시작 시간과 기간(1~90일)을 지정해 TPU를 예약하려면 캘린더 모드에서 미래용 예약을 요청하세요. 동적 워크로드 스케줄러를 기반으로 하는 이 모드를 사용하면 최대 120일 전에 TPU 가용성을 확인하고 일정에 따라 용량을 요청할 수 있습니다. Compute Engine API 또는 Google Cloud 콘솔을 사용하여 캘린더 모드에서 미래용 예약을 요청할 수 있습니다. 승인되면 Compute Engine은 선택한 날짜와 시간에 예약을 자동으로 생성합니다. 그런 다음 Create Node API 또는 Queued Resources API를 사용하여 TPU를 프로비저닝하여 예약을 사용할 수 있습니다. Compute Engine은 예약 기간이 끝나면 예약과 예약을 사용하는 TPU를 자동으로 삭제합니다.
최대 90일간의 미래용 예약(캘린더 모드)을 요청하는 것은 정확한 시작 시간이 필요하고 기간이 정의된 학습 및 실험 워크로드에 적합합니다.
[[["이해하기 쉬움","easyToUnderstand","thumb-up"],["문제가 해결됨","solvedMyProblem","thumb-up"],["기타","otherUp","thumb-up"]],[["이해하기 어려움","hardToUnderstand","thumb-down"],["잘못된 정보 또는 샘플 코드","incorrectInformationOrSampleCode","thumb-down"],["필요한 정보/샘플이 없음","missingTheInformationSamplesINeed","thumb-down"],["번역 문제","translationIssue","thumb-down"],["기타","otherDown","thumb-down"]],["최종 업데이트: 2025-09-04(UTC)"],[],[],null,["# Request a future reservation for up to 90 days (in calendar mode)\n=================================================================\n\n|\n| **Preview**\n|\n|\n| This product or feature is subject to the \"Pre-GA Offerings Terms\" in the General Service Terms section\n| of the [Service Specific Terms](/terms/service-terms#1).\n|\n| Pre-GA products and features are available \"as is\" and might have limited support.\n|\n| For more information, see the\n| [launch stage descriptions](/products#product-launch-stages).\n\nTo reserve TPUs for a specific start time and duration, between 1 and 90 days,\nrequest a future reservation in calendar mode. This mode, powered by\nthe [Dynamic Workload\nScheduler](https://cloud.google.com/blog/products/compute/introducing-dynamic-workload-scheduler),\nlets you check TPU availability up to 120 days in advance and request capacity\nbased on your schedule. You can request a future reservation in calendar mode\nusing the Compute Engine API or Google Cloud console. If approved,\nCompute Engine automatically creates a reservation for your chosen date and\ntime. You can then [consume the reservation](/tpu/docs/consume-reservation) by\nprovisioning TPUs using the [Create Node API](/tpu/docs/managing-tpus-tpu-vm) or\nthe [Queued Resources API](/tpu/docs/queued-resources). Compute Engine\nautomatically deletes the reservation, and any TPUs that consume it, at the end\nof your reservation period.\n\nRequesting a future reservation for up to 90 days (in calendar mode) is a good\nfit for training and experimentation workloads that require precise start times\nand have a defined duration.\n\nFor more information, see [About future reservation requests in calendar\nmode](/compute/docs/instances/future-reservations-calendar-mode-overview).\n\nView future availability of TPUs\n--------------------------------\n\nYou can view the availability of TPUs 1 to 120 days in advance. For more\ninformation, see [View resources future\navailability](/compute/docs/instances/create-future-reservations-calendar-mode#view-availability).\n\nCreate a future reservation request in calendar mode\n----------------------------------------------------\n\nYou can request a future reservation in calendar mode using the Compute Engine\nAPI or Google Cloud console. For more information, see [Create a future reservation\nrequest in calendar\nmode](/compute/docs/instances/create-future-reservations-calendar-mode).\n\nLimitations\n-----------\n\nRequesting a future reservation in calendar mode for TPUs has the following\nlimitations:\n\n- You can't modify, cancel, or delete a future reservation request.\n- You must specify a reservation duration with a minimum of 1 day and a maximum of 90 days.\n- You can only reserve the following TPU versions in the specified zones:\n - **[TPU v6e](/tpu/docs/v6e)**: asia-northeast1-b, us-east5-a, us-east5-b\n - **[TPU v5p](/tpu/docs/v5p)**: us-east5-a\n - **[TPU v5e](/tpu/docs/v5e)** : us-west4-b (`BATCH`), us-central1-a (`SERVING`)\n\n| **Note:** For TPU v5e, you must select a workload type. For workloads that handle large amounts of data in single or multiple operations, such as machine learning (ML) training workloads, specify `BATCH`. For workloads that handle concurrent requests and require minimal network latency, such as ML inference workloads, specify `SERVING`.\n\nFor the allowed number of TPU chips per request, see\n[Limitations](/compute/docs/instances/create-future-reservations-calendar-mode#limitations)\nin the Compute Engine documentation.\n\nWhat's next\n-----------\n\n- [Learn about Cloud TPU reservations](/tpu/docs/about-tpu-reservations)\n- [Share a reservation between projects](/tpu/docs/share-reservation)\n- After your reservation start date, [consume the reservation](/tpu/docs/consume-reservation)"]]
