DDN - IT Press Tour #62 June 2025

Transcript

1. 1 │ © 2025 DDN The World’s Leading Data Intelligence Platform DELIVERING POWERFUL AI BUSINESS OUTCOMES [ ANY DATA ] [ ANY APPLICATION ] [ ANYWHERE ]
2. 2 │ CONFIDENTIAL DDN at a Glance 700,000+ GPUs deployed +$90bn 2032E AI Storage TAM 11,000+ Customers 1000+ | 600+ Employees | Engineers
3. 3 │ You Cannot Succeed in AI Without DDN Jensen Huang, CEO of Nvidia, and Alex Bouzari, CEO of DDN, at DDN Beyond Artificial, February 20, 2025 “NVIDIA Is Powered By DDN …Without DDN, NVIDIA Supercomputers Wouldn’t Be Possible.” Jensen Huang, NVIDIA CEO
4. 4 │ Our Journey 1998 1998-2008 2008-2018 2018-2023 2024-Present The Beginning Leadership in High Performance Computing Expansion in Enterprise Markets & Technological Innovation Accelerated Growth & Leadership in AI Analytics, LLM Training & Gen AI Go-to Leading Force in AI Data Intelligence
5. 5 │ The world’s largest alternative asset firm with $1.1 trillion+ in assets invests $300M in DDN at a $5 Billion valuation $300 MILLION INVESTED
6. 6 │ Powering AI for Enterprises, Cloud Providers and Integrators Multimodal and Agentic AI Drug Discovery / Autonomous Vehicles Financial Modeling AI Fraud Detection Genome Sequencing LLM Training and Inference On Prem and in Multiclouds
7. 7 │ DDN Supercharges More Than Half a Million GPUs The Latest NVIDIA Technology Optimized for Your Needs 8 of the Top 10 Cloud Providers
8. 8 │ DDN Wins Momentum & Growth • Record enterprise expansion across AI, HPC, and data-intensive verticals • 3rd consecutive year of double-digit revenue growth in AI-driven deployments • Strongest global pipeline in company history, driven by demand for intelligent data platforms Recent Wins • Selected by leading financial services firm for real-time risk analytics at scale • Chosen by global pharmaceutical leader for AI-powered protein modeling • Winner of several notable awards this year, including recognition by AI Breakthrough for Best AI Platform Innovation, Coolest Cloud Company by CRN, AI Excellence Award, highlighted on the AI 100 list as the Best Data Company. Market Recognition • DDN Infinia featured in Gartner’s 2025 Market Guide for Hybrid Infrastructure for AI • Key mention in Gartner's March 2025 “CTO’s Guide to the Generative AI Technology Landscape” • Highlighted for unified, intelligent data services • Recognized for enterprise-grade governance and performance at scale Why It Matters • Innovation without infrastructure is just potential. DDN turns AI and HPC ambition into execution.
9. 9 │ DDN Powers More AI Across More Markets Than Anyone Else Financial Services Healthcare Manufacturing Financial Services NCP & AI Cloud Providers Autonomous Driving AI Chatbots & Assistants Energy Defense Public Sector & Research
10. 10 │ Driving Business Outcomes for Major Industries Financial Services Manufacturing Life Sciences Telco Public Sector Use Cases: Fraud, real-time analytics, transaction monitor, market predictions Use Cases: Genomic storage & analysis, high- throughput sequencing. Real time bioinformatics Outcome: DDN fast data retrieval enables high- frequency trading, compliance, security in this high-regulated area Outcome: DDN scalable fast retrieval and processing speeds up drug discovery and diagnostic accuracy Use Cases: Automate, optimize, predictive maintenance Outcome: DDN enables improved QC, increased efficiency, and reduced downtime Use Cases: Advanced algorithms for data patterns, for telco detection/prediction of network anomalies Use Cases: Intel analysis, storage of large surveillance data for real- time defense AI apps Outcome: DDN enables telcos to fix problems before customers negatively impacted Outcome: DDN low latency enables faster threat detection and predictive analytics helps cybersecurity Financial Services
11. • Lower AI Adoption Barriers: Cost-optimized infrastructure makes AI more accessible for enterprise use. • Faster Data Processing: High-performance architecture ensures smooth handling of large, complex workloads. • Global Scalability: Core42’s infrastructure supports deployments across multiple regions and industries. • Support for Innovation: Flexible, evolving systems empower rapid development of new AI solutions. 11 BENEFITS • Infrastructure Cost and Scalability: The high cost of compute resources and the need to scale quickly across regions present barriers to AI adoption. • Performance and Evolving Demands: AI workloads are data-intensive and continually evolving, requiring infrastructure that can deliver high performance while adapting to new use cases like robotics and climate modeling. "I would say the starting point is performance and scalability. We are deploying AI factories that have to handle, by definition, a lot of data in and a lot of data out. We are running heavy compute on the data. We are in the constant search of storage solutions that can enable us both from a performance and a scalability standpoint. Obviously, we also request these solutions to be price competitive." Edmondo Orlotti, Chief Growth Officer | Core42 © 2025 DDN Accelerating AI Adoption with High- Performance Infrastructure • Scalable Infrastructure: Core42 deploys high-performance AI factories designed to manage large-scale, data-intensive workloads. • Optimized Storage Performance: The infrastructure is built to deliver fast, efficient access to data, enabling smoother AI operations. • Future-Ready Architecture: Core42 continuously adapts its platform to meet the evolving needs of AI applications across industries.
12. Encouraging Quantitative Analysts to Operate Independently With Parallel Processing 12 SOLUTION • DDN 400NVX2 QLC systems • Simple NVME-oF backend BENEFITS • Decreased processing latency by 10x • Parallel protocol enabled concurrent diverse workloads across a high number of nodes • Instant results upon deployment CHALLENGES While decentralizing the management of approaches their quantitative analysts take provides a strategic advantage in the firm’s ability to outpace competition, it makes enforcing efficient IO near impossible. As a result, to make this strategy work, Jump required a robust and highly efficient compute infrastructure capable of handling diverse workloads concurrently and with little to no processing latency. “DDN QLC systems are a really important part of that environment to get IO to our researchers as quickly as possible.” Alex Davies | Chief Technology Officer, Jump Trading © 2025 DDN
13. Genome Sequencing at Scale with NVIDIA H100 & DDN 13 SOLUTION • An end-to-end solution backed by hybrid NVME and dense disk storage for Nanopore Sequencing • DDN A³I with NVIDIA H100 GPUs for a fast system in a small footprint • Simple expansion with standard building blocks Roche BENEFITS • Decreased analysis turnaround time by more than 7X (from 15 days to 2 days) • ​​Allowed them to keep and analyze more data • Reduced the amount of data uploaded to the cloud by 100X CHALLENGES Being 100% cloud based worked fine early in their development, but after adding additional sensors and generating more data – up to 2PB every 24 hours – it became too cumbersome and slow to continue with that strategy. Limitations of being fully cloud-based required users to compromise their analysis and discard 99% of their data. Turnaround time for critical experiments was over two weeks.​ “The solution DDN had, for what we needed, just made sense. We needed a small footprint and something very fast so we could process the data as quickly as possible – time is money.” Chuck Seberino | Director of Accelerated Computing Roche Sequencing Solutions © 2025 DDN
14. 14 │ Summary – How DDN Benefits Customers AI Focus GPU Efficiency Inference & RAG Acceleration AI Factory Alignment Multi Cloud & Portability Deployment Complexity Partner Ecosystem Cost Efficiency
15. 15 │ Welcome to the Data Intelligence Revolution!
