TuringData Flash F9100/F9200 Series

The TuringData Platform Appliance for AI and HPC

Deploy TuringData Platform in an integrated, all-flash appliance engineered to accelerate model training, support lower-latency inference pipelines and keep GPU clusters productive.
AI-Native All-Flash Storage
TuringData Flash combines TuringData Platform software with NVMe storage and high-speed InfiniBand / RoCE or Ethernet networking in a scale-out appliance. It provides a streamlined path to deploying high-performance storage for AI training, inference and HPC workloads, starting with three storage nodes.
Benchmark-Proven Performance for GPU-Intensive Workloads
Start with Just3Nodes
7.5M IOPS
Handle metadata-intensive AI workloads
541.5GiB/s
MLPerf® Storage v3.0 3D U-Net read bandwidth
90%+
GPU utilisation for greater infrastructure efficiency

*Results may vary by configuration and workload.

Keep Data Moving. Keep GPUs Productive.

High-Throughput Data Delivery

Supply GPU clusters with high-speed data access for demanding training, checkpointing, inference and HPC workloads.

Efficient Small-File Processing

Use distributed metadata and parallel I/O processing to accelerate workloads involving millions of files.

Balanced Data Access

Distribute concurrent I/O across the appliance cluster to reduce traffic hotspots when many compute nodes access the same data.

Scale-Out Growth

Start with a three-node appliance cluster and expand capacity and performance as datasets and workload requirements grow.

Start with Three Nodes. Scale as Demand Grows.

Deploy TuringData Flash with a three-node starting configuration, then expand capacity and performance as your AI workloads grow.

Key Features

GPUDirect

Supports NVIDIA GPUDirect® Storage, achieving high performance to meet AI workload requirements.

Elastic Data Network

Integrates multiple network planes within storage clusters to accelerate both AI model training and inference.

Multi-Tenancy

Supports multi-tenant deployments with per-directory QoS and quota management, ensuring each user or team receives predictable performance and fair resource allocation.

Distributed Metadata

A non-volatile, distributed metadata architecture ensures superior resilience, scalability, and performance.

Unified Global Namespace

Enables real-time read/write access to data across edge, core, and cloud with a unified global namespace.

Smart DataLoad

Seamlessly integrates object storage and file systems, enabling data movement driven by business needs and efficient management.

GPUDirect

Supports NVIDIA GPUDirect® Storage, achieving high performance to meet AI workload requirements.

Elastic Data Network

Integrates multiple network planes within storage clusters to accelerate both AI model training and inference.

Multi-Tenancy

Supports multi-tenant deployments with per-directory QoS and quota management, ensuring each user or team receives predictable performance and fair resource allocation.

Distributed Metadata

A non-volatile, distributed metadata architecture ensures superior resilience, scalability, and performance.

Unified Global Namespace

Enables real-time read/write access to data across edge, core, and cloud with a unified global namespace.

Smart DataLoad

Seamlessly integrates object storage and file systems, enabling data movement driven by business needs and efficient management.

Setting a New Benchmark for

Performance and Cost Efficiency

TuringData Flash can be deployed and fully operational with just three nodes, running stably while providing unparalleled performance and cost efficiency.
Storage Appliance

Find out how much GPU utilization you're leaving on the table. Schedule a free 30-minute architecture review with our experts.