms of IOPS, throughput, and latencies, rendering them unsuitable for high-performance environments such as AI, machine learning, and real-time applications. DDN’s A1400X2 QLC array aims to address these limitations and provide a more efficient storage solution for these demanding workloads.
With VAST Data recently announcing SuperPOD certification for its scale-out NAS system and highlighting the complexity of parallel file systems compared to NAS, the competition between DDN and VAST Data for the same customer base intensifies. Customers new to AI model training who currently rely on NAS rather than a parallel file system may lean toward VAST Data as their preferred choice. However, existing parallel file system users may find that the A1400X2 QLC array seamlessly integrates into their workflows, offering a hassle-free transition compared to a NAS-based system.
The advent of QLC flash arrays from major players like NetApp, Dell, Pure Storage, and VAST Data further solidifies QLC flash as a mainstream technology. The introduction of 60TB QLC SSDs, with Solidigm’s upcoming release, enables a significant capacity increase over the conventional 30TB drives. While other NAND fabricators and SSD suppliers are likely to follow suit, it remains to be seen if Micron will focus on higher-capacity TLC drives built with 232-layer technology.
DDN plans to commence shipping its A1400X2 QLC systems between June and August, marking a significant milestone in the evolution of AI storage solutions. As businesses continue to grapple with growing AI workloads, DDN’s innovative array provides a compelling option to address their storage requirements efficiently and effectively.
Conlcusion:
DDN’s upgraded A1400 X2 ExaScaler array for AI and machine learning workloads introduces significant advancements in storage capacity and performance. By incorporating QLC SSDs and compression technology, the array offers up to 10 times more data storage compared to competitors while delivering cost savings and faster performance. With 60TB QLC drives and a multi-core RAID engine, the system achieves impressive IOPS and throughput.
Expansion trays can be added for further capacity expansion, catering to growing AI workloads. DDN’s design eliminates the need for internal switches and networks, reducing complexity. The competition intensifies with VAST Data’s scale-out NAS system, and QLC flash technology becomes mainstream with increased capacity. DDN plans to ship the A1400 X2 QLC systems between June and August, providing an efficient storage solution for AI workloads. Overall, this market development signifies a significant advancement in AI storage capabilities, offering enhanced performance and capacity for businesses dealing with growing AI demands.