16. 16 │ Complexity Performance High Costs Reliability & Security AI Challenges Facing the Enterprise • Agentic to Cloud • Multimodal • Cost prohibitive to provide real time insight to drive ROI • Inference and AI pipeline • Model iterations • Processing at massive scale • Governance • Hybrid Environment • Secure connection Agentic to Core
17. 17 │ AI Requires a New Technology Stack New AI Stack AI-Native Software and Services Software Defined and Power Optimized Software-Defined & High Throughput Highly Parallelized with Rich Data Management Representative Leaders Cloud Compute Data Layer & Storage Networking
18. 18 │ The Data Layer is a Mission Critical Piece of the New AI Stack $9.0bn $33.1bn $92.6bn 2022A 2027E 2032E 30% CAGR AI Storage TAM to Grow 10x+ in a Decade 23% CAGR Requirements for AI Optimized Storage TCO at Scale Massive Scalability High Throughput I/O TCO at Scale Speed to Service GPUs Source: Bloomberg, Generative AI: Accessing Opportunities and Disruptions in an Evolving Trillion-Dollar Market, (September 28, 2023).
19. 20 │ DDN’s Data Intelligence Platform Delivers Game Changing AI Value We boost AI value for the Enterprise across the stack on prem and in multi-clouds AI large language models, training, inference, and GenAI are accelerated up to 100X DDN Data intelligence platform sits between APPLICATIONS and INFRASTRUCTURE and supercharges both GPU efficiency increases 30%, while data center footprint and power consumption are reduced by 10X
20. 22 │ DDN Data Intelligence Platform Overview DDN Infinia provides massive scale and a robust AI data intelligence platform DDN EXAScaler provides a highly efficient and scalable parallel file system for AI at scale Deploy Securely Reduce Complexity Accelerate AI Innovation Reduce Costs Performance at Scale Efficient Data Management Enterprise Class Simplicity Secure & Reliable
21. 23 │ Training Data Load Checkpoint Model Load NVIDIA GPU Utilization NVIDIA GPU + DDN Utilization Higher Utilization with Lower Load Times DDN Optimizes GPU Usage and Improves Customer ROI by 30% DDN helps to free up valuable GPU resources by • Optimizing data movement • Accelerating checkpoint and model load operations Training Data Load Checkpoint Model Load
22. 24 │ Faster ingestion and low latency indexing times enables over 20X faster NVIDIA NIMs and RAG pipelines Over 20X Faster RAG Pipelines in One Click 0 500 1000 1500 2000 2500 3000 3500 4000 4500 Infinia + GPU with NVIDIA cuVS AWS + GPU with NVIDIA cuVS Indexing Rate (1000s per second) RAG Pipeline Indexing Acceleration 22X SPEEDUP Query Service NVIDIA cuVS Data Service Indexing Service NVIDIA cuVS User PDF Embedder Embedder GPU Acceleration DDN Acceleration
23. 25 │ DDN EXAScaler PRODUCTS MODULE
24. 26 │ Preferred by NVIDIA for AI/ML with NVIDIA SuperPOD, Highest Performing Filesystem with multi-protocol support (NFS, SMB, and S3) and Multi Tenancy DDN EXAScaler is the fastest HPC and AI Platform at Scale Any Scale Proven at EXAScale with 10s of 1000s of GPUs and CPUs Best Performance Fully rounded Multi-TB/second throughput, millions of Metadata operations Robust EXAScaler supports more GPUs than any other data platform
25. 28 │ AI Training is Data Intensive Ingest/Data Prep Data Load Model Load Checkpoint Model Distribution Distributed Training Training Load tokens and images for each training Epoch Load model into GPU prior to start and restart Save model to persistent for many reasons Up to 60% Up to 5% Up to 43% % of time that can be consumed by data movement
26. 30 │ AVAILABLE NOW in DDN EXASCALER Secure and Simple Multi-Tenancy: NVIDIA Approved NCP tenant management Up to 10X Data Reduction: Scalable data reduction lowers costs Zero Downtime: Online upgrades keep data always available 10x Data Capacity: Higher density FLASH and appliance support DDN Accelerates AI with Spectrum-X and NVIDIA Bluefield NEW! NEW Reference Architectures with NVIDIA Blackwell GPU Solutions NEW!
27. 31 │ • Up to 3.2x Read Acceleration • Up to 2.4x Write Acceleration • Up to +16% overall throughput with concurrent workloads • No performance degradation for all other workloads DDN & NVIDIA Spectrum-X Speed Up 0 10 20 30 40 50 60 70 80 90 Read GB/s write GB/s GB/s 4 clients ECMP collision 4 clients Spectrum-X DDN AI400X2 Performance (fio) RoCEv2 vs Spectrum-X Collision scenario x3.2 read x2.4 write NEW!
28. 32 │ • DDN Solutions Deliver the Maximum Performance over both Spectrum-X and Infiniband into NVIDIA DGX B200 Platforms • Over 90% of Max Network with BOTH Reads and Writes to accelerate Model and Data Loads as well as Checkpoints NEW Reference Architectures with NVIDIA Blackwell GPU HIGHEST PERFORMANCE ON MINIMUM INFRASTRUCTURE NEW!
