NVMe Storage

NVMe Storage on AWS

AWS offers both local NVMe SSDs (instance store) and network-attached NVMe via EBS. Understanding when to use each — and when to deploy your own NVMe-oF layer — directly impacts storage cost and performance for EC2 and EKS workloads.

NVMe-Equipped Instance Types

Instance Family NVMe Characteristics
i4i (up to 30TB NVMe) Storage-optimized: 2M+ IOPS per instance, highest NVMe density
c6id / c7gd Compute-optimized with local NVMe: ideal for analytics, databases
r6id / r7iz Memory-optimized with NVMe: in-memory databases with fast persistence
m6id / m7i-flex General purpose with local NVMe: balanced compute + fast local storage
trn1 / p4d AI accelerators with local NVMe for model checkpointing

Instance store NVMe volumes are ephemeral — data is lost on stop/terminate. Use for scratch space, caches, or workloads that replicate data to persistent storage (EBS or NVMe-oF).

NVMe-oF/TCP on AWS

For persistent shared NVMe block storage on EC2, deploy an NVMe/TCP storage cluster (e.g. simplyblock) across i4i or c6id instances. The storage nodes expose NVMe namespaces to EC2 compute nodes over the VPC network at 25–40µs latency — 3–5× lower than EBS gp3 (100–500µs typical).

For the NVMe/TCP protocol deep-dive, see nvme-tcp.com → For a full NVMe-oF architecture overview, see the NVMe-oF guide.

Recommended Storage Architecture

Tier / Use Case Recommendation
Ephemeral scratch EC2 instance store NVMe (fastest, non-persistent)
Persistent single-node EBS io2 Block Express (sub-1ms, 256K IOPS max)
Shared cluster (low latency) NVMe-oF/TCP on EC2 (25–40µs, 1M+ IOPS)
EKS PersistentVolumes NVMe-oF CSI driver or EBS CSI driver
Network AWS ENA (up to 200Gbps on some instances)

simplyblock: NVMe/TCP storage for AWS

simplyblock deploys as a software-defined NVMe/TCP storage cluster on standard AWS instances. It provides Kubernetes CSI, dynamic provisioning, and sub-40µs persistent block storage — without proprietary hardware or cloud-managed disk limits.

Managed PostgreSQL on AWS

vela.run provides managed PostgreSQL on AWS with NVMe/TCP storage — predictable sub-millisecond query latency without managing your own database infrastructure.

vela.run →

Related