Resource Based Pricing
Customize your hardware configurations with à la carte pricing
CoreWeave's entire infrastructure is purpose-built for compute-intensive workloads, and everything from our servers to our storage and networking solutions are designed to deliver best-in-class performance that are up to 35x faster and 80% less expensive than generalized public Clouds.
Hardware selection is significantly flexible, allowing for custom configurations of CPU, GPU, RAM, and storage requests when scheduling workloads. Resources are scheduled using provided configurations, granting both savings and simplicity on top of legacy Cloud instance type selection.
While CoreWeave allows scheduling and provides pricing based on standard instances, all CoreWeave Cloud instances are highly configurable. All billing is à la carte; priced by the hour, and billed by the minute.
All billing is based on the greater of resources requested in an instance. If the instance is configured to be burstable, billing is based on the actual resources consumed during any minute billing window.
The following components are configurable in GPU based instances.
Description | Label | Cost per hour | VRAM |
|---|---|---|---|
H100 HGX (80GB) | H100_NVLINK_80GB | $4.78 | N/A |
H100 for PCIe (80GB) | H100_PCIE | $4.25 | N/A |
A100 HGX (80GB) | A100_NVLINK_80GB | $2.21 | N/A |
A100 HGX (40GB) | A100_NVLINK | $2.06 | 40GB HBM2e |
A100 for PCIe (40GB) | A100_PCIE_40GB | $2.06 | 40GB HBM2e |
A100 for PCIe (80GB) | A100_PCIE_80GB | $2.21 | N/A |
V100 for NVLINK | Tesla_V100_NVLINK | $0.80 | 16GB HBM2 |
A40 | A40 | $1.28 | 48GB GDDR6 |
RTX A6000 | RTX_A6000 | $1.28 | 48GB GDDR6 |
RTX A5000 | RTX_A5000 | $0.77 | 24GB GDDR6 |
RTX A4000 | RTX_A4000 | $0.61 | 16GB GDDR6 |
Quadro RTX 5000 | Quadro_RTX_5000 | $0.57 | 16GB GDDR6 |
Quadro RTX 4000 | Quadro_RTX_4000 | $0.24 | 8GB GDDR6 |
The following CPU components are also configurable in GPU based instances. For CPU-only instances, see CPU-only instance pricing.
Node type | Label | Cost per hour |
|---|---|---|
AMD Epyc Milan vCPU | amd-epyc-milan | $0.010 |
AMD Epyc Rome vCPU | amd-epyc-rome | $0.010 |
Intel Xeon Scalable | intel-xeon-scalable | $0.010 |
Intel Xeon v4 | intel-xeon-v4 | $0.010 |
Instances without a GPU attached are configurable as combinations of vCPU and system RAM.
For these instances, system RAM is included in the vCPU price. Combinations can be configured in multiples of the following.
CPU Type | Label | RAM per vCPU | Cost per vCPU per Hour |
|---|---|---|---|
AMD Epyc Milan | amd-epyc-milan | 4GB | $0.035 |
AMD Epyc Rome | amd-epyc-rome | 4GB | $0.030 |
Intel Xeon Scalable | intel-xeon-scalable | 4GB | $0.030 |
Intel Xeon v4 | intel-xeon-v4 | 4GB | $0.020 |
Intel Xeon v3 | intel-xeon-v3 | 4GB | $0.0125 |
RAM may be specified as an additional component on its own.
Component | Label | Cost per hour |
|---|---|---|
System RAM per GB | memory | $0.005 |
Below is an example configuration provided in an excerpt of a YAML manifest, in which 6 AMD Epyc Rome vCPUs with 24GB of RAM are requested.
Additional Resources
In this example, the node type is chosen using an
affinity. For more information on using affinities to select hardware, see Node Types and Advanced Label Selectors. containers:
- name: epyc-example
resources:
requests:
cpu: 6
memory: 24Gi
limits:
cpu: 6
memory: 24Gi
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: node.coreweave.cloud/cpu
operator: In
values:
- amd-epyc-rome
In this example, the requested resources are:
CPU amount | Instance type | Unit cost per hour | System RAM |
|---|---|---|---|
6 | AMD Epyc Rome | $0.03 | 24Gi |
The total per hour cost of this instance would be $0.03 x 6, equaling $0.18 per hour.
In this example, a hardware configuration of
4 Tesla V100 NVLINK GPUs with 32 Intel Xeon Scalable vCPU and 128GB of RAM is requested: containers:
- name: v100-example
resources:
requests:
cpu: 32
memory: 128Gi
nvidia.com/gpu: 4
limits:
cpu: 32
memory: 128Gi
nvidia.com/gpu: 4
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: gpu.nvidia.com/class
operator: In
values:
- Tesla_V100_NVLINK
In this example, the requested resources are:
Instance type | Instance count | Unit cost per hour |
|---|---|---|
Tesla V100 NVLINK | 4 | $0.80 |
vCPU count | vCPU cost per hour |
|---|---|
32 | $0.01 |
RAM amount | RAM cost per hour |
|---|---|
128Gi | $0.005 |
The per hour cost of these resources equals
($0.80 x 4) + ($0.01 x 32) + ($0.005 x 128), or $4.16 per hour.IP Addresses are billed at $4.00 per IP per month.
For periods of use less than one month, this charge is prorated by the minute just as all other billing. If a public IP address is assigned to an instance but the instance is not running, billing will continue to accrue for this reserved Public IP Address.
All CoreWeave Cloud billing periods cover the calendar month.
For example, a billing period that begins on
1 January 12:00am UTC ends on 1 February 12:00am UTC.