29. 33 │ Optimized Multi-Tenancy MINIMUM DATACENTER FOOTPRINT FOR MAXIMUM PERFORMANCE Authentication Access control Multitenancy Encryption Establish user and node identity with full confidence. Enforce policy and multiple levels of classification. Share infrastructure to enable limitless at- scale flexibility. Secure all your data end-to-end, live and at rest. Auditing Record and retain activity for review and compliance. Isolated Data Partitions Dynamic Tenant Networks Tenant Quota Tenant Quality of Service
30. 34 │ • EMF Tenants – single command Tenant Creation • Default secure - no access to filesystem to nodes not part of a Tenancy Create Update Add/remove clients from Tenant, increase/decreate Capacity allocation Delete Remove configuration and optionally delete data Disable/Enable Disable client access to data, configuration remains present DDN EXAScaler Multitenancy Orchestration emf tenant [command] emf tenant list emf tenant create TENANT [--nids NID_RANGE1,NID_RANGE2,...] --capacity-allocation 100TiB [--inode-allocation 100M] emf tenant destroy TENANT [--remove-data] Namespace Isolation Capacity Management Identity Domain Isolation [global] multitenancy = true EMF-level configuration option to enable Multitenancy capabilities globally for the filesystem EXASCALER APIS AND ORCHESTRATION EXAMPLE
31. 35 │ • Online Server updates from EXA 6.3.1 • Rolling updates allowing running workloads to complete, scheduling update prior to next workload start • New health and configuration checks in the upgrade process. • ‘EMF analyze’ comprehensive system- diagnostics and pre-upgrade-checker. 10X Faster Deployments, Online Upgrades DDN EXASCALER Orchestration Manager Monitor cluster state run/track command plans Service Mesh Secure comms Servers, clients & utility nodes Security infrastructure Role Based Access Control Audit logging
32. 36 │ • Monitor Live & Historical Workloads. DDN is the only platform to show running jobs • Easy Integration into Scheduling Environment – just requires environment variables to be set • Use Open Tools, Prometheus, Grafana • Physical Views of HW, Monitor All Protocols • Easy installation & upgrade - Major OS flavours and Docker-based installation has been introduced to simplify deployment See More, Diagnose Faster with DDN DDN EXASCALER MONITORING AT SCALE Analyze File Distribution Monitor NFS/SMB/S3
33. 37 │ Proven Large Scale DDN and NVIDIA Solutions: 4608 GPU MINIMUM DATACENTER FOOTPRINT FOR MAXIMUM PERFORMANCE 63 NVIDIA GB200 NVL72 4608 B200 GPUS 2 RACKS OF DDN EXASCALER 31 DDN DATA APPLIANCES: AI400X3 6 DDN METADATA APPLIANCES: AI400X3
34. 40 │ DDN Infinia PRODUCTS MODULE
35. 41 │ DDN Infinia is a Software-Defined, Data Intelligence Platform Designed For AI Any Data Unstructured, Semi- Structured, Structured, Any Protocol Any Location Built for Clouds to Deliver Data Capability as a Service Any Scale In Production at EXAScale One Platform. SLA Driven. On Premise and MultiCloud Multimodal and Agentic AI. LLM and Inference
36. 43 │ S3 / GCS CSI / CINDER - The Data Platform to Accelerate AI Workflows KV Store – Unlimited Metadata
37. 44 │ ... Node Infinia Data Plane (KV Store) Pool A Erasure Coding Control Plane Enterprise Capabilities • Scale Out, Elastic Meta/Bulk • Always On • Data Reduction • Secure Encryption • Snapshot • Multi-tenancy Data Center, GPU and Cloud Efficiencies and Cost Reduction • Fault Domain Aware Maximize Uptime • Network Erasure Coding • Fully Hardware Agnostic • 100% Cloud Native Application & Data Acceleration • Low Latency Multi-Protocol • Streamlined Analytics • Advanced Data Services & SDK S3/GCP object CSI/Cinder FILE SQL . . . DDN Supercharges LLM Training, Inference & Gen AI Anywhere Node ... Node Pool N Erasure Coding Node Software Development Kit (SDK) Data Ocean Event Engine Pipeline & Workflow Acceleration • Event Engine Drives Pipeline • Intelligent Data Movement
38. 45 │ Unlocking Real-Time Insights with 100X Faster Object Listing Performance • Query and Find Data faster within workflows: DDN Infinia Listing performance delivers more than 100X the performance compared to public cloud 0 100,000 200,000 300,000 400,000 500,000 600,000 700,000 1 worker 10 workers 100 workers AWS Listings per second Listing Performance (1 Bucket) AWS DDN is 100x faster
39. 46 │ AWS S3 Express
40. 47 │ • DDN Infinia running on Public Cloud outpaces Native Object Storage for the most common operations Small Object operations 0 5000 10000 15000 20000 25000 Rate (per second) Create Rates 0 2000 4000 6000 8000 10000 12000 14000 16000 Infinia (12x EC2) AWS S3 Express One Aws S3 Rate (per second) Delete Rates 19X Faster 8X Faster
41. 48 │ • Rapid data retrieval, enables real-time interactions critical for RAG apps • Enhanced User Experience: empowers seamless, high- performance workflows, improving end-user satisfaction and business outcomes. 25X Faster Response Times for Data Access 0 20 40 60 80 100 120 140 Infinia Competition Time to First Byte (Lower is Better) 25x lower latency
42. 49 │ DDN Stays Fast, Even Under High Load 1ms 800 𝜇𝑠 600 𝜇𝑠 400 𝜇𝑠 200 𝜇𝑠 0 𝜇𝑠 0:00 1:00 2:00 3:00 4:00 5:00 6:00 Average Latency • Faster AI load times, database response times, queries etc improving end user satisfaction • Accelerate Intensive Enterprise Analytics workloads like Apache Spark, Starburst Presto/Trino, Clickhouse Consistent low latency High Concurrent Load Requests Per Second
43. 50 │ And Infinia Scales to the largest AI Systems in the World 10X LOWER LATENCIES, OVER 1TB/S OF REAL BW FROM ONE CLUSTER • Unmatched Performance Under Heavy Loads • Handle millions of concurrent queries with low-latency responses, ensuring real-time performance for AI and RAG workflows. • consistent, high-speed data access, improving time-to-insight and decision-making >TB’s/sec Data Throughput 8mb transfers, 60 connections/node 30 Million Object List per Second
44. 51 │ Simplify and Accelerate Inference & RAG Pipelines Lowest Latency Event Engine Network Acceleration Seamless Multi-Protocol Native Real-Time Search GPU Acceleration Data Ingest Booster
45. 52 │ Faster Ingestion and Low latency indexing times enables over 20X Faster NVIDIA NIMs and RAG Pipelines Over 20X Faster RAG Pipelines : EXAMPLE ONE-CLICK RAG ACCELERATED ONE-CLICK RAG PIPELINE: DDN INFINIA AND NVIDIA NIMS 0 500 1000 1500 2000 2500 3000 3500 4000 4500 Infinia + GPU with NVIDIA cuVS AWS + GPU with NVIDIA cuVS Indexing Rate (1000s per second) RAG Pipeline Indexing Acceleration 22X SPEEDUP Query Service NVIDIA cuVS Data Service Indexing Service NVIDIA cuVS User PDF Embedder Embedder GPU Acceleration DDN Acceleration
46. 57 │ Building a Cloud Multi-Tenancy Platform DDN INFINIA 2.1 Realm Tenant Sub Tenant Sub Tenant Sub Tenant Sub Tenant Tenant Sub Tenant Sub Tenant Tenant Sub Tenant Realm Admin Tenant Admins Sub Tenant Admins RBAC and scoped user roles (realm, tenant, subtenant level Realm Admin: Manage HW, upgrade, tenants Tenant Admin: Manage users/groups/subtenants Subtenant Admin: Manage DataSets • Integration with external identity providers AD/LDAP • Associate several external identity groups (from LDAP or Active Directory) with a single tenant in Infinia. DataSets
47. 58 │ 100% Service Level Platform: Capacity, Performance, Resilience DYNAMICALLY ADJUST TENANT SERVICE LEVELS Open up Tenant Dialog box & Adjust Allocate Performance Allocate Capacity DONE!
