DEVUP Docs
Back to Dashboard

Storage

High-Performance Storage

NVMe-backed storage that delivers up to 3x the I/O throughput of standard volumes — available in select data centers for both GPU and CPU instances.

What is High-Performance Storage?

High-Performance Storage is an NVMe-backed tier for Network Volumes that provides significantly faster read/write speeds compared to standard volumes. It is designed for workloads with heavy I/O demands — large-scale model training, dataset preprocessing, and checkpoint-intensive workflows.

The DEVUP AI dashboard labels this tier with a ⚡ 3x Speed badge. This is the throughput improvement displayed in the Create Volume modal when the High-Performance toggle is enabled.

This tier applies at the volume level, not the instance level. Once a volume is created with high-performance storage, any instance that mounts it — GPU or CPU — benefits from the faster I/O.

Standard vs High-Performance

AttributeStandardHigh-Performance
Backing hardwareStandard block storageNVMe SSD
Speed (dashboard label)Baseline⚡ 3x Speed
Cost per GB/monthLower (check dashboard)Higher (check dashboard)
AvailabilityCA-MTL-4 onlyAll 4 volume data centers
GPU compatibility
CPU compatibility

Standard storage is only available in CA-MTL-4. All other volume-capable data centers (EU-FR-1, EUR-NO-2, US-CA-2) exclusively use high-performance storage — in those regions, the toggle is locked on automatically.

Data center availability

Data centerStandard availableHigh-PerformanceUI behavior
CA-MTL-4✓ OptionalToggle on/off freely
EU-FR-1RequiredToggle locked on, cannot disable
EUR-NO-2RequiredToggle locked on, cannot disable
US-CA-2RequiredToggle locked on, cannot disable

The server enforces these rules — if you attempt to create a standard volume in a high-performance-only data center, the request is rejected with an error.

Enabling High-Performance Storage

  1. Open the Create Volume modal from the Network Volumes tab.
  2. Select a data center. Data centers supporting Network Volumes show a ⚡ lightning bolt badge in the top-right corner of their card.
  3. Below the name and size fields, a High-Performance Storage section appears with a toggle switch:
    • In CA-MTL-4: the toggle is off by default — click to enable.
    • In EU-FR-1, EUR-NO-2, US-CA-2: the toggle is locked on (dimmed, non-interactive) since high-performance is the only tier available.
  4. The Total per month cost updates immediately to reflect the high-performance rate.
  5. Click Create.

High-performance is set at creation time and cannot be changed after a volume is created. To switch tiers, create a new volume, transfer your data via SSH, and delete the old one.

When to use High-Performance Storage

WorkloadRecommended tierWhy
Large-scale model training⚡ High-PerformanceFaster checkpoint writes reduce wall-clock training time
Dataset preprocessing / ETL⚡ High-PerformanceI/O-bound pipelines benefit most from NVMe throughput
Image / audio datasets (many small files)⚡ High-PerformanceRandom read performance is dramatically better on NVMe
Inference with model already in GPU memoryStandard (if available)Compute-bound — storage speed rarely a bottleneck
Long-term cold storage for infrequently used dataStandard (if available)Cost optimization — lower per-GB rate for archival data

Works with GPU and CPU instances

High-performance is a property of the volume itself, not the instance. A high-performance Network Volume can be attached to any GPU or CPU instance in the same data center — both instance types benefit from the faster I/O. No additional configuration is needed on the instance side.

FAQ

Can I convert an existing standard volume to high-performance?

No. The storage tier is set at creation time and cannot be changed. Create a new high-performance volume, transfer your data via SSH, then delete the old volume.

Why is the toggle grayed out in some data centers?

EU-FR-1, EUR-NO-2, and US-CA-2 only support high-performance storage. In these data centers, the toggle appears locked on because standard storage is not available. This is enforced both in the UI and on the server — attempts to create a standard volume in these regions are rejected.

Is the "3x Speed" label an exact benchmark figure?

The "⚡ 3x Speed" label in the dashboard indicates the NVMe tier provides up to 3x the throughput of standard storage. Actual performance depends on your workload's I/O pattern (sequential vs. random, read vs. write) and the data center's current load.

Where can I check the exact high-performance pricing?

Open the Create Volume modal, select a data center, enable the high-performance toggle, and the Total per month cost updates in real time based on your chosen size. Comparing with the toggle off (in CA-MTL-4) shows the exact price difference.