Droplet and hardware prices
Hosted sizes, specialized memory and CPU, GPUs, dedicated hosts, and discounts.
Hosted compute prices depend on the selected hardware class and running allocation. Open Compute → Droplets → Create droplet → Size to inspect the region's available configurations and estimated cost. Dedicated hosts are under Fleet → Hosted capacity → Dedicated hosts.
Shared CPU
Shared sizes suit development, small services, and workloads that tolerate CPU contention. Memory is configured per droplet; shared vCPU is not a dedicated physical-core reservation.
Shared CPU sizes have these memory allocations and compute rates:
| Size | vCPU | Memory | Hourly | 730-hour month |
|---|---|---|---|---|
s-1-1 | 1 | 1 GiB | $0.004110 | $3.00 |
s-1-2 | 1 | 2 GiB | $0.006849 | $5.00 |
s-2-4 | 2 | 4 GiB | $0.013699 | $10.00 |
s-4-8 | 4 | 8 GiB | $0.027397 | $20.00 |
s-8-16 | 8 | 16 GiB | $0.054795 | $40.00 |
s-16-32 | 16 | 32 GiB | $0.109589 | $80.00 |
Each shared droplet can request up to 20 GiB of included ephemeral scratch space. Scratch is not a persistent root disk and is not allocated unless requested.
Dedicated general purpose
Dedicated allocation follows the pool's CPU and sibling-thread policy. A vCPU is the guest's scheduled processor unit; it is not a claim that every CPU class has identical instruction throughput.
Dedicated general-purpose sizes have these allocations and compute rates:
| Size | vCPU | Memory | Hourly | 730-hour month |
|---|---|---|---|---|
d-2-8 | 2 | 8 GiB | $0.027397 | $20.00 |
d-4-16 | 4 | 16 GiB | $0.054795 | $40.00 |
d-8-32 | 8 | 32 GiB | $0.109589 | $80.00 |
d-16-64 | 16 | 64 GiB | $0.219178 | $160.00 |
d-32-128 | 32 | 128 GiB | $0.438356 | $320.00 |
d-64-256 | 64 | 256 GiB | $0.876712 | $640.00 |
Memory and CPU configurations
Memory-optimized configurations provide 8 GiB per vCPU.
Memory-optimized sizes have these compute rates:
| Size | vCPU | Memory | Hourly | 730-hour month |
|---|---|---|---|---|
m-2-16 | 2 | 16 GiB | $0.038356 | $28.00 |
m-4-32 | 4 | 32 GiB | $0.076712 | $56.00 |
m-8-64 | 8 | 64 GiB | $0.153425 | $112.00 |
m-16-128 | 16 | 128 GiB | $0.306849 | $224.00 |
m-32-256 | 32 | 256 GiB | $0.613699 | $448.00 |
CPU-optimized configurations provide 1 GiB per dedicated vCPU.
CPU-optimized sizes have these compute rates:
| Size | vCPU | Memory | Hourly | 730-hour month |
|---|---|---|---|---|
c-4-4 | 4 | 4 GiB | $0.032877 | $24.00 |
c-8-8 | 8 | 8 GiB | $0.065753 | $48.00 |
c-16-16 | 16 | 16 GiB | $0.131507 | $96.00 |
c-32-32 | 32 | 32 GiB | $0.263014 | $192.00 |
c-64-64 | 64 | 64 GiB | $0.526027 | $384.00 |
These dedicated classes can request up to 40 GiB of included ephemeral scratch space. Additional scratch is $0.025/GiB per 730-hour month while allocated to running compute. It is discarded under the ephemeral-disk lifecycle.
Custom sizes and hardware policy
A region can offer intermediate supported sizes. General-purpose dedicated CPU is $4 per vCPU plus $1.50 per GiB RAM per 730-hour month, matching the published dedicated sizes. General-purpose custom sizes require at least 4 GiB RAM per vCPU; the separate memory and CPU families follow their own size ratios. A custom request must satisfy the pool's increments, minimum memory, image, and physical admission limits.
CPU pinning, supported NUMA policy, huge-page backing, and migration-compatible CPU profiles have no separate feature fee. They can reduce destination availability and require overlap capacity during migration. They do not make an unsupported custom size admissible.
GPU droplets
GPU rates include the listed host CPU and memory and up to 100 GiB of requested ephemeral scratch. Persistent disks and any separately licensed guest software remain additional.
GPU configurations include these accelerators and host resources:
| Accelerator | GPU memory | Host vCPU / RAM | Hourly | 730-hour month |
|---|---|---|---|---|
| 1 × NVIDIA L4 | 24 GiB | 8 / 32 GiB | $0.35 | $255.50 |
| 1 × NVIDIA A10 | 24 GiB | 8 / 32 GiB | $0.40 | $292.00 |
| 1 × NVIDIA L40S | 48 GiB | 16 / 64 GiB | $0.70 | $511.00 |
| 1 × NVIDIA A100 | 80 GiB | 16 / 128 GiB | $1.10 | $803.00 |
| 1 × NVIDIA H100 | 80 GiB | 24 / 128 GiB | $1.80 | $1,314.00 |
| 2 × NVIDIA H100 | 160 GiB total | 48 / 256 GiB | $3.60 | $2,628.00 |
Availability depends on the region. GPU models are distinct classes, not an automatic upgrade path at a common rate. Passthrough restrictions apply to ordinary live migration.
Dedicated physical hosts
A host reservation bills for the whole assigned server. Guests created within that paid host do not also incur ordinary droplet compute charges. Attached managed storage and external traffic remain separate.
Whole-host reservations provide these physical resources:
| Host | Physical cores / threads | RAM | Local NVMe | 730-hour month |
|---|---|---|---|---|
metal-8 | 8 / 16 | 32 GiB | 2 × 480 GB | $59 |
metal-16 | 16 / 32 | 64 GiB | 2 × 960 GB | $99 |
metal-24 | 24 / 48 | 128 GiB | 2 × 1.92 TB | $149 |
metal-32 | 32 / 64 | 256 GiB | 2 × 1.92 TB | $249 |
metal-64 | 64 / 128 | 512 GiB | 2 × 3.84 TB | $449 |
Local device capacities use manufacturer decimal units. Local disks and one public IPv4 are included for the host reservation. Their data is not automatically replicated. Stopping guests does not release the whole-host reservation; explicitly release it after evacuation to stop that meter.
Spot and commitments
When offered, shared and dedicated CPU spot allocations cost 40% of the listed compute rate, a 60% discount. GPU spot costs 50% of its listed compute rate. Spot is preemptible; persistent resources retain their normal rates after preemption. The creation review shows the selected capacity type and rate before allocation.
Eligible dedicated CPU, GPU, and whole-host commitments receive 20% off for 12 months or 35% off for 36 months. They do not stack with spot or discount unrelated storage and traffic. Subscription annual billing is a separate discount.
An idle hardware capacity hold is billed at the applicable held-compute rate. Consumed capacity is billed once, through its reservation or guest allocation, not both. A pricing commitment alone is not a physical hold. See reserved capacity.
Disks, addresses, and operating systems
Persistent root and data disks are charged by the selected storage class; they are not hidden inside the compute rate. The default does not attach scratch or public addresses unless requested. The first public IPv4 assigned to a running droplet is included; extra or retained idle addresses have their own rate.
Public Linux catalog images have no image license fee. Guest OS licenses, commercial drivers, and proprietary application licenses are not included unless the selected licensed image explicitly shows that charge. A workload supplying its own approved license uses the ordinary hardware rate.
See storage and networking, metering rules, and worked estimates.