48. 59 │ • Tenants can create datasets and specify the fault tolerance that is required Deliver Resilience as a Dial-In SLA DDN INFINIA DATASET FAULT TOLERANCE TENANT A: DATASET 1 → 3 Drive Failure TENANT A: DATASET 2 → 1 Rack Failure TENANT B: DATASET 1 → 1 Data Hall Failure TENANT B: DATASET 2 → 2 Node Failure TENANT B: DATASET 3 → 1 Node Failure TENANT : DATASET → FAULT TOLERANCE LEVEL Fault Tolerance test: Entire Rack Outage → workload continues with minimal i/o pause
49. 61 │ • S3 greatest flaw = high latency and low bandwidth for Data Movements – S3 Limits use cases that need fast response times – S3 requires data protection and associated latency at the server – S3 does not leverage any network acceleration Traditional Hadoop, SPARK, S3 Limits Application Performance Applications PUT() RESTFul API Calls No Parallelism No RDMA Slow Transfers Server-Side Erasure Coding Additional Latency
50. 62 │ HDFS To DDN; No Code Changes, Lower Latency, Greater Scale Hadoop/SPARK Applications INFINIA SDK CONNECTOR … High-Speed AI Data Processing In-memory computing plus Infinia acceleration boosts processing speeds many times faster than S3 or HDFS Streamlined Integration Fast time-to-market and improved operational efficiency​ using Infinia Connector for: • MLlib machine learning, • GraphX for graph processing • Streaming of real-time data streams Enhanced Scalability Efficient data processing across 1000s of nodes for processing massive AI datasets Cost Efficiency Optimize hardware usage and reduce operational costs by moving more data with less HW Improved Reliability DDN Fault Domains ensure that applications continue uninterrupted through HW failures DDN NDA ONLY Backend acceleration Data path is accelerated with RDMA and fully parallel. Need for back-end erasure coding network RDMA ERASURE CODING DDN INFINIA ROADMAP – DDN SDK NDA ONLY 2025 2026
51. 63 │ • PostgreSQL Front-End – Comprehensive SQL integration • Infinia SQL – Distributed SQL data service built on Infinia KV • Infinia SQL metadata catalog – System metadata explorer with SQL interface – No snapshot/copy overhead, consistent catalog Fast, Scalable SQL for AI and Data Analytics Workloads User Admin SQL Data Service Metadata Explorer RedSQL CLI INFINIA SQL LIBRARY ... Node Node Node Node Node Parallel Data Query Control Commands PostgresSQL
52. 64 │ • Infinia accesses distributed metadata and exposes it as a native Data Service • See and Action Data at remote locations, only moving the data that you really need to process locally Action Data Anywhere, Maximize Efficiency, Minimize Movement DATA MANAGEMENT AND DATA INGEST SECURE DATA SHARING DATA LAKE AI SUPERCOMPUTING DATA MOVEMENT ANALYTICS & INFERENCE CACHING FILTER/COMPRESS Reduce Egress Costs by 10X Free Up Infrastructure Simplify Distributed Data
53. 66 │ Infinia S3: Multi-Tenant Object Storage with Advanced Data Protection and Governance and Strong Flexibility REDUCE COSTS BY 50%, ACCELERATE AI PIPELINES, AND INCREASE CUSTOMER DENSITY • Sub-Millisecond Latency. Real-time AI inference and faster training • 80%+ S3 Compatibility. Easy cloud migration and hybrid cloud • Unlimited Metadata Tags. Rich search and classification • Multi-Tenant with Quotas. Secure, isolated performance for tenants • No Object Size Limits. Handle massive AI datasets and multi-part uploads • Bucket Policies & ACLs. Increased data security and compliance • IPv6 Support & Logging. Future-proof connectivity and full audit trail
54. 67 │ © 2025 DDN Q&A
55. 81 │ EXA on GCP (GCP Managed Lustre) PRODUCTS MODULE
56. 82 │ • Latest 300 TiB Lustre, 280 OSTs, 5 MDTs • 500 rocky clients, n2-standard-32, 128 GB mem, 32 vCPUs, Master node: perf500- 300t-auto-client-rocky-1afa30e6aa • Single Thread: over 3GB/s • Single Client over 10GB/s GCP Managed Lustre Performance – Single Client NOT TO BE SHARED OUTSIDE DDN 0 2000 4000 6000 8000 10000 12000 Write Read Sequentual Perofmrnace (MB/s) Single Client Performance (n2-standard-32) 1 thread 1 client DDN NDA ONLY
57. 83 │ • Latest 300 TiB Lustre, 280 OSTs, 5 MDTs • 500 rocky clients, n2-standard-32, 128 GB mem, 32 vCPUs, Master node: perf500- 300t-auto-client-rocky-1afa30e6aa • Over 350GB/s for Reads • Over 250GB/s for Writes GCP Managed Lustre Performance – Whole Filesystem NOT TO BE SHARED OUTSIDE DDN 0 50000 100000 150000 200000 250000 300000 350000 400000 0 50 100 150 200 250 300 350 400 450 Sequesntial Throughput (MB/s) Client Clound (16 procs per client) GCP Managed Lustre Performance Write Read DDN NDA ONLY
58. 84 │ New Product Announcements PRODUCTS MODULE
59. 85 │ One-Click Solutions – The DDN AI Integrated Platform Financial Services Life Sciences Autonomous Driving Faster Insights, Lower Data Center And Cloud Costs, Accelerated AI Deployment Integrated With NVIDIA Nims, Nemo, DGX, Spectrumx, Bluefield 3, NCP’s And Hyperscalers New Release!
60. 86 │ EXAScaler + Infinia Integrated Solution – LLM + Inference Multimodal and agentic AI, LLM and inference, on prem and multicloud in one platform ✓ Maximum Efficiency LLM AI Training ✓ Edge-Cloud-Core, Inference and Data Governance 100x Speed 1/10 Power Any Scale New Release!
61. 87 │ AI Training & Inference: Scaling to Trillion-Parameter Models Optimized multimodal AI pipeline Sub-millisecond inference latency Maximizes GPU utilization to 99% Maximizes Value In All Generative AI And Real-time Inference Use Cases New Release!
62. 102 │ INDUSTRIES MODULE
63. 103 │ DDN Driving Business Outcomes Across Industries Financial Services High frequency trading Lifesciences Personalized healthcare Manufacturing Defect detection Cloud Providers Large AI Cloud environments Public Sector Early tsunami warnings
64. 104 │ Life Sciences INDUSTRIES MODULE
65. 105 │ • DDN Accelerates the data-Ingest, distribution and analysis of the petabytes of data • DDN is recognized by Genomics research facilities – such as Life Technologies®, Sanger®, TGen® and Cornel®, as the go-to visionary- partner and trusted-advisor for best- in-class storage solutions. • Our subject matter experts help researchers and scientists accelerate their discovery process Proven at Leading Institutions Worldwide
66. 106 │ • AlphaFold2 predicts 3D structure of target protein. • Molecules passed to MolMIM generates diverse molecules to explore for binders • These are evaluated by an Oracle model, scored on binding affinity, etc • DiffDock predicts optimal binding poses and enhancing the binding configurations. Streamlines identification and optimization of drug-like molecules, significantly reducing the time/cost of traditional drug discovery. Life Sciences: Virtual Screening Reduces time for Drug Discovery AlphaFold2 Target Protein Sequence Protein Database MSaaS Multiple Sequence Alignment Target Protein Structure DiffDock Docking Molecule Structure Library MolMiM Ranking Molecules User Human Feedback Optimized Hits Protein Folding Molecule Generation Gradient-Free Optimization QED, Solubility, Human Feedback
67. 107 │ • Data Orchestration: Infinia intelligently curates and moves large datasets between AlphaFold2, MolMIM, Oracle, and DiffDock, ensuring seamless workflows and eliminating data silos • Scalable : Provides high-performance object and database functionality to manage the molecular libraries and AI-generated structures • Secured Data: DDN Infinia Secures Pipeline Data, Distributed Data across Clinical and Medical Sites and Identifying Data Labels Life Sciences: Virtual Screening Reduces time for Drug Discovery AlphaFold2 Target Protein Sequence Protein Database MSaaS Multiple Sequence Alignment Target Protein Structure DiffDock Docking Molecule Structure Library MolMiM Ranking Molecules User Human Feedback Optimized Hits Protein Folding Molecule Generation Gradient-Free Optimization QED, Solubility, Human Feedback Low latency object Persistent container volume Create triggers Fast Search
68. 108 │ Financial Services INDUSTRIES MODULE
