Log In
Order 0

Buy online or call

1-800-371-1212

DataFlow HPS: High-end AI/HPC Storage Built for Speed and Reliability

DataFlow HPS 1U Unit

The DataFlow family represents a curated lineup of storage platforms, each optimized for specific use cases and engineered to deliver scale, speed, and long-term reliability.

DataFlow HPS Storage with VDURA delivers extreme endurance and performance through a parallel file system engineered for sustained velocity. Built on a scalable architecture combining director nodes, flash nodes, and HDD JBODs, it ensures uncompromising durability and responsiveness at scale.

Powering AI and HPC Workloads at Scale

DataFlow HPS combines all-flash and hybrid storage architectures to deliver NVMe-class performance for AI, HPC, simulation, and analytics workloads. The architecture scales linearly to thousands of nodes, allowing capacity and performance to grow without data migration or application disruption.

Key Highlights:

Architecture Overview

The VDURA architecture separates metadata and data for optimal performance, using VeLO — an NVMe-optimized metadata engine — to accelerate I/O operations. Its shared-nothing design ensures resilience, while the DirectFlow client enables parallel access for low latency and consistent throughput across all nodes.



Flexible Configuration Options

Thinkmate’s DataFlow HPS line offers pre-configured solutions in all-flash or hybrid configurations to balance performance and cost. Our experts can help you select the right system or design a custom architecture tailored to your requirements.

Model Type Capacity Configuration Throughput (Read/Write) Availability Best For
DataFlow HPS-500F All-Flash 500 TB NVMe Director + Flash Nodes 210 / 175 GB/s 99.99999 / 99.9 % Small HPC clusters, fast checkpoints
DataFlow HPS-1000F All-Flash 1 PB NVMe Director + Flash Nodes 210 / 175 GB/s 99.99999 / 99.9 % AI training, real-time inference, large-scale checkpointing, Gen AI
DataFlow HPS-4500 Hybrid 4.5 PB Flash + HDD Nodes 105 / 70 GB/s (NVMe), 37 / 25 GB/s (HDD) 99.999 / 99.99 % Mixed AI/HPC, large-scale research
DataFlow HPS-7000 Hybrid 7 PB Flash + HDD Nodes 140 / 105 GB/s (NVMe), 50 / 37 GB/s (HDD) 99.999 / 99.99 % Long-term data and model archives
Request a Custom Configuration

Typical HPS Storage Use Cases

Use Case Typical Workloads
AI Training & Inference LLM training, RAG pipelines, AI datasets
GPU Clusters & AI Factories Multi-GPU environments and AI infrastructure
HPC & Scientific Computing Simulation, modeling, research computing
Data Analytics & Data Lakes Large-scale analytics and unstructured data
Engineering & Simulation EDA, CFD, CAE workloads
Media & Rendering Video, animation, and visual effects

Why Partner with Thinkmate?

Co-Design and Sizing

Work with Thinkmate engineers to size performance, capacity, and storage architecture based on your AI, HPC, analytics, or simulation requirements.

Validated Storage Architectures

DataFlow HPS solutions are pre-configured and tested to deliver the throughput, scalability, and reliability required for demanding AI and high-performance workloads.

On-Site Ready

Systems arrive ready for rack-and-stack, with management tooling and network design recommendations to get you into production quickly.

Designed to Scale

Expand capacity and performance as requirements grow while maintaining consistent operation across AI, HPC, and analytics environments.

Speak with a Storage Expert

Common Questions About High-Performance Storage

AI training, inference, and analytics workloads often require multiple GPUs or compute nodes to access data simultaneously. High-performance storage helps eliminate data bottlenecks and provides the throughput needed to keep GPUs and compute resources fully utilized.

Not necessarily. Many AI and HPC environments use a combination of flash and high-capacity storage tiers to balance performance, capacity, and cost. Thinkmate can help determine the right architecture for your workload.

High-performance storage is commonly used for AI and machine learning, HPC, simulation, analytics, engineering, life sciences, media production, and other workloads that process large datasets or require high-throughput data access.

Yes. AI and HPC environments can generate enough parallel I/O to overwhelm traditional storage systems. High-performance storage helps ensure that GPUs and compute nodes receive data fast enough to maintain application performance.

Storage requirements depend on factors such as dataset size, GPU count, throughput requirements, and application workloads. Thinkmate experts can help determine the appropriate balance of capacity, performance, and cost for your environment.