Core Feature
High Performance over IP and RDMA
Quobyte supports high performance parallel IO over TCP/IP and RDMA (RoCE, infiniband, OmniPath). Clusters can operate in a mixed mode where clients and services automatically switch from IP to RDMA when available.
AI Data Lake
Quobyte AI Data Lake delivers seamless data access through every major protocol—Linux (POSIX), Windows, NFS, S3, and HDFS—so teams can work with the same datasets from any application, device, or environment.
Converged AI Storage™
Quobyte’s Converged AI Storage™ runs storage on GPU nodes for instant data flow, eliminating latency and network hops during AI training, inference, and checkpointing. It leverages local CPUs and NVMe for faster speed, efficiency, and seamless scaling.
Run on ARM processors like NVIDIA Grace Blackwell
Quobyte runs natively on ARM processors like NVIDIA Grace Blackwell, delivering high-performance, energy-efficient AI storage and compute on a unified platform. This enables scalable, real-time data access for modern AI workloads—both in the data center and at the edge.
Deploy Anywhere®
On prem on your own servers, on the public clouds or on Kubernetes. Read more...
Cloud and Hybrid Cloud Support
Quobyte runs on virtual machines or bare metal instances, with local drives or cloud block storage. Learn more about our recommended cloud instances. With our multi-cluster support you can easily combine on-prem and cloud-based clusters into a powerful hybrid-cloud storage system.
Full fault tolerance with automatic failover
All Quobyte services and data is stored redundantly using synchronous quorum replication or erasure coding. Failover is fully automatic and does not require virtual IPs, multipath other mechanisms.
Non-disruptive anything
Adding servers, removing them, hardware refreshes, rebalancing, data tiering, software updates, hardware maintenance... all of these tasks can be done while the cluster is running and serving data at full speed.
Security
Multi-tenancy
Full isolation of tenants with self-service and optional hardware isolation. Read more...
Unified ACLs
ACLs in Quobyte are unified, i.e. Quobyte translates automatically between the different ACLs like POSIX, NFSv4 or Windows. You don't have to manage different ACLs for every platform. Read more...
LDAP
Quobyte supports LDAP for the management layer, for the file system (client side translation) and for access keys.
Access Keys
Access Keys in Quobyte can be used for S3 access but also for secure multi-user and multi-tenant access on shared machines with Kubernetes, Linux or Hadoop. Read more...
X.509 Certificates
X.509 certificates ensure only authorized users and machines can access the Quobyte storage. Read more...
TLS (selective)
TLS can be enabled globally or only for specific networks, e.g. for access from outside a cluster. Read more...
End-to-End AES encryption
End-to-end-encryption ensures that no data leaves the client machine in clear text. Data is encrypted at rest and in transit. Read more...
Data Management
File Query Engine
Quickly search through billions of files based on file system and custom metadata with the Quobyte File Query Engine
Policy Based Data Management
Full control over how and where files are stored, immutability, tiering, caching or encryption. Read more
Automatic Tiering
Automatic policy based tiering of files between storage media, pools or clusters. Read more
Immutable Files
Admins can configure policies to automatically make files immutable after creation or to prevent deletion forever or for a pre-defined timespan. Immutable files are an invaluable tool to protect against ransomware attacks.
File/Object expiration
Automatically expire (and delete) files and object based on policies or user-configurable metadata.
Quality-of-Service Classes
Quobyte provides several Quality-of-Service classes for IO and metadata operations, as well as fairness between IO and metadata operations within a QoS class.
Centralized Client Configuration Management
Client configuration such as prefetching, caching, metadata caching are configured through policies and automatically applied on all clients. Changes to policies are picked up by clients within seconds. No need for local configuration on the client machines.
Data Protection
Synchronous replication of files and metadata
Synchronous quorum protection ensures data is stored on multiple servers and protects against data loss, unavailability or split-brain situations. Read more...
Erasure coding
Erasure coded files offer high performance access at very low space overhead. Read more..
Recoding of files
Policy-based recoding of files transforms replicated files to erasure coding or to wider erasure coding schemas, e.g. for archival.
Automatic file layout
The automatic file layout store small files on flash (with replication) and automatically switches to erasure coding on hard drive as the file grows. This gives you the perfect combination of high performance and price without tiering.
End-to-end checksums
End-to-end checksums do what NFS doesn't: Protect your data throughout its entire lifetime, while in transit and at rest. Read more...
