Infinia Platform Overview - Infinia is a software-defined, metadata-driven, containerized, cloud-native data intelligence platform designed for scalability, performance, and efficiency across core, cloud, and edge environments [1] - The platform supports critical data protocols like object and block, integrating with AI data acceleration libraries like TensorFlow and PyTorch [1] - Infinia enhances AI execution engines by serving data in its native form, reducing the need for data conversion and speeding up applications [1] Metadata and Multi-Tenancy Capabilities - Infinia allows for tagging massive amounts of metadata to objects, enabling faster data discovery and processing, with no limitations on metadata capability [1] - The platform has built-in multi-tenancy capabilities, providing SLAs for individual tenants and sub-tenants on capacity and performance, ensuring quality of service [1] Scalability and Cloud Native Design - Infinia is fully containerized, allowing for scale-out at web scale, starting from a few terabytes and scaling to exabytes [1] - The product is designed to be cloud-native and will soon be available in leading cloud provider marketplaces [2] AI Data Challenges and Solutions - Infinia addresses the complexity of managing large amounts of distributed multimodal data across core, cloud, and edge environments by creating a unified platform [1] - It tackles the demand for extremely low latency required to run AI applications, as well as the high costs associated with running AI [1] - The platform ensures data protection at any time and at any scale [1] Performance Metrics - Infinia can deliver time to first byte in less than a microsecond [2] - It can deliver 30 to 40 million objects per second in list object operations [2] - Infinia can deliver terabytes per second throughput at large scale [2] Efficiency and Sustainability - Infinia can achieve 10x data reduction, fitting over 100 petabytes of storage into a single rack [2] - It can reduce the overall data center footprint by a quarter compared to competitors, saving 10x power and cooling costs [2] Security Features - Infinia focuses on security authentication and access control, preventing unauthorized data access [2] - Data is always encrypted, and all actions within the system are audited [2] - The platform provides 99.9999% uptime enabled by reliability-focused features [2] Key Business Outcomes - Infinia aims to reduce complexity and achieve more accurate results on a unified platform for AI inference, data analytics, and model training [2] - It accelerates innovation by running AI apps faster, enabling businesses to beat the competition [2] - The platform enables rapid deployment across the cloud core and the edge to increase productivity, boost efficiency, and maximize ROI [2]

let's bring up Sven our CTO so James just mentioned and talked a lot about exos scaler now it's time time to talk about the Next Generation infinia good afternoon uh my name is Sven I'm ddn CTO and I would like to talk a little bit about infinia um infinia is a really interesting project uh we started this about seven years ago um and we took a very different approach on how we get it to Market so instead of just gaing it we basically picked a couple of very large scale customers we have actually a system r ...

Core Message - AI is essentially HPC (High Performance Computing), emphasizing the importance of data in both fields for extracting intelligence and value [1] - DDN (DataDirect Networks) provides the "rocket fuel" or data intelligence infrastructure that enables better, faster, and more accurate insights from massive datasets in real-time [1] - Data intelligence is critical for AI transformation, enablement, and acceleration, requiring the unification, curation, and analysis of distributed data from various sources [1] Challenges and Solutions - Current challenges hindering AI acceleration include GPU scarcity, limited data center space, and insufficient power; a data intelligence framework is needed to alleviate these issues [2] - DDN's solutions focus on delivering more capabilities from existing GPUs, shrinking data center footprint, and lowering power consumption [2] - DDN accelerates data ingestion, freeing up GPU cycles, and optimizing networks to reduce time to insight and enhance value [2] DDN's Technology and Positioning - DDN is the only data intelligence platform deployed internally at NVIDIA, and also supports massive deployments like XAI with over 100,000 GPUs [1] - DDN's new technology, Infinia, is a high-performance, multi-tenancy data intelligence platform that supports multiple protocols and minimizes data movement [2] - DDN's solutions maximize the value from infrastructures deployed at scale in data centers and the cloud, benefiting both HPC and AI applications [3] Market Impact and Growth - DDN powers more than half a million GPUs and has deployments at the exobyte level, demonstrating significant growth and scale [3] - DDN's ability to solve challenges at massive scale translates to bulletproof stability and cost-effectiveness across a broad range of installations [3] - DDN aims to accelerate scientific and business outcomes by handling data at the edge, in data centers, and in the cloud [3]

AI Model Efficiency & Adoption - DeepSync's approach highlights opportunities for significantly more efficient AI models than previously thought [1] - This efficiency is accelerating the adoption of AI across various sectors [1] Product Development & Integration - The company has launched R1, indicating a new product or platform [1] - R1 is designed to interact with Infinia data intelligence layer to solve problems [1]

