Creating a Qdrant Cloud Cluster
Qdrant Cloud offers two types of clusters: Free and Standard. Premium customers create Standard clusters with additional options.
Free Clusters
Free tier clusters are for prototyping and testing. You don’t need a credit card to join.
| Resource | Value |
|---|---|
| RAM | 1 GB |
| vCPU | 0.5 |
| Disk space | 4 GB |
| Nodes | 1 |
This configuration supports serving about 1 M vectors of 768 dimensions. The choice of cloud providers and regions is limited.
A free cluster includes:
- Standard support
- Basic monitoring, log access, and alerting
- Version upgrades with downtime
- Manual snapshots and restores through the API only
- No dedicated resources
If unused, free tier clusters are automatically suspended after 1 week, and deleted after 4 weeks of inactivity if not reactivated. To reactivate a suspended cluster, see Reactivate a Suspended Cluster. You can upgrade to a standard cluster at any time.
Standard Clusters
On top of the Free cluster features, Standard clusters offer:
- Response time and uptime SLAs
- Dedicated resources
- Backup and disaster recovery
- Multi-node clusters for high availability
- Horizontal and vertical scaling
- Monitoring and log management
- Audit logging
- Zero-downtime upgrades for multi-node clusters with replication
- Support for GPUs to optimize indexing (AWS only)
You have a broad choice of regions on AWS, Azure, and Google Cloud. For payment information, see Billing & Payments.
Create a Cluster
Prerequisite: Provide billing information before creating a standard cluster.
- Start in the Clusters section of the Cloud Dashboard.
- Select Clusters and then click + Create.
- In the Create a cluster screen, select Free or Standard. Most of the remaining options are only available for standard clusters.
- Select a provider and region, or a Hybrid Cloud environment.
- Choose the resources, using the table in Cluster Options.
- Review your cluster configuration and pricing, then select Create. It takes some time to provision your cluster.
Once provisioned, you can access your cluster on ports 443 and 6333 (REST) and 6334 (gRPC). See Connect to a Cluster.
Cluster Options
| Option | Choices and rules |
|---|---|
| Provider and region | AWS, GCP, Azure, or your own Hybrid Cloud environment. If you pay through a cloud marketplace, paid clusters are limited to that provider’s regions. |
| RAM per node | Set from your sizing estimate. Each node gets a disk with enough space for Qdrant’s default collection configuration. |
| vCPU per node | The available vCPU options depend on the RAM you choose. |
| GPU per node | AWS only, for higher RAM configurations. Speeds up indexing for write-heavy workloads. All GPU settings are configured automatically. See GPUs for Indexing. |
| Number of nodes | Every node has the same size. Use at least 3 for production. |
| High Availability | Sets up 3 or more nodes with replication, as described in Creating a Production-Ready Cluster. |
| Multi AZ (Premium) | Spreads nodes across availability zones. Chosen at creation only. Needs at least 3 nodes, and the node count must be a multiple of 3. Available in regions with at least 3 availability zones, at a higher rate to cover cross-zone traffic. |
| Additional disk | Add disk when your collection configuration needs more disk per GB of RAM, for example with the cold memory tier. You can add or remove disk later. |
| Disk speed tier (AWS) | Balanced for disks of 32 GiB or more, Performance for disks of 256 GiB or more. Faster tiers help write-heavy workloads and configurations with a low RAM cache ratio. |
Choose the Resources
Estimate the memory and disk for each collection and sum them. For dense vectors:
dense_size = points * replication_factor * dimensions * bytes_per_dim
bytes_per_dim is 4 for float32, 2 for float16, 1 for uint8, and 0.5 for turbo4. For example, one million 768-dimension float32 points with a replication factor of 2 need about 5.7 GB. On top of the vectors, budget for these:
- The HNSW index, which typically adds 20% to 30%.
- Payloads, payload indexes, and the ID tracker.
- About 20% headroom for the write-ahead log, snapshots, and temporary segments created by the optimizer.
Then size the nodes:
| Decision | Rule |
|---|---|
| RAM per node | Keep usage under 80% of the node’s RAM. Above that, the operating system evicts cached pages and latency increases sharply. |
| Number of nodes | ceil(total_ram_estimate / (node_ram * 0.8)), and at least 3 for production. A single node typically tops out around 100 million vectors, fewer for high-dimensional float32 data. |
| Shards per collection | At least 2 per node, so you can add nodes without resharding. 12 is a common choice because it divides evenly across 1, 2, 3, 4, 6, and 12 nodes. By default, a collection gets as many shards as the cluster has nodes. |
| Memory-saving options | Quantization reduces vector memory by a factor of 4 to 32. The cold memory tier reads vectors from memory-mapped files on demand instead of pre-loading them into RAM, so plan more disk per GB of RAM. |
Migrations need more room than normal operation. When you restore a snapshot or use the Migration Tool, the target cluster needs twice the disk space that the source collection uses, and twice the RAM when you use the Migration Tool.
These numbers are estimates. For the full formulas for sparse vectors, payloads, and indexes, see Capacity Planning. For ingestion-heavy workloads, it also covers CPU and disk speed. For an interactive estimate, use the Sizing Calculator, and for cost, use the Pricing Calculator.
Creating a Production-Ready Cluster
A production-ready cluster has at least 3 nodes, collections with a replication factor of at least 2, enough shards to grow, a backup schedule, and alerts that reach you. To get the High Availability setup, check the High Availability checkbox when you create the cluster. A cluster without these settings runs with a single replica of each shard, which gets none of the availability guarantees. See Setting Up a Resilient Qdrant Cluster.
The Cloud Production Checklist lists every setting with its value, where to set it, and why. For the reasoning behind the checklist, see Vector Search in Production.
Deleting a Cluster
You can delete a Qdrant database cluster from the cluster’s detail page.
Next Steps
- Create a Database API key to connect.
- Scale the cluster horizontally or vertically.
- Update the cluster when a new Qdrant version is available.
- Back up the cluster on a schedule.
Ready to try Qdrant?