Full Stack Observability
JSON REST API
The Quobyte JSON REST API gives you full control over your cluster and makes monitoring and automation a breeze.
Prometheus Exporter
All Quobyte services and clients have built-in prometheus exporters. Quobyte also implements the Consul API for automated service and client discovery.
Grafana Dashboard
The Quobyte dashboard for Grafana gives you full insight into your Quobyte cluster, the workloads and long-term trends. Read more on grafana.com...
Webconsole / GUI
Manage your Quobyte clusters through the intuitive browser-based webconsole.
Storage Architected for AI™
The only storage built from a hyperscaler blueprint, engineered to eliminate I/O bottlenecks across the entire AI/ML pipeline, from data prep to large-scale inference.
Maximize GPU Investment
Achieve peak utilization and faster time-to-result, directly boosting the return on your most expensive compute resources with Quobyte’s AI-ready performance and bulletproof resiliency.
Scalable Performance: Guaranteed Speed for Every GPU.
Predictable linear performance scaling allows you to grow your storage from a few nodes to exabytes, non-disruptively and without re-architecting, for every GPU you add.
Software Defined, Not Hardware Confined
Run storage on any server hardware, across cloud, hybrid, and edge, for CapEx savings and flexibility.
Smartphone Simplicity
Get the most out of your GPUs with Quobyte’s high throughput. Achieve up to 10 GB/s single-stream read on Linux using RDMA over RoCE. Access your data in parallel, eliminate bottlenecks, and deliver maximum performance via TCP/IP, RDMA, or a mix of both.
Quobyte: Architected for Bottleneck-Free GPU-Storage
Deep learning and LLM workloads generate extreme I/O demands — massive parallel reads during training, heavy synchronous writes for checkpoints, and low-latency access for inference.
Traditional enterprise storage can’t deliver the parallel performance, while HPC file systems often lack the reliability and uptime enterprise AI requires. The result: idle GPUs, wasted compute, and stalled projects. Quobyte delivers the massive parallel performance and the bulletproof resiliency and uptime required to keep your GPUs fed at all times.
Unleash Massive, Parallel Throughput at Scale for Faster Training
Unleash parallel throughput at scale—Quobyte is architected for massive AI and LLM training. It scales to 10,000+ clients on exabyte datasets, proven by MLPerf™ benchmarks. Optimized for QLC flash and high-speed protocols like RDMA, Quobyte keeps GPUs fully utilized and eliminates I/O bottlenecks.
Faster Checkpointing
Quobyte eliminates LLM checkpoint bottlenecks with massive parallel write I/O, spreading traffic across all servers and drives for peak throughput. Using TLC flash speeds up heavy-write operations, dramatically reducing checkpoint times. Converged AI Storage™ on GPU nodes delivers minimal latency for even faster, interruption-free checkpointing.
Accelerate LLM Fine-Tuning with Quobyte's AI Data Lake
Quobyte is the AI Data Lake: a fast, unified storage platform delivering massive throughput and direct fine-tuning access. It breaks down data silos with native multi-protocol support, giving every team seamless, full-speed access to shared data.
Data scientists can instantly provision, version, and snapshot in a global namespace, and leverage the File Query Engine for rapid metadata searches—streamlining discovery and collaboration.
Guaranteed Uptime and Consistent Performance for Inference
When models move to production, storage must provide consistent, low-latency I/O for real-time applications without fail. Quobyte’s bulletproof resiliency ensures maximum uptime, eliminating maintenance windows and guaranteeing service continuity. The predictable, low-latency I/O required for real-time model serving and inference is guaranteed.
The same storage that powers your training phase seamlessly supports inference workloads, ensuring immediate data availability and predictable performance across your entire AI lifecycle.
Zero-Delay Data Ingestion
The first AI bottleneck is moving massive datasets onto the cluster. Quobyte Data Mover syncs petabyte-scale data in parallel from sources like S3 or Azure, slashing setup time. For use cases like autonomous driving or IoT, Quobyte’s parallel ingest absorbs high-volume data bursts without stalls—even at peak load.
High-Performance Data Curation at Scale
Quobyte manages billions of files—including Parquet and ZIP—with fast access at any scale. Its metadata-accelerated File Query Engine enables instant searches using custom tags, avoiding slow crawls and metadata file issues. This lets data scientists quickly find, organize, and prepare data, speeding model development and iteration without bottlenecks.