Everyone running AI and HPC infrastructure is under pressure to get more from what they already have. Adding new hardware takes months, increases power demand, and does little if storage is still the bottleneck for your GPUs.
Quobyte 5 gets more out of the x86 and ARM servers you already own. It delivers substantially higher I/O and metadata performance from the infrastructure already in place, helping you push those systems further before additional hardware is required.
Quobyte 5 also changes how teams understand what is happening across the file system. As file systems grow, understanding what is stored, where capacity is going, and how data is protected becomes harder without specialist knowledge. Quobyte 5 makes that information easier to access with the new MCP server interface and an expanded Quobyte File Query Engine, letting your teams ask questions in plain language and get answers straight from live file-system metadata.
Get More Performance From the Hardware You Already Have
Quobyte 5 advances how the file system uses the hardware it runs on. Modern server CPUs deliver much of their performance through higher core counts, and Quobyte 5 is designed to scale across those cores on both current and previous-generation systems. On the network, RDMA multi-link support with Quobyte’s own link aggregation enables the file system to utilize multiple InfiniBand or RoCE links in the same server, which matters most on dense GPU nodes that carry four or more high-speed interfaces.
Higher performance means getting more work from the servers you already own. For a given workload, you can run on fewer servers and use less power.
The same engineering principles produced our results in the MLPerf® Storage 2025 3D U-Net benchmark, the most I/O-intensive workload in the MLPerf Storage suite. A mid-tier all-flash Quobyte cluster with just four nodes drove six H100 GPUs per client at 90.29% utilization over a 200 Gb/s RoCE network. Quobyte delivered the fastest unverified result in the comparison, 33% ahead of the closest competitive alternative, while operating almost at line speed with the least amount of hardware and efficient power consumption.
Quobyte 5 builds on the same foundation, delivering Fastest AI-Ready Performance from the infrastructure you already operate, and extending how far it can go before you add new systems.
Ask Your File System a Question
Finding an answer in a large storage environment has always required expertise: which report to open, which command to run, which metadata to query, how to interpret the result. With Quobyte 5, you start with the question. The MCP server interface lets LLMs work with live file-system information through the Quobyte File Query Engine.
On one of our test clusters, we asked:
“Are we using our flash tier efficiently? How much data could we shift to HDD?”
The answer showed that 92% of the data was sitting on SSD. Every file looked recently accessed, which suggested nothing was inactive. The LLM did not take that at face value. It identified a pattern consistent with a scan touching every file, switched to modification time, and explained why that was the more reliable signal. On that basis, files not modified for more than 90 days made up 76% of the data on SSD, giving us a clear place to investigate.
From there, we kept going in plain language: Which tenants are consuming the most capacity? Which volumes hold the oldest data? Every answer came from live file-system metadata, in seconds.
A specialist could have found all of this, including the access-time problem. The difference is that users can now start with the question instead of the query syntax, while the same queries run across billions of files.
Get Answers From Live Metadata With the Quobyte File Query Engine
Behind the MCP server interface is the Quobyte File Query Engine. It queries live file-system metadata directly and is highly parallelized, scanning and filtering billions of files. There is no additional metadata index to maintain or keep synchronized.
Quobyte 5 adds TOP and HISTOGRAM query operators, along with new queryable properties for volume, tenant, and xattr names. The same permissions and ACLs that govern file access also govern query results, so users see only metadata for files they already have permission to access.
Put More of Your GPU Nodes to Work
Your GPU nodes contain substantial CPU resources and local NVMe alongside the accelerators. In a typical cluster, those CPUs run more than 70% idle.
Quobyte GPU Converged Storage runs directly on those nodes and turns that surplus CPU and NVMe into high-performance, resilient storage. As your GPU fleet grows, additional nodes can bring storage capacity and throughput along with compute.
Quobyte 5 strengthens this deployment model with support for multi-socket and multi-NIC systems, together with RDMA multi-link support, and adds automatic recovery after power loss. Throughout all of it, Quobyte’s Bulletproof Resiliency keeps your file system available as individual GPU nodes reboot, fail, or join and leave the cluster.
You can also operate GPU nodes and dedicated storage servers in the same cluster, with policy-based data management controlling placement across NVMe, SSD, HDD, and remote storage.
Idle resources inside your AI infrastructure now contribute directly to the storage performance those same workloads depend on.
Spend Less Time Managing Storage
Quobyte 5 simplifies day-to-day operations with fully integrated PKI and certificate management, covering CA and CSR management, client certificate permissions, and revocation. S3 adds a new S3 Browser, bucket versioning, and object lock with retention support, while system management gains JSON and YAML support in the CLI, IPMI monitoring, and license updates without a restart.
A streamlined installation process with pre-flight checks catches configuration issues before deployment.
Storage Architected for AI™
Quobyte is Storage Architected for AI™, delivering Fastest AI-Ready Performance, Bulletproof Resiliency, and smartphone-like simplicity at any scale.
With Quobyte 5, you get more from what you already have. Your existing hardware delivers substantially more I/O and metadata performance. Surplus CPU and NVMe inside your GPU nodes contribute high-performance storage. The knowledge inside your file system becomes easier for your team to reach with a question in plain language.
Turning existing hardware into lightning-fast AI storage is ultimately a software problem. Quobyte 5 solves more of it with the infrastructure you run today, while preserving your freedom to choose the systems you deploy next.
Join our deep-dive webinar on the performance work and the new MCP server interface on October 7, 2026.