69. 109 │ • Fraud Detection & AML • Ultra-low-latency AI pipelines • Risk Management & Compliance • Alternative Data & Sentiment Analysis • Personalization & Customer Intelligence Why Choose DDN for AI in Financial Services & Retail? ROI from AI requires a scalable, data-first strategy for real-time insights, efficiency, and competitive advantage AI-Optimized Performance Ultra-low latency AI pipelines for real-time insights and decision-making. Better Business Outcomes Boost NPS, and drive repeat purchases in retail. Near- perfect fraud prevention in FSI Cost Efficiency at Scale Reduce AI training, inference, and data retrieval costs while eliminating cloud penalties Enterprise-Grade Scalability & Security Secure, high-performance infrastructure for petabyte- scale AI workloads. DDN DELIVERS
70. 110 │ Speed Up GPU Applications More Than 10X with Optimized IO DDN delivers fastest and most responsive small IO, random IO and metadata performance 0 100 200 300 400 500 600 700 800 900 1000 VAST WEKA DDN AI400X2 0 500 1000 1500 2000 2500 3000 3500 NFS Storage DDN AI400X2 16X MORE IOPS IOPS to GPUs (more is better) IO500 MD Score (higher is faster) Random 4K IO Performance to Single GPU Client Filesystem Metadata Responsiveness 13X FASTER More IOPS = More Processing, No Stalling Faster Metadata = Immediate Access, No Idling
71. 111 │ AI Workflow - Hyper Personalization for a Large Retailer Data Prep Customer 360 Merchandizin g / Catalogue Pricing Ordering Shipping Support Fast Iterative Model Experiments - Multi modal models - Increasing parameters and size - Changing datasets driving innovation Inference Distributed Deployment of Recommendation Engine, Predictive Analytics, RAG, Chatbots With QoS for page refresh, throughput, and latency for search, find, and buy Supply Chain & Inventory Mobile Browser Conversational Commerce DDN Data Intelligence Model Training Model Training Model Training Validation Data Types: Demographics & Profile Data Behavioral Data Loyalty & Engagement Data Real-time & Contextual Data. Psychographic & Sentiment Data Supply Chain & Inventory Data. Dynamic Pricing & Competitive Data Model Training Model Training Inference Personalization Engine
72. 112 │ DDN Architecture for Multicloud Real Time Fraud Detection STREAMING RISK SCORING Transactions (payment terminals, apps, online) are ingested via Spark Streaming. adds location, device fingerprints, transaction history XGBoost (Gradient Boosting) is the primary fraud detection model. Data is stored in MapR-DB/Apache HBase and flagged high-risk transactions are stored separately. Object Store REPLICAS On Premises On Premises
73. 113 │ • DDN reduces latency at every data intensive step of the AI workflow • Cumulative Optimizations enable up to 10x more results from your infrastructure AI Workflow – Fraud Detection for a Payment Processor Transaction Logs Banks/Payment- providers User Behavior Data External Threat Feeds (dark web, logs) Model Validation Inference & Feedback Model Deployment Query Layer Dremio/Trino Model Training 0 5 10 15 20 25 Transaction User Behavior External Threats Model Training Inference Transactions per millisecond AWS S3 Infinia/EXA
74. 114 │ • Faster Data Pipelines – Maximize GPU efficiency near 100% • Improved Model Accuracy – Unified data access across all environments • Hybrid Cloud – Seamless on-prem, edge, and public cloud integration • Multi-Protocol Support – S3, GCS, POSIX, and SQL compatibility • Disruption-Free Scaling – Multi-exabyte scale-out with zero downtime • Multi-Tenancy – Supports large data teams and partners DDN’s value drivers for Financial Services
75. 115 │ NCP & AI Cloud Providers INDUSTRIES MODULE
76. 116 │ DDN Powers AI & NVIDIA Cloud Providers
77. 117 │ Powering GPU-as-a-Service with DDN & NVIDIA
78. 118 │ Sovereign AI INDUSTRIES MODULE
79. 119 │ Sovereign AI : Accelerate product development cycles 100x. Optimize for cost savings, seamless integration of Integrated AI + HPC workflows DDN DATA INTELLIGENCE PLATFORM Pre-Processing & Data Prep. Post Processing & Inference/Production Simulation & Training Data Collection AI Surrogate Models accelerate National Productivity across ALL verticals: Automotive Manufacturing | Oil and Gas | Life Sciences | Defense | Aerospace,
80. 120 │ Surrogate Models : 100X Accelerated Time to Insight Application Hosting User ISV Application Omniverse Streaming API On Premise Cloud Design Engineer CFD Engineer CAD Application CFD Application 3D Assets USD Database Geometry Surface/Volume AI Engineer AI Training Application Surrogate Model NIM World State Controller Modulus Training Aerodynamics Foundational Model Execution Engine Voxelizer Extensions CFD Simulation Model Data AI Model Training Surrogate Aerodynamics Model AI Engineer AI Training Application Container Kit Application Scene Data
81. 121 │ Surrogate Models : 100X Accelerated Time to Insight Application Hosting User ISV Application Omniverse Streaming API On Premise Cloud Design Engineer CFD Engineer CAD Application CFD Application 3D Assets USD Database Geometry Surface/Volume AI Engineer AI Training Application Surrogate Model NIM World State Controller Modulus Training Aerodynamics Foundational Model Execution Engine Voxelizer Extensions CFD Simulation Model Data AI Model Training Surrogate Aerodynamics Model AI Engineer AI Training Application Container Kit Application Scene Data
82. 122 │ Agentic AI INDUSTRIES MODULE
83. 123 │ Agentic AI systems ingest vast amounts of data from multiple data sources and third- party applications to independently analyze challenges, develop strategies and execute tasks. Businesses are implementing agentic AI to • personalize customer service • streamline software development • facilitate patient interactions DDN Delivers a Fully Featured Data Platform for Agentic AI
84. 124 │ • DDN's Infinia and EXAScaler are foundational technologies driving the evolution of Agentic AI – Transform data management, retrieval, and AI model performance, empowering enterprises to deploy scalable, efficient AI workflows • DDN Infinia integrates with NVIDIA AI Enterprise to manage and orchestrate large- scale pipelines, supporting NIM microservices and NeMo Retriever. – Selectively curates and moves high-value data from cities, vehicles, and other sources to strengthen autonomous systems – Enhance retrieval-augmented generation (RAG) and provides critical support for Kubernetes container volumes, AI event management, and model governance. Agentic AI: Building AI Agents with NVIDIA and DDN
85. 125 │ Building AI Agents with NVIDIA and DDN Curator Customizer Evaluator Guardrails RAG Retriever NVIDIA NeMo NVIDIA NIM Information Retrieval AI Safety Digital Human Visual Content Generation Digital Biology Physical AI Understanding & Reasoning DDN INFINIA DDN DATA INTELLIGENCE Object Store Container Plane Catalog Event Engine Data Ocean
86. 126 │ NeMo Retriever Multimodal Data Extraction Microservices State-of-the-art data extraction for petabytes of PDFs created annually Multimodal Data Extraction Throughput Extraction of Enterprise Documents Pages per second, evaluated on publicly available dataset of PDFs consisting of text, charts, and tables. NIM On: nv-yolox-structured-image-v1, nemoretriever-page-elements-v1, nemoretriever- graphic-elements-v1, nemoretriever-table-structure-v1, PaddleOCR, nv-llama3.2-embedqa- 1b-v2. NIM Off: open-source alternative; HW - 1xH100 Customizable & Scalable Document Ingestion—any format, with any modality, of any size Built on NVIDIA NIM Supports docx, pptx, png, jpg, infographics Future: html, xlsx Extract text, structured charts, tables Future: flow charts, block diagrams, infographics Customizable extraction operations GPU accelerated linear scaling 15x improved throughput 12 pages/sec 0.81 pages/sec NIM Off NIM On Higher Throughput Multimodal Retrieval Accuracy NeMo Retriever Extraction Recall@5 Accuracy Retrieval of Enterprise Documents Evaluated on publicly available dataset of PDFs consisting of text, charts, tables, and infographics. NIM On: nemoretriever-page-elements-v2, nemoretriever-table-structure-v1, nemoretriever-graphic-elements-v1, paddle-ocr NIM Off: open-source alternative: HW - 1xH100 50% fewer incorrect answers 91% NIM Off NIM On 81%
