Use Case · AI & Research

Storage Fast Enough for Your GPUs

GPU clusters sit idle when storage can’t feed them fast enough. TrueNAS R60 delivers 60 GB/s of NVMe throughput over 400 GbE — purpose-built for model training, inference pipelines, and large-scale scientific data.

Request a Quote View TrueNAS R60

The Challenge

Modern AI and data science workloads have outpaced most storage architectures. Training a large language model or running genomics pipelines requires storage that can sustain hundreds of gigabytes per second to keep GPU utilization high — storage bottlenecks are directly tied to compute cost and time-to-result.

At the same time, research environments need data integrity, long-term retention, and the ability to share datasets across multiple compute nodes simultaneously. TrueNAS combines PCIe Gen5 NVMe throughput, Terabit Ethernet connectivity, and OpenZFS data integrity in a single platform — without the per-TB licensing costs of proprietary parallel file systems.

How TrueNAS Supports AI and Research Workloads

High throughput, low latency, and the data integrity that long-running compute jobs demand.

Throughput and Connectivity

TrueNAS R60 — PCIe Gen5, DDR5, 400 GbE, 60 GB/s

The TrueNAS R60 is the first R-Series platform built with PCIe Gen5 and DDR5 RAM, delivering up to 60 GB/s of throughput with ultra-low latency. Supporting Terabit Ethernet at speeds up to 400 GbE, the R60 provides a high-performance foundation for AI inference, data science pipelines, and high-resolution scientific data that demand extreme throughput and consistent latency at petabyte scale.

NVMe-oF for GPU-Direct Storage

TrueNAS supports NVMe over Fabrics (NVMe-oF), enabling GPU-Direct Storage pathways that bypass CPU overhead — delivering storage data directly to GPU memory for the highest possible training throughput.

NFSv4 with RDMA

TrueNAS R-Series supports NFSv4 over RDMA (Remote Direct Memory Access), reducing CPU cycles per I/O operation and delivering lower latency for compute-intensive parallel workloads across multiple nodes.

Data Management at Scale

S3-Compatible Object Storage

Expose TrueNAS datasets as S3-compatible object storage via MinIO AIStor. ML frameworks including PyTorch DataLoader, TensorFlow I/O, and Hugging Face Datasets read directly from S3 endpoints — point them at on-premises TrueNAS instead of cloud storage.

Immutable Snapshots for Experiment Reproducibility

Snapshot a dataset before a training run in milliseconds. If the experiment needs to be reproduced exactly, roll back to the identical data state — essential for reproducible research and ML experiment tracking.

Scale to 7 PB in a Single Namespace

TrueNAS R60 scales to 7 PB raw (14 PB effective) with up to 4 expansion shelves — large enough to hold entire training corpuses, raw genomics data, or multi-year telescope observation archives in a single addressable namespace.

Cloud Tiering and Replication

Replicate completed datasets to Amazon S3, Google Cloud, or Azure for long-term archival. Bring data back on-premises for the next compute run — keeping active working sets on fast NVMe and cold archives in low-cost cloud tiers.

Data Integrity and Security

OpenZFS — Every Byte Checksummed, Every Training Run Protected

Silent data corruption during a training run can propagate silently through model weights, producing incorrect results that are expensive to diagnose. OpenZFS checksums every block end-to-end — corrupted data is detected and repaired from redundant copies before it reaches the compute layer. For research environments where reproducibility is non-negotiable, this guarantee matters.

TrueSecure for Sensitive Research Data

FIPS 140 encryption for data-at-rest and in-flight, NIST 800-209 compliance, KMIP key management, and encrypted datasets protect sensitive genomics, patient, and proprietary model data end-to-end.

Multi-Tenant Access Control

TrueNAS supports Active Directory, LDAP, and FreeIPA integration with per-dataset permissions. Separate research group datasets with distinct access controls while sharing the same physical storage pool — no data leakage between projects.

Trusted by leading organizations

Recommended TrueNAS Systems for AI & Research

Models commonly chosen for this workload, with reasoning.

V-Series

TrueNAS V160

4U tri-mode · 24+4 bay · 400 GbE

Hyperconverged NVMe with NVMe-oF over RDMA and MinIO AIStor — feeds DGX, H100, and Blackwell clusters at line-rate.

View V160 →

F-Series

TrueNAS F100

2U all-flash · 48-core · 20 PB

All-NVMe flash dataset store with 48-core throughput. Pairs well with a separate object tier for training corpora.

View F100 →

R-Series

TrueNAS R60

12-bay all-NVMe · 60 GB/s

All-NVMe rackmount when a single-controller appliance is acceptable. 12-bay 2U at 60 GB/s, easier to scale horizontally.

View R60 →

Authorized TrueNAS Reseller

Stop letting storage bottleneck your GPUs

Tell us your GPU cluster size, dataset scale, and network fabric — we’ll size the right TrueNAS configuration and stay on the order until it ships.

Talk to a Specialist View TrueNAS R60

Recommended Hardware for AI & Research

Primary Recommendation

TrueNAS R60

PCIe Gen5, DDR5, 12× NVMe, 60 GB/s, 400 GbE, 7 PB raw. The purpose-built platform for GPU cluster storage and large-scale data science.

View R60

High-Capacity All-Flash

TrueNAS F-Series

F60 (9 PB, 60 GB/s) and F100 (20 PB, 8× 100 GbE) for research environments requiring maximum NVMe capacity with dual-controller HA.

View F-Series

Hot + Warm Tiered

TrueNAS M-Series

Hybrid NVMe + HDD with TrueCache NVDIMM for environments that need flash-tier performance for active datasets and high-density spinning disk for deep working archives up to 30 PB.

View M-Series

Find Your Fit

Not sure which series fits your workload?

Four deployment paths. Pick the one that fits.

Home, Office & Remote Site

Quiet, compact ZFS storage for creators, home labs, and branch offices. ECC RAM, 10 GbE, and IPMI in a desktop or 1U chassis.

Mini X+ Mini R

Explore Mini Series

SMB & Mid-Market Rackmount

High-density hybrid and all-NVMe rackmount NAS. From 3.4 PB hybrid up to 60 GB/s NVMe.

R20 R50 R60

Explore R-Series

Enterprise HA & High-IOPS

Dual-controller failover with TrueCache NVDIMM, hybrid or all-flash, and 25/100 GbE. Built for VMware, SAP, and 24/7 business-critical workloads.

H-Series M-Series F-Series

Compare Enterprise

AI/ML & Petabyte Scale

Tri-mode SAS + Gen4 NVMe with 400 GbE fabric and 60 GB/s system bandwidth. MinIO AIStor object storage for GPU clusters and AI/ML pipelines.

V160

Explore V-Series