Unveiled at ISC 2026 in Hamburg, DDN’s new A1400X3M storage appliance delivers 160 million IOPS per rack and integrates NVIDIA Dynamo-powered KV Cache software to enable dual-path AI acceleration. The update also brings enhanced security, observability and infrastructure efficiency for large-scale AI deployments, targeting common data pipeline bottlenecks across ingestion, preparation, training, inference, RAG and agentic AI workflows.
DDN Co-Founder and CEO Alex Bouzari emphasized modern AI success hinges on end-to-end data operational efficiency rather than just compute power. DDN’s latest innovations maximize GPU utilization, cut inference costs, speed up token generation and improve overall economics for large-scale AI factories.
A1400X3M Hardware Upgrade
Serving as the successor to the A1400X2, the new A1400X3M runs DDN’s EXAScaler Lustre software. Its 2RU 24-SSD chassis upgrades from Intel Ice Lake CPUs to AMD Genoa processors and PCIe Gen4 to double-speed PCIe Gen5 interfaces. It delivers up to 190 GBps read throughput and 110 GBps sequential writes, marking a 35% read performance improvement over the prior generation. A full rack deployment achieves 160 million IOPS, equating to roughly 4 million IOPS per chassis.
NVIDIA Dynamo-Integrated KV Cache Acceleration
KV cache stores LLM activation keys and values during inference to eliminate redundant recomputation and boost performance. NVIDIA’s Dynamo distributed inference framework extends GPU KV cache to NVMe storage via a four-tier memory hierarchy (HBM, DRAM, local SSD, external NVMe), turning storage-resident cache into persistent, addressable context memory. As a qualified NVIDIA NVMe partner, DDN integrates this capability across its EXAScaler parallel file system and Infinia object storage platforms.
DDN’s dual-platform AI infrastructure addresses distinct workload needs: Infinia delivers ultra-low-latency metadata performance and high concurrency for inference and retrieval tasks, while EXAScaler provides leading parallel file system speed for AI training and checkpointing. Combined, they unify end-to-end AI data workflows, eliminate silos, sustain high GPU utilization and accelerate time-to-first-token.
The optimized KV Cache framework delivers key advantages: distributed shared caching for large-scale inference, ultra-low-latency access for long-context AI tasks, native support for agentic AI, RAG and multi-step reasoning pipelines, deep compatibility with Dynamo and vLLM, up to 55× faster cache loading, higher GPU efficiency and reduced per-token operational costs.
Security & Efficiency Enhancements
DDN rolled out upgraded enterprise-grade features for production AI environments. Security updates include bare-metal multi-tenancy, KMIP-compliant encryption key management and VictoriaLogs-enabled operational visibility, alongside multi-tenant CSI and standard APIs. Efficiency improvements cover intelligent file pinning and NAND-accelerated Hot Pools, which tier data from expensive all-flash media to cost-effective HDDs.
Real-world deployments validate its performance gains: Salesforce leveraged Google Cloud Managed Lustre (powered by EXAScaler) to achieve 1.5× faster model training, 75% lower I/O latency and 42% reduced training expenses. The A1400X3M will reach general availability by late Q3 2026, with full datasheets publicly accessible.
Beijing Qianxing Jietong Technology Co., Ltd.
Sandy Yang/Global Strategy Director
WhatsApp / WeChat: +86 13426366826
Email: yangyd@qianxingdata.com
Website: www.qianxingdata.com/www.storagesserver.com
Business Focus:
ICT Product Distribution/System Integration & Services/Infrastructure Solutions
With 20+ years of IT distribution experience, we partner with leading global brands to deliver reliable products and professional services.
“Using Technology to Build an Intelligent World”Your Trusted ICT Product Service Provider!
Your message must be between 20-3,000 characters!
English