87. 127 │ DDN Enablement in AI Workflows: RAG Chat Streaming Audio PDF Image Reverse Proxy Server S3 Content → Object Store Retrieved Content LLM NIM Knowledge Graph Relationships → KV Store Graph Search NeMo Retriever Embedding Vector DB Vectors → Vector DB Vector Search User RAG enhanced Chat Query RAG Chat Interface NeMo Retriever Embedding LLM NIM NeMo Guardrails NeMo Retriever Reranking
88. 128 │ DDN Enablement in AI Workflows: RAG Chat Streaming Audio PDF Image Reverse Proxy Server S3 Content → Object Store Retrieved Content Knowledge Graph Relationships → KV Store Graph Search Vector DB Vectors → Vector DB Vector Search User RAG enhanced Chat Query RAG Chat Interface Low latency object Persistent container volume Storing embeddings Create triggers Fast Search LLM NIM NeMo Retriever Embedding NeMo Retriever Reranking LLM NIM NeMo Retriever Embedding NeMo Guardrails
89. 129 │ One-Click RAG Pipeline: DDN Infinia and NVIDIA NIMs Query Service NVIDIA cuVS 1x L40S 48GB Data Service 128x CPUs Indexing Service NVIDIA cuVS 1x L40S 48GB User/Agent API Service NVIDIA NIM Reranking Service [llama-3.2-nv-rerankqa-1b-v2] 4x L40S 48GB NVIDIA NIM Embedding Service [nv-embedqa-e5-v5] 8x L40S 48GB Application Service NVIDIA NIM LLM Service [Llama-3.3 70B Instruct] 4x L40S 48GB NVIDIA NIM Embedding Service [nv-embedqa-e5-v5] 8x L40S 48GB Knowledge Base (PDFs) Document Parser Service API Service Business Insights Infinia 720TB 12x (8x 7.5TB SSDs) Shared service Shared service
90. 35x Improved Data Storage Efficiency with NeMo Retriever Deliver actionable intelligence at greater scale with optimized embedding microservices The NVIDIA NeMo Retriever embedding model (llama-3.2-nv-embedqa-1b-v2) impacts vector storage volume with long context support, dynamic embeddings, and efficient storage for high-performance, scalable data processing - improving storage efficiency by 35x. 1,024 Raw Text 2,560 Long Context Baseline Vector Embeddings Reduced Embedding Dimensions 70 188 GPU/CPU Memory (300 tokens, 1024 dimensions, FP8) nv-embedqa-e5-v5 (4096 tokens, 1024 dimensions, FP8) llama-3.2-nv-embedqa-1b-v2 (4096 tokens, 384 dimensions, FP8) llama-3.2-nv-embedqa-1b-v2 35X Improved Data Storage Efficiency
91. 5.3X lower indexing cost CPU GPU Higher Index Build Throughput CPU indexing HW - 5th gen Intel Xeon (192vCPU); GPU indexing HW - 8xL4 Embedding - nv-embedqa-e5-v5; segment size - 240K vectors (1024 Dim, fp32); Indexing - CAGRA (GPU), HNSW (CPU); Target Recall - 98% NV-EmbedQA-E5-V5​ English text embedding + indexing for question-answering retrieval​ CPU GPU GPU-Accelerated Vector Databases NVIDIA enables massive scale with accelerated extraction, retrieval, and vector search 7x improved throughput 54k vectors/sec 7k vectors/sec CPU GPU Integrations Available in major vector databases and libraries, including FAISS, Milvus, Solr, Elastic Scalable Enables high-volume vector indexing and search Interoperable interoperable between CPU and GPU enabling index building on a GPU and searching on a CPU Flexible and Scalable Advanced Algorithms GPU Indexing optimized for high accuracy and low latency at all batch sizes Flexible Integration Supports multiple languages including C, C++, Python, and Rust, for easy integration into vectorized data applications
92. 132 │ NVIDIA AI Blueprints INDUSTRIES MODULE
93. 133 │ • Use generative AI to enhance customer service • Infinia brings together fragmented data sources and protects against potential data risks • Leverages RAG and generative AI technologies including NVIDIA NIM and NVIDIA NeMo Retriever NVIDIA AI Blueprint: AI Virtual Assistant
94. 134 │ SQL Store Persistent container volume Store embeddings Create triggers KV Cache SQL Search • Use generative AI to enhance customer service • Infinia brings together fragmented data sources and protects against potential data risks • Leverages RAG and generative AI technologies including NVIDIA NIM and NVIDIA NeMo Retriever NVIDIA AI Blueprint: AI Virtual Assistant
95. 135 │ NVIDIA AI Blueprint: Build A Generative Virtual Screening Pipeline • Virtual Screening Reduces time for Drug Discovery • AlphaFold2 predicts 3D structure of target protein. • Molecules passed to MolMIM generates diverse molecules to explore for binders • These are evaluated by an Oracle model, scored on binding affinity, etc • DiffDock predicts optimal binding poses and enhancing the binding configurations. • Streamlines identification and optimization of drug- like molecules, significantly reducing the time/cost of traditional drug discovery. AlphaFold2 Target Protein Sequence Protein Database MSaaS Multiple Sequence Alignment Target Protein Structure DiffDock Docking Molecule Structure Library MolMiM Ranking Molecules User Human Feedback Optimized Hits Protein Folding Molecule Generation Gradient-Free Optimization QED, Solubility, Human Feedback
96. 136 │ • Data Orchestration: Infinia intelligently curates and moves large datasets between AlphaFold2, MolMIM, Oracle, and DiffDock, ensuring seamless workflows and eliminating data silos • Scalable : Provides high-performance object and database functionality to manage the molecular libraries and AI-generated structures • Secured Data: DDN Infinia Secures Pipeline Data, Distributed Data across Clinical and Medical Sites and Identifying Data Labels AlphaFold2 Target Protein Sequence Protein Database MSaaS Multiple Sequence Alignment Target Protein Structure DiffDock Docking Molecule Structure Library MolMiM Ranking Molecules User Human Feedback Optimized Hits Protein Folding Molecule Generation Gradient-Free Optimization QED, Solubility, Human Feedback Low latency object Persistent container volume Create triggers Fast Search NVIDIA AI Blueprint: Build A Generative Virtual Screening Pipeline
97. 137 │ DDN Enablement in AI Workflows: AI Radio Frameworks Fast data layer Low latency object Fast Search Embeddings pyAerial interacts with NVIDIA cuBB layer-1 data functions to generate over-the-air training data for the Aerial Data Lake. DDN’s fast data intelligence layer helps with Data prep, inference and integrates with NIM and NeMO ecosystem
98. 138 │ External DDN Customer Case Studies CASE STUDIES MODULE
99. • Lower AI Adoption Barriers: Cost-optimized infrastructure makes AI more accessible for enterprise use. • Faster Data Processing: High-performance architecture ensures smooth handling of large, complex workloads. • Global Scalability: Core42’s infrastructure supports deployments across multiple regions and industries. • Support for Innovation: Flexible, evolving systems empower rapid development of new AI solutions. 139 BENEFITS • Infrastructure Cost and Scalability: The high cost of compute resources and the need to scale quickly across regions present barriers to AI adoption. • Performance and Evolving Demands: AI workloads are data-intensive and continually evolving, requiring infrastructure that can deliver high performance while adapting to new use cases like robotics and climate modeling. "I would say the starting point is performance and scalability. We are deploying AI factories that have to handle, by definition, a lot of data in and a lot of data out. We are running heavy compute on the data. We are in the constant search of storage solutions that can enable us both from a performance and a scalability standpoint. Obviously, we also request these solutions to be price competitive." Edmondo Orlotti, Chief Growth Officer | Core42 © 2025 DDN Accelerating AI Adoption with High- Performance Infrastructure • Scalable Infrastructure: Core42 deploys high-performance AI factories designed to manage large-scale, data-intensive workloads. • Optimized Storage Performance: The infrastructure is built to deliver fast, efficient access to data, enabling smoother AI operations. • Future-Ready Architecture: Core42 continuously adapts its platform to meet the evolving needs of AI applications across industries.
