Advanced Label Selectors
Label selectors are available for precise placement of workloads
Selecting the right hardware for your workload is important. All compute nodes are tagged with a set of labels specifying the hardware type. Affinity Rules should be leveraged on workloads to ensure that the desired type of hardware (ie. GPU model) gets assigned to your job.
The basic node selectors demonstrated in Node Types are usually all that is needed to properly schedule workloads. Please contact support for any questions about advanced scheduling or special requirements.
Label | Possible Values | Description |
---|---|---|
node.coreweave.cloud/cpu | The CPU family of the CPU in the node | |
ethernet.coreweave.cloud/speed | 10G, 40G, 100G | The uplink speed from the node to the backbone |
gpu.nvidia.com/count | 4-8 | Number of GPUs provisioned in the node. Using this selector is not recommended as the GPU resource requests are the correct method of selecting GPU count requirement |
gpu.nvidia.com/class | GPU model provisioned in the node | |
gpu.nvidia.com/vram | GPU VRAM in Gigabytes on the GPUs provisioned in the node | |
gpu.nvidia.com/nvlink | true, false | Denotes if GPUs are interconnected with NVLink. Currently applicable only for Tesla_V100 |
topology.kubernetes.io/region | ORD1, LAS1, LGA1 | The region the node is placed in |
Schedule in ORD1 only
16 Core Xeon CPU with 10GE
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: topology.kubernetes.io/region
operator: In
values:
- ORD1
spec:
containers:
- name: example
resources:
limits:
cpu: 16
memory: 48Gi
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: node.coreweave.cloud/cpu
operator: In
values:
- intel-xeon-v4
- key: ethernet.coreweave.cloud/speed
operator: In
values:
- 10G
Last modified 6mo ago