Multimodal AI & Metadata - Multimodal AI is essential for enterprises to truly benefit from AI [1] - Metadata tagging and movement are crucial for the economics of AI to work, given limitations in data center space and power [1] - Metadata attributes are very key to the success of AI initiatives [1] Data Intelligence & Business Value - The core focus should be on a metadatarrier infrastructure that can handle low-latency object transformation to gain insight [2] - The ultimate goal of bringing data into the environment to train models is to gain data intelligence [2] - AI initiatives must result in business value for enterprises and leisure value for consumers [3]

Solution Overview - DDN provides a one-click high-performance RAG (Retrieval-Augmented Generation) pipeline for enterprise use, deployable across various environments [1] - The RAG pipeline solution incorporates NVIDIA NIM within the NVIDIA Nemo framework, hosting embedding, reranking, and LLM models, along with a MILV vector database [2] - DDN Infinia cluster, with approximately 0.75 petabytes capacity, serves as the backend for the MILV vector database [3] Technical Details - Infinia's AI-optimized architecture, combined with KVS, accelerates NVIDIA GPU indexing [3] - The solution utilizes an NVIDIA AI data platform reference design to facilitate the creation of custom knowledge bases for extending LLM capabilities [4] - The one-click RAG pipeline supports multiple foundation models for query optimization [7] Performance and Benefits - Integration between DDN Infinia and NVIDIA Nemo retriever, along with NVIDIA KVS, results in faster response times, quicker data updates, and rapid deployment of custom chatbots [9] - The RAG pipeline enables detailed and accurate responses to specific queries, as demonstrated by the Infinia CLI hardware management features example [8][9]

Core Technology - The company utilizes accelerated computing and artificial intelligence to learn from data [1] - The company transforms raw data into data intelligence [1] - The company embeds intelligence into models and extracts semantics, intelligence, and information from data [2] - Instead of serving raw data, the company serves metadata, knowledge, and insights [2] - The semantic layer of data is extremely compressed [2]

Product Overview - Infinia is a ground-up developed product intended as more than just an object store, offering multiple data interfaces [1] - The system provides database access through an SQL interface and can be used as a notification bus, critical for AI workflows [2] - It is a pure software product designed to solve complicated problems, making it attractive for service providers [3] Key Features & Capabilities - Built-in architectural capabilities around multi-tenancy provide technology as a service [4] - Deep integration capabilities through an SDK enable customers to accelerate applications and AI workflows [4] Ecosystem & Partnerships - DDN has ongoing partner development with ecosystem partners to accelerate technology adoption through the Infinia SDK [4]

Overview - DDN Infinia 在 Oracle Cloud Infrastructure (OCI) 上的性能展示，但强调这仅为技术预览，未来可能发生变化 [1] - Infinia 架构提供广泛的数据管理能力，包括多种数据 IO 路径、核心存储组件（如 scale-out KV 存储、always-on 加密和数据缩减）、原生多租户等 [2] - Infinia 完全软件定义和容器化，可在物理或虚拟化硬件上运行，适用于云部署 [2] Technical Details & Performance - 在 OCI 内部测试使用了 6 个 BM dense ioe5 计算实例作为 Infinia 集群的主机，以及 6 个 BM standard E5.192% 实例作为客户端，客户端实例具有单个 100 GB 连接 [2] - 在 dense ioe5 实例中，Infinia 软件仅使用了 128 个可用核心中的 32 个 [2] - 使用 warp 在分布式基准测试模式下进行 IO 生成，确保每个客户端并发地向所有 Infinia 集群节点发送操作，并在所有客户端和所有 Infinia 集群节点之间创建完整的 IO 网格 [3] - Put 操作的吞吐量约为 28 GB/s 到 30 GB/s，每个客户端和每个 Infinia 节点平均处理速度约为 4800 MB/s (约 5 GB/s) [5] - Get 操作的吞吐量约为 35 GB/s 到 37.5 GB/s，负载均匀分布在所有客户端和 Infinia 节点上，约为 6100 MB/s (约 6.5 GB/s) [6] - 实现了 5 毫秒的 time to first byte，对于 S3 对象 IO 来说非常出色 [6] Conclusion - 软件定义的 Infinia 不仅可以在云中的 Oracle 计算基础设施上运行，而且即使是小型六节点集群也能实现出色的性能 [7]

AI Infrastructure Vision - Infinia initiated a new product development in 2017, driven by the need for a different architecture for AI, diverging from existing solutions [1] - The company aimed for an architecture that scales efficiently for training and offers very low latency [1] - Infinia envisioned a distributed, on-premise, and multi-cloud solution where data payloads (images, video) do not move due to cost, relying instead on metadata and tagging [2] Challenges and Innovation - The initial architectural concept for AI was met with skepticism, deemed impossible by some [2] - Infinia adopted a problem-solving approach that disregarded past constraints like file systems [3] - The development process took seven years [3]