100. Transforming AI Infrastructure for Maximum Performance and Efficiency 140 • Scalability & Infrastructure: Scalable AI cloud with integrated high- performance storage. • Developer Experience: Simplified AI development with fast model training and deployment. • Cost Efficiency: Efficient, cost-effective platform supporting a range of user needs. BENEFITS • Speed & Performance: Faster AI model training and deployment. • Affordability: Lower infrastructure costs for broader AI accessibility. • Operational Efficiency: Streamlined workflows that reduce complexity for developers. • Reliability & Flexibility: Scalable and reliable platform supporting diverse industries.  Technical Complexity: Growing complexity of AI models and massive data volumes.  Cost Management: High infrastructure and computing costs for startups and SMEs.  Performance & Scalability: Performance bottlenecks slowing AI training and deployment. "DDN Data Intelligence platform provides scalability and high-performance storage, keep up with the massive AI workloads. The combination of low-latency access and high bandwidth and the parallel file system compatibilities enable our customers to have faster training of their models and reduce bottlenecks and optimize their AI workflow." Louis Xu, Head of Bitdeer AI © 2025 DDN
101. DDN EXAScaler Supercharges CINECA’s Early Warning Tsunami Forecasting with the Speed to Save More Lives 141 SOLUTION • DDN EXAScaler software • DDN ES400NVX2 All Flash Tier systems • DDN ES7990X + 62 SS9012 + 4 ES400NVX HDD Tier systems BENEFITS • Decreased forecasting turnaround time by 400% • Increased GPU Utilization to 99% • CINECA’s Leonardo compute cluster became #1 ranked supercomputer for bandwidth on IO500 list. CHALLENGES To highlight an example, tsunami forecasting is an extremely time sensitive task that involves modeling impact at coastal locations based on real-time information once a tsunami is detected. Previously, CINECA and most other computing centers had to wait 20 minutes from the time of detection until their forecast was complete. Now, with DDN EXAScaler, CINECA can complete a forecast within 5 minutes, providing a potentially life-saving increase in time needed for evacuations. “This kind of power needs an extremely well-optimized storage environment to gain maximum efficiency. We chose DDN because of its ability to accelerate all stages of the AI and HPC workloads lifecycle.” Mirko Cestari | HPC & Cloud Technology Coordinator, CINECA © 2025 DDN
102. Siam AI deploys world-class AI infrastructure powered by DDN to accelerate Thailand’s digital future 142 • Deployed First 1,000-GPU Cluster following NVIDIA architecture & using DDN’s Intelligent Data Platform • Future-Ready Architecture for scalability • Optimized Data Management for Complex AI and ML Workloads BENEFITS • Accelerated AI accessibility for institutions and students. • Maintained data sovereignty with in-country infrastructure. • Delivered reliable, high-performance services for diverse workloads. • Enabled faster rollout of national AI initiatives.  Lack of Local AI Infrastructure -Thailand needed sovereign, high-performance infrastructure to support national AI development.  Complex AI Workloads - Supporting LLMs, genome sequencing, and smart city applications required robust data management.  Fast Deployment at Scale - Siam AI had to quickly stand up a scalable, future-proof platform with limited precedent in the region. "Well I believe DDN data intelligence platform give us the best chance to manage this complex complex AI cluster and then all the machine learning workloads. They give us unparalleled reach and write speed, maximize GPUs optimization, improving our checkpoint with efficiency. In doing so it gives us the best chance to give our clients the best performance and reliable service in such a big scale of the workload that we have to provide for them." Rattanapon Mongnapachan, Founder & CEO | SIAM AI © 2025 DDN
103. Consolidating Research Data and Accelerating AI- Driven Discoveries to Deliver Concrete Benefits to Society and Human Health 143 SOLUTION • A centralized DDN storage system using a global EXAScaler Lustre file system • Four fully populated systems that span a global namespace • SFA NVMe ES400NVX system with GPU integration BENEFITS • Faster data throughput with direct datapaths between storage and GPU • Improved performance, speed, and reliability significantly impacted researchers’ ability to accelerate the discovery process • A future-proof infrastructure for any additional AI applications CHALLENGES Within Helmholtz Munich there are multiple institutes that focus on different areas of research that require the need to move, store, and process large amounts of data. The storage infrastructure was hindering compute performance as well as hardware reliability. While these systems may have been appropriate when initially deployed for modest workloads, they couldn’t scale to meet the needs of more modern and sophisticated approaches to analyzing research data. “Because we pre-emptively set up high-performance DDN systems, Helmholtz Munich was well-equipped to manage and quickly access the massive data sets generated by this new wave of AI applications.” DR. Alf Wachsmann | Head of DigIT Infrastructure & Scientific Computing Helmholtz Munich © 2025 DDN
104. Encouraging Quantitative Analysts to Operate Independently With Parallel Processing 144 SOLUTION • DDN 400NVX2 QLC systems • Simple NVME-oF backend BENEFITS • Decreased processing latency by 10x • Parallel protocol enabled concurrent diverse workloads across a high number of nodes • Instant results upon deployment CHALLENGES While decentralizing the management of approaches their quantitative analysts take provides a strategic advantage in the firm’s ability to outpace competition, it makes enforcing efficient IO near impossible. As a result, to make this strategy work, Jump required a robust and highly efficient compute infrastructure capable of handling diverse workloads concurrently and with little to no processing latency. “DDN QLC systems are a really important part of that environment to get IO to our researchers as quickly as possible.” Alex Davies | Chief Technology Officer, Jump Trading © 2025 DDN
105. Finding a Nano-Scale Needle in a Peta- Scale Haystack – Anomaly Detection 145 SOLUTION • “Virtual Inspector” mass storage device; stores inspection data during defect detection scans • Runs on DDN EXAScaler Lustre (ES400X2) platform with sustained write speeds of over 40 GB/s • 6.5+ PB of scalable NVME flash storage BENEFITS • Reduced false positives and increased the accuracy of defect detection • Sped up defect detection process allowing KLA to make quicker informed decisions • Building block architecture allows KLA to easily scale up to support future research and development needs. CHALLENGES Within Helmholtz Munich there are multiple institutes that focus on different areas of research that require the need to move, store, and process large amounts of data. The storage infrastructure was hindering compute performance as well as hardware reliability. While these systems may have been appropriate when initially deployed for modest workloads, they couldn’t scale to meet the needs of more modern and sophisticated approaches to analyzing research data. “We have limited access to our system so we really needed something that was a single platform that would scale with ease.” Krishna Muriki | HPC Systems Design & HPC Architect, KLA © 2025 DDN
