[[["易于理解","easyToUnderstand","thumb-up"],["解决了我的问题","solvedMyProblem","thumb-up"],["其他","otherUp","thumb-up"]],[["很难理解","hardToUnderstand","thumb-down"],["信息或示例代码不正确","incorrectInformationOrSampleCode","thumb-down"],["没有我需要的信息/示例","missingTheInformationSamplesINeed","thumb-down"],["翻译问题","translationIssue","thumb-down"],["其他","otherDown","thumb-down"]],["最后更新时间 (UTC):2025-08-11。"],[],[],null,["# Preemptible TPUs\n================\n\n| **Important:** TPU Spot VMs are the latest version of preemptible TPUs. New and existing preemptible TPUs continue to be supported, and TPU Spot VMs use the same pricing model as preemptible TPUs. However, TPU Spot VMs provide new features that preemptible TPUs don't support. For example, TPU Spot VMs don't have the 24 hour run time limit of preemptible TPUs. For more information, see [Manage TPU Spot VMs](/tpu/docs/spot).\n\nPreemptible TPUs cost much less than non-preemptible TPUs. The\nCloud TPU service might preempt (shut down) these TPUs at any\ntime, if it requires additional TPU resources.\n\nCreating a preemptible TPU VM\n-----------------------------\n\n### gcloud\n\n```bash\n$ gcloud compute tpus tpu-vm create demo-tpu \\\n --zone=europe-west4-a \\\n --accelerator-type=v3-8 \\\n --version=tpu-vm-tf-2.17.1-pjrt \\\n --preemptible\n \n```\n\nwhere:\n\n- `demo-tpu` is a name for the TPU.\n- `accelerator-type` is the type of TPU. For more information about supported accelerator types for each TPU version, see [TPU versions](/tpu/docs/system-architecture-tpu-vm#versions).\n- `version` is the [Cloud TPU software version](/tpu/docs/runtimes#tpu_software_versions) to install.\n- `preemptible` allows Cloud TPU to preempt the TPU.\n\nPricing and quota for preemptible TPUs\n--------------------------------------\n\nPricing for preemptible TPUs is significantly lower than for normal TPUs.\nFor details, see the [pricing page](/tpu/pricing). You are not\ncharged for TPUs if they are preempted in the first minute after you create\nthem.\n\nQuota for preemptible TPUs is generally higher, and is separate from the quota\nfor normal TPUs. See the [quota page](/tpu/docs/quota).\n\nDetecting if a TPU has been preempted\n-------------------------------------\n\nYou use the following `gcloud` command to check whether the\nCloud TPU service has preempted your TPU:\n\nList your available TPUs:\n\n\u003cbr /\u003e\n\n```bash\ngcloud compute tpus tpu-vm list\n --zone=us-central1-b\n \n```\n\n\u003cbr /\u003e\n\nThe output of the command displays the details of the TPUs created in your\nproject. If a\nTPU has been preempted, the status changes from `READY` to `PREEMPTED`.\n\nFor example: \n\n```bash\nNAME ZONE ACCELERATOR_TYPE NETWORK_ENDPOINT NETWORK RANGE STATUS\ndemo-tpu us-central1-b v2-8 10.240.1.2:8470 default 10.240.1.0/29 PREEMPTED\n```\n\n\u003cbr /\u003e\n\n| **Note:** If your Cloud TPU is preempted, you must delete it and create a new one as described in [Managing TPUs](/tpu/docs/managing-tpus-tpu-vm).\n\nDesigning your machine learning application to run on preemptible TPUs\n----------------------------------------------------------------------\n\nMake sure your application is resilient to restarts of the VM and TPU, by saving\nmodel checkpoints regularly and by configuring your application to restore the\nmost recent checkpoint on restart."]]