As the artificial intelligence (AI) sector continues to gain momentum, DDN, a storage array manufacturer, is gearing up to meet the impending demands of enterprises expected to invest billions in AI technologies. With a long history of providing storage solutions for high-performance computing (HPC) clusters, DDN is now adapting to the growing AI landscape.
According to DDN’s co-founder and president, Paul Bloch, existing HPC compute clusters utilizing DDN storage could see orders triple or even quadruple by 2027. He mentioned significant deployments, such as the 200,000 GPUs in Elon Musk's xAI cluster, which is anticipated to expand to one million units. Bloch noted that the rapid performance growth of GPUs is heavily influencing data storage needs, as the requirement for real-time data ingestion becomes critical.
During a recent meeting in Silicon Valley, DDN’s CTO, Sven Oehme, highlighted the promise of the enterprise market for AI. He remarked, "AI hasn’t yet reached its full potential, but businesses can foresee the possibilities ahead. The real surge will occur when enterprises leverage AI across various sectors." DDN, currently valued at $5 billion, forecasts a revenue of $1 billion by 2025, bolstered by a recent $300 million investment from Blackstone.
Oehme also indicated that this collaboration with Blackstone would enable DDN to penetrate new customer segments previously untouched. He emphasized the potential for significant AI investments from unfamiliar companies, particularly in industrial applications where management is looking for cost-saving opportunities through AI technology.
In a bid to solidify its position, DDN has refreshed its core product line, unveiling the EXAscaler AI400X3. Building on the previous AI400X2 Turbo, this new model incorporates 24 NVMe SSDs and additional SSD shelving support via NVMe-over-TCP.
The performance enhancements include a 55% increase in read speeds and over 70% for writes, achieving throughput of 140GBps for reads and between 75GBps to 100GBps for writes. This improvement is largely attributed to the integration of four Nvidia BlueField cards operating under the SpectrumX protocol.
"We have a reference architecture for the new Blackwell GPU," said Oehme, noting the ongoing testing and development using DDN equipment. Security and pooling improvements have also been incorporated, offering MSPs enhanced dynamic provisioning for cloud services.
For enterprises, the EXAscaler AI400X3 introduces data compression capabilities that allow for usable capacity significantly exceeding its raw capacity, aligning with offerings from companies like Dell and HPE.
In terms of resilience, DDN has made substantial advancements, allowing the replacement of hardware or software components without interrupting the storage array's operations. This ensures minimal downtime, a critical factor for enterprise customers.
While the EXAscaler AI400 primarily operates in file mode, DDN recognizes the shift towards object storage among enterprise AI users. Although access speeds may be lower, object storage offers advantages in capacity and content labeling for various AI applications.
With the launch of Infinia 2.1, DDN claims a remarkable performance boost, achieving speed enhancements up to 100x compared to previous versions. It now rivals other accelerated object storage options, such as HyperStore and Ring, particularly for GPU support.
Infinia 2.1 is expected to be ten times faster in access time and twenty-five times faster in handling requests compared to AWS’s S3 Express. The solution's compatibility extends to numerous AI software stacks and has added integrations with observability platforms like Datadog.
Besides Infinia, DDN's range includes xFusionAI and Inferno, both designed to enhance accessibility and performance in AI environments.
DDN reports that they have successfully implemented storage solutions supporting 700,000 GPUs across 7,000 clients, with 4,000 of those actively utilizing AI workloads. The company, employing around 1,100 individuals, intends to hire an additional 200 to 250 staff members this year.
Founded in 1998, DDN has roots in providing storage solutions for supercomputer centers and has significantly contributed to advancements in data throughput capabilities. Bloch emphasizes the complex infrastructure needed to support high GPU counts effectively, stating, "There are few systems globally capable of achieving such throughput. That's our expertise."