106. DDN Chosen by Supercomputer Provider to Accelerate Research Discoveries 146 SOLUTION • Gautschi is an NVIDIA H100 SuperPOD used for modeling simulation + AI workloads • 3 DDN EXAScaler ES400NVX2-SE nodes • Powering 9 PB raw QLC flash capacity BENEFITS • Increased bandwidth by 2.9x • 5.4x the IOPS (Input/Output Operations Per Second) of any other Purdue System • Technology that is fully integrated with the biomedical research CHALLENGES Purdue faced a significant challenge in managing the escalating volume and diversity of data being processed and accessed on their Gautschi cluster. Maintaining maximum data transfer efficiency and IOPS performance was of primary concern to ensure that, as data libraries increase in size and quantity, students and faculty can perform operations and iterate at the speed of innovation. “This investment in resources will ensure that Purdue faculty have access to cutting-edge resources to ensure their competitiveness, and they can focus on scientific outcomes instead of technology problems.” Preston Smith | Executive Director, Purdue Rosen Center for Advanced Computing © 2025 DDN
107. Scaleway Accelerates AI Innovation with DDN 147 SOLUTION • High-throughput storage • Sustainable infrastructure • Seamless integration with AI workloads BENEFITS • Scalable infrastructure • High performance and reliability • Sustainability • Operational efficiency • Future-proof infrastructure CHALLENGES As AI workloads grow in complexity, Scaleway faces several challenges, including: • Scalability: Providing enough storage to support large AI models without compromising performance. • High-performance requirements: Ensuring their data infrastructure could keep pace with the intense data demands of distributed GPU clusters. • Sustainability: Maintaining high energy efficiency and resource conservation in line with Scaleway’s environmental goals. “We’ve been blown away by the quality and linearity of DDN’s solution. It allows us to maintain consistent performance across all nodes accessing data, even as demand grows.” Adrienne Jan | Chief Product Officer, Scaleway © 2025 DDN
108. An AI Curriculum for Every Student 148 SOLUTION • NVIDIA DGX SuperPOD with DDN A³I HiPerGator 3.0, which would become recognized as the world’s fastest AI supercomputer in academia • 140 NVIDIA DGX A100 systems (1120 GPUs), 4PB DDN AI400X all- flash storage • Mellanox 200GB/s InfiniBand networking BENEFITS • Reduced processing latency for end users up to 9X with parallel protocol • Significantly boosted the processing capacity available for the university and its departments (0.7 Exaflops) • Simple building block architecture allows for seamless scalability for future needs CHALLENGES UF faced several challenges on its mission to create a pinnacle computing environment to meet its ambitious goals. As research data and computational needs expanded rapidly, UF’s existing infrastructure struggled to keep pace with the growing demands of its many users across departments. Additionally, because the cluster is shared between departments for various computational tasks, ensuring jobs could be completed in a timely manner would be essential to maintaining a swift research pace and overall beneficial experience for the end users. “We have a history of success with DDN storage systems powering our HPC computing and anticipate similar high productivity for our AI workloads.” Erik Deumens | Director of Information Technology, University of Florida © 2025 DDN
109. Empowering Research With Easy Access to Infrastructure Required to Tackle Most Pressing Life Science Questions 149 SOLUTION • A 150PB integrated storage system • Multiple 200Gb infiniband and 40Gb ethernet links • 82 Object Storage Targets BENEFITS • Easy to access storage for 1500 users • Immediately visible results of the investment • Technology that is fully integrated with the biomedical research CHALLENGES The BioHPC group brings together a large group of mathematicians and computer scientists who work on various aspects of data science and pattern recognition in biomedical data. These pattern recognition problems are heavily data-driven and require very large data sets, making it critical to have the data storage and data processing completely integrated. “For an academic medical center, it’s definitely a privilege to have this level of technology development, fully integrated with the biomedical research.” Professor Gaudenz Danuser | Professor & Chair, Lyda Hill Dept. of Bioinformatics University of Texas Southwestern © 2025 DDN
110. Genome Sequencing at Scale with NVIDIA H100 & DDN 150 SOLUTION • An end-to-end solution backed by hybrid NVME and dense disk storage for Nanopore Sequencing • DDN A³I with NVIDIA H100 GPUs for a fast system in a small footprint • Simple expansion with standard building blocks Roche BENEFITS • Decreased analysis turnaround time by more than 7X (from 15 days to 2 days) • ​​Allowed them to keep and analyze more data • Reduced the amount of data uploaded to the cloud by 100X CHALLENGES Being 100% cloud based worked fine early in their development, but after adding additional sensors and generating more data – up to 2PB every 24 hours – it became too cumbersome and slow to continue with that strategy. Limitations of being fully cloud-based required users to compromise their analysis and discard 99% of their data. Turnaround time for critical experiments was over two weeks.​ “The solution DDN had, for what we needed, just made sense. We needed a small footprint and something very fast so we could process the data as quickly as possible – time is money.” Chuck Seberino | Director of Accelerated Computing Roche Sequencing Solutions © 2025 DDN
111. Pursuing World Class IT Solutions for Economic and Human Advancement 151 SOLUTION • NVIDIA DGX SuperPod with DDN A³I data platform – codename PARRAM Siddhi – known as fastest supercomputer in India • 42 NVIDIA DGX A100 systems (340 GPUs) • 10.5PB of DDN A³I data platform • EXAScaler Software with 250 GB/s read performance • NVIDIA Mellanox HDR InfiniBand network BENEFITS • EXAScaler Parallel Protocol provides industry-leading latency reduction for 377 users from 25+ institutions • DDN A³I platform running EXAScaler software provides fast and efficient data infrastructure for healthcare and agricultural data • Simple building block architecture allows for seamless scalability for future needs CHALLENGES C-DAC was tasked with procuring a cluster powerful enough to serve as India’s AI and HPC specific cloud computing infrastructure. The cluster needed to be shared across multiple user bases including academia, R&D institutes and start-ups, and to do so with little to no processing latency. The cluster would also need to locally store India’s massive data sets from areas like healthcare and agriculture in a high-throughput and efficient data platform. “We had to establish data centers in 15 different locations within the country. For all of those we needed storage, and we chose DDN for this requirement.” Sanjay Wandhekar | Senior Director C-DAC © 2025 DDN
112. Harnessing Data to Drive Scientific Breakthroughs 152 SOLUTION • Taiga, a global parallel file system designed to span across multiple compute environments. • Three 400NVX/400NVX2 units and 1 18KX unit (Combined 25PB of NVME and HDD capacity) • DDN EXAScaler software • Nvidia HDR Infiniband networking BENEFITS • Streamlined data access and management by offering a centralized data platform across multiple compute environments. • Easy scalability accommodates increasing user demands. • Automated data backup to the NCSA’s Granite long-term archive. CHALLENGES The NCSA operates nine distinct compute environments, each with its own specialized pipelines, making data movement cumbersome for researchers. Their users needed a unified data platform to seamlessly access and store data across multiple compute environments, while ensuring scalability and performance for efficient processing and adaptability. “DDN’s Exascaler platform provides us with the performance and flexibility we need to give thousands of researchers a central, performant namespaces for all their active data needs.” JD Maloney | Lead HPC Engineer, NCSA © 2025 DDN