Specify the machine configuration for a pipeline step
Stay organized with collections
Save and categorize content based on your preferences.
Kubeflow pipeline components are factory functions that create pipeline
steps. Each component describes the inputs, outputs, and implementation of the
component. For example, train_op is a component in the following code sample.
For example, a training component could take a CSV file as an input and use it
to train a model. By setting the machine type parameters on the pipeline step,
you can manage the requirements of each step in your pipeline. If you have two
training steps and one step trains on a huge data file and the second step
trains on a small data file, you can allocate more memory and CPU to the first
task, and fewer resources to the second task.
CPU_LIMIT: The maximum CPU limit for this operator. This string
value can be a number (integer value for number of CPUs), or a number
followed by "m", which means 1/1000. You can specify at most 96 CPUs.
MEMORY_LIMIT: The maximum memory limit for this operator. This
string value can be a number, or a number followed by
"K" (kilobyte), "M" (megabyte), or "G" (gigabyte).
At most 624GB is supported.
SELECTOR_CONSTRAINT: Each constraint is a key-value pair label.
For the container to be eligible to run on a node, the node must have each
constraint as a label. For example:
'cloud.google.com/gke-accelerator', 'NVIDIA_TESLA_T4'
ACCELERATOR_LIMIT: The accelerator (GPU or TPU) limit for the
operator. You can specify a positive integer. For more information about the
available GPUs and how to configure them, see
GPUs. For more information
about the available TPUs and how to configure them, see
TPUs.
CustomJob supports specific machine types that limit you to a maximum of 96 CPUs and 624GB of memory. Based on the CPU, memory, and accelerator configuration that you specify, Vertex AI Pipelines automatically selects the closest match from the supported machine types.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-08-25 UTC."],[],[],null,["# Specify the machine configuration for a pipeline step\n\nKubeflow pipeline components are factory functions that create pipeline\nsteps. Each component describes the inputs, outputs, and implementation of the\ncomponent. For example, `train_op` is a component in the following code sample.\n\nFor example, a training component could take a CSV file as an input and use it\nto train a model. By setting the machine type parameters on the pipeline step,\nyou can manage the requirements of each step in your pipeline. If you have two\ntraining steps and one step trains on a huge data file and the second step\ntrains on a small data file, you can allocate more memory and CPU to the first\ntask, and fewer resources to the second task.\n\nBy default, the component will run on as a Vertex AI\n[`CustomJob`](/vertex-ai/docs/reference/rest/v1/projects.locations.customJobs)\nusing an **e2-standard-4** machine, with 4 core CPUs and 16GB memory. For more information about selecting one of the Google Cloud-specific machine resources listed in [Machine types](/vertex-ai/docs/training/configure-compute#machine-types), see [Request Google Cloud machine resources with Vertex AI Pipelines](/vertex-ai/docs/pipelines/request-gcp-machine-resources).\n\nThe following sample shows you how to set CPU, memory, and GPU configuration\nsettings for a step:\n**Note:** If you want to specify the disk space in the machine configuration, you must [create a custom training job from a component by requesting Google Cloud machine resources](/vertex-ai/docs/pipelines/request-gcp-machine-resources#create-customtrainingjob-from-component) instead. \n\n from kfp import dsl\n\n @dsl.pipeline(name='custom-container-pipeline')\n def pipeline():\n generate = generate_op()\n train = (\n train_op(\n training_data=generate.outputs['training_data'],\n test_data=generate.outputs['test_data'],\n config_file=generate.outputs['config_file'])\n .set_cpu_limit('\u003cvar translate=\"no\"\u003eCPU_LIMIT\u003c/var\u003e')\n .set_memory_limit('\u003cvar translate=\"no\"\u003eMEMORY_LIMIT\u003c/var\u003e')\n .add_node_selector_constraint(\u003cvar translate=\"no\"\u003e\u003cspan class=\"devsite-syntax-n\"\u003eSELECTOR_CONSTRAINT\u003c/span\u003e\u003c/var\u003e)\n .set_accelerator_limit(\u003cvar translate=\"no\"\u003e\u003cspan class=\"devsite-syntax-n\"\u003eACCELERATOR_LIMIT\u003c/span\u003e\u003c/var\u003e))\n\nReplace the following:\n\n- \u003cvar translate=\"no\"\u003eCPU_LIMIT\u003c/var\u003e: The maximum CPU limit for this operator. This string\n value can be a number (integer value for number of CPUs), or a number\n followed by \"m\", which means 1/1000. You can specify at most 96 CPUs.\n\n- \u003cvar translate=\"no\"\u003eMEMORY_LIMIT\u003c/var\u003e: The maximum memory limit for this operator. This\n string value can be a number, or a number followed by\n \"K\" (kilobyte), \"M\" (megabyte), or \"G\" (gigabyte).\n At most 624GB is supported.\n\n | **Note:** Vertex AI Pipelines does not support calling [`set_memory_request`](https://kubeflow-pipelines.readthedocs.io/page/source/dsl.html#kfp.dsl.PipelineTask.set_memory_request) on an operator ; you must use [`set_memory_limit`](https://kubeflow-pipelines.readthedocs.io/page/source/dsl.html#kfp.dsl.PipelineTask.set_memory_request) to request a specific memory amount.\n- \u003cvar translate=\"no\"\u003eSELECTOR_CONSTRAINT\u003c/var\u003e: Each constraint is a key-value pair label.\n For the container to be eligible to run on a node, the node must have each\n constraint as a label. For example:\n `'cloud.google.com/gke-accelerator', 'NVIDIA_TESLA_T4'`\n\n The following constraints are available:\n - `NVIDIA_GB200`^+^ (includes [GPUDirect-RDMA](/vertex-ai/docs/training/configure-compute#gpudirect-rdma))\n - `NVIDIA_B200`^\\*^ (includes [GPUDirect-RDMA](/vertex-ai/docs/training/configure-compute#gpudirect-rdma))\n - `NVIDIA_H100_MEGA_80GB`^\\*^ (includes [GPUDirect-TCPXO](/vertex-ai/docs/training/configure-compute#gpudirect-tcpxo))\n - `NVIDIA_H100_80GB`\n - `NVIDIA_H200_141GB`^\\*^ (includes [GPUDirect-RDMA](/vertex-ai/docs/training/configure-compute#gpudirect-rdma))\n - `NVIDIA_A100_80GB`\n - `NVIDIA_TESLA_A100` (NVIDIA A100 40GB)\n - `NVIDIA_TESLA_P4`\n - `NVIDIA_TESLA_P100`\n - `NVIDIA_TESLA_T4`\n - `NVIDIA_TESLA_V100`\n - `NVIDIA_L4`\n\n \u003cbr /\u003e\n\n - `TPU_V2`\n - `TPU_V3`\n- \u003cvar translate=\"no\"\u003eACCELERATOR_LIMIT\u003c/var\u003e: The accelerator (GPU or TPU) limit for the\n operator. You can specify a positive integer. For more information about the\n available GPUs and how to configure them, see\n [GPUs](/vertex-ai/docs/training/configure-compute#specifying_gpus). For more information\n about the available TPUs and how to configure them, see\n [TPUs](/vertex-ai/docs/training/configure-compute#tpu).\n\n | **Note:** TPUs are only available in specific locations. For more information, see [Using accelerators](/vertex-ai/docs/general/locations#accelerators).\n\n`CustomJob` supports specific machine types that limit you to a maximum of 96 CPUs and 624GB of memory. Based on the CPU, memory, and accelerator configuration that you specify, Vertex AI Pipelines automatically selects the closest match from the supported machine types."]]
