Vultr Explained: Cloud Compute, Bare Metal, GPU, and Locations

Account, users and API keys

A Vultr account is a single billing entity with an owner login. Under Account, Users you can add additional users, each with their own password, two-factor setting and a permission list that limits what they can do (for example, manage servers but not billing). The customer portal is the web console; the same operations are available through the Vultr API v2, the vultr-cli tool and an official Terraform provider.

API keys are issued per account and can be restricted to a list of allowed source IP addresses, which you should set for any key used from a fixed server or CI runner. Enable two-factor authentication on the owner login and on every user with create or destroy rights. SSH keys are stored at the account level and selected when you deploy a server, so add your team’s public keys once and reuse them.

Cloud Compute and Optimized Cloud Compute

Vultr’s virtual machines are split into two groups. Cloud Compute uses shared vCPUs and comes in Regular Performance, High Performance (AMD or Intel) and High Frequency tiers, with the tiers differing in CPU generation, clock speed and storage type. This is where small web servers, side projects and most staging environments live.

Optimized Cloud Compute uses dedicated vCPUs that are not shared with other tenants, with General Purpose, CPU Optimized, Memory Optimized and Storage Optimized sub-families. These suit databases, build farms and anything whose performance should not vary with neighbouring workloads. Both groups are described on the Cloud Compute product page.

Servers deploy from Vultr’s operating system images, from the Marketplace of preconfigured applications, from your own snapshots, or from a custom ISO you upload. Snapshots are stored per account and can be restored to a server in any location that supports the same plan type, which makes them a simple cross-region migration tool.

Bare Metal and Cloud GPU

Bare Metal gives you a whole physical server with no hypervisor, provisioned through the same portal and API as virtual machines and billed on the same hourly model. It suits workloads that need consistent single-tenant performance, licensing tied to physical cores, or a custom hypervisor of your own.

Cloud GPU provides instances with NVIDIA accelerators such as the A16, A40, L40S, A100, H100 and GH200, and AMD Instinct MI300X, available either as fractional GPU slices for inference and smaller models or as full and multi-GPU nodes for training. GPU availability varies by location and changes as hardware generations move, so check the Cloud GPU page for the current list. Vultr also offers a serverless inference service and a container registry aimed at teams deploying models. GPU plans are the main reason buyers who search for a Vultr account for AI development look at the provider; the account itself is the same as any other Vultr account, and GPU quota is granted by Vultr on request.

Locations

Vultr runs more than thirty data center locations at the time of writing, listed on its locations page. They include multiple United States cities (New Jersey, Silicon Valley, Los Angeles, Chicago, Dallas, Seattle, Atlanta, Miami, Honolulu), Toronto and Mexico City, São Paulo and Santiago, Amsterdam, London, Manchester, Frankfurt, Paris, Warsaw, Stockholm and Madrid, Tokyo, Osaka, Seoul, Singapore, Mumbai, Bangalore, Delhi NCR, Sydney, Melbourne, Tel Aviv and Johannesburg.

Every location offers Cloud Compute; Optimized Cloud Compute, Bare Metal, GPU and some managed services are available in a subset. There are no availability zones within a location. Block storage, VPCs, reserved IPs and snapshots are location-bound, so a resilient design places servers in two locations and replicates data between them.

Networking, storage and managed services

Each server gets a public IPv4 and IPv6 address. VPC networks provide private addressing between servers in the same location, and the newer VPC 2.0 adds larger, more flexible private networks. Reserved IPs are public addresses you keep and reassign for failover. A cloud Firewall applies rule groups at the network edge before traffic reaches the server. Load Balancers distribute HTTP or TCP traffic across a server pool with health checks. Vultr also hosts DNS zones for your domains.

Storage comes as Block Storage volumes in NVMe and HDD classes, Object Storage with an S3-compatible API, and automatic Backups enabled per server for a percentage surcharge. Managed services include Vultr Kubernetes Engine (VKE), with a free control plane and worker node pools billed as compute, and Managed Databases for MySQL, PostgreSQL, Valkey-compatible caching and Apache Kafka.

Shared vCPU compute

Cloud Compute: Regular, High Performance, High Frequency

Dedicated vCPU compute

Optimized Cloud Compute families

Physical servers

Bare Metal, hourly billed

Accelerators

Cloud GPU, NVIDIA and AMD, fractional to multi-GPU

Managed

VKE Kubernetes, Managed Databases, Load Balancers

Billing, credit, limits and acceptable use

Vultr bills hourly with a monthly cap per resource, and charges accrue to the account balance. You can pay by card as you go, or add a deposit in advance that usage draws from; promotional credit, when Vultr offers it, works the same way and expires on the terms Vultr sets. Billing alerts can be configured so an email fires when the monthly charge or balance crosses a value you choose.

New accounts have limits on the number of instances, vCPUs, GPU instances and block storage they can create. Limits rise as an account builds a payment history and can be raised sooner by opening a support ticket that explains the workload. The CloudAccountHub product page lists Vultr in credit tiers labelled from Vultr 200$ Credit to Vultr 305$ Credit; each label states the listed credit amount and nothing beyond it, and readers searching for a Vultr free trial account should read the listing labels literally, since Vultr’s own trial terms are separate and subject to change.

Vultr’s terms of service and acceptable use policy apply to every account. Outbound SMTP on port 25 is blocked by default on new accounts, abuse reports are enforced, and mail sent from Vultr servers must comply with anti-spam law. Use an authenticated relay for application email.

Where to go next

• See the listed credit tiers on the Vultr account product page.
• Compare it with its closest rivals in Vultr vs DigitalOcean and UpCloud vs Vultr.
• Learn what to look for in accelerators in Cloud GPUs for AI workloads.
• Understand VPCs and reserved IPs in Cloud networking basics.
• Protect API keys with API keys and secrets management in the cloud.
• Browse every guide in the Cloud Library.