--- title: Google Cloud TPU description: >- The benefits of Tensor Processing Units via scalable, user-friendly cloud resources for ML model development. breadcrumbs: Docs > Integrations > Google Cloud TPU --- > For the complete documentation index, see [llms.txt](https://docs.datadoghq.com/llms.txt). # Google Cloud TPU Integration version1.0.0 {% callout %} # Important note for users on the following Datadog sites: us2.ddog-gov.com {% alert level="info" %} To find out if this integration is available in your organization, see your [Datadog Integrations](https://app.datadoghq.com/integrations) page or ask your organization administrator. To initiate an exception request to enable this integration for your organization, email [support@ddog-gov.com](mailto:support@ddog-gov.com). {% /alert %} {% /callout %} ## Overview{% #overview %} Google Cloud TPU products make the benefits of Tensor Processing Units (TPUs) available through scalable and easy-to-use cloud computing resource for all ML researchers, ML engineers, developers, and data scientists running cutting-edge ML models. Use the Datadog Google Cloud Platform integration to collect metrics from Google Cloud TPU. ## Setup{% #setup %} ### Installation{% #installation %} To use Google Cloud TPU, you only need to set up the [Google Cloud Platform integration](https://docs.datadoghq.com/integrations/google-cloud-platform.md). ### Log collection{% #log-collection %} Google Cloud TPU logs are collected with Google Cloud Logging and sent to a Dataflow job through a Cloud Pub/Sub topic. If you haven't already, [set up logging with the Datadog Dataflow template](https://docs.datadoghq.com/integrations/google-cloud-platform.md#log-collection). Once this is done, export your Google Cloud TPU logs from Google Cloud Logging to the Pub/Sub topic: 1. Go to the [Google Cloud Logging page](https://console.cloud.google.com/logs/viewer) and filter the Google Cloud TPU logs. 1. Click **Create Export** and name the sink. 1. Choose "Cloud Pub/Sub" as the destination and select the Pub/Sub topic that was created for that purpose. **Note**: The Pub/Sub topic can be located in a different project. 1. Click **Create** and wait for the confirmation message to show up. ## Data Collected{% #data-collected %} ### Metrics{% #metrics %} | | | | | **gcp.tpu.cpu.utilization**(gauge) | Current CPU utilization on the TPU worker, represented as a percentage. Values are typically numbers between 0.0 and 100.0, but might exceed 100.0.*Shown as percent* | | **gcp.tpu.memory.usage**(gauge) | Memory usage in bytes.*Shown as byte* | | **gcp.tpu.network.received_bytes_count**(count) | Cumulative bytes of data this server has received over the network.*Shown as byte* | | **gcp.tpu.network.sent_bytes_count**(count) | Cumulative bytes of data this server has sent over the network.*Shown as byte* | | **gcp.tpu.accelerator.duty_cycle**(count) | Percentage of time over the sample period during which the accelerator was actively processing*Shown as percent* | | **gcp.tpu.instance.uptime_total**(count) | Elapsed time since the VM was started, in seconds.*Shown as second* | | **gcp.gke.node.accelerator.tensorcore_utilization**(count) | Current percentage of the Tensorcore that is utilized.*Shown as percent* | | **gcp.gke.node.accelerator.duty_cycle**(count) | Percent of time over the past sample period (10s) during which the accelerator was actively processing.*Shown as percent* | | **gcp.gke.node.accelerator.memory_used**(count) | Total accelerator memory allocated in bytes.*Shown as byte* | | **gcp.gke.node.accelerator.memory_total**(count) | Total accelerator memory in bytes.*Shown as byte* | | **gcp.gke.node.accelerator.memory_bandwidth_utilization**(count) | Current percentage of the accelerator memory bandwidth that is being used.*Shown as percent* | | **gcp.gke.container.accelerator.tensorcore_utilization**(count) | Current percentage of the Tensorcore that is utilized.*Shown as percent* | | **gcp.gke.container.accelerator.duty_cycle**(count) | Percent of time over the past sample period (10s) during which the accelerator was actively processing.*Shown as percent* | | **gcp.gke.container.accelerator.memory_used**(count) | Total accelerator memory allocated in bytes.*Shown as byte* | | **gcp.gke.container.accelerator.memory_total**(count) | Total accelerator memory in bytes.*Shown as byte* | | **gcp.gke.container.accelerator.memory_bandwidth_utilization**(count) | Current percentage of the accelerator memory bandwidth that is being used.*Shown as percent* | | **gcp.tpu.accelerator.memory_bandwidth_utilization**(gauge) | Current percentage of the accelerator memory bandwidth that is being used. Computed by dividing the memory bandwidth used over a sample period by the maximum supported bandwidth over the same sample period.*Shown as percent* | | **gcp.tpu.accelerator.memory_total**(gauge) | Total accelerator memory currently allocated in bytes.*Shown as byte* | | **gcp.tpu.accelerator.memory_used**(gauge) | Total accelerator memory currently used in bytes.*Shown as byte* | | **gcp.tpu.accelerator.tensorcore_utilization**(gauge) | Current percentage of the Tensorcore that is utilized. Computed by dividing the Tensorcore operations that were performed over a sample period by the supported number of Tensorcore operations over the same sample period.*Shown as percent* | | **gcp.tpu.multislice.accelerator.device_to_host_transfer_latencies.avg**(gauge) | Cumulative distribution of device to host transfer latency for each chunk of data. A latency starts when the request for data to be transferred to the host is issued and ends when an acknowledgement is received that the transfer of data has completed.*Shown as microsecond* | | **gcp.tpu.multislice.accelerator.host_to_device_transfer_latencies.avg**(gauge) | Cumulative distribution of host to device transfer latency for each chunk of data of multislice traffic. A latency starts when the request for data to be transferred to the device is issued and ends when an acknowledgement is received that the transfer of data has completed.*Shown as microsecond* | | **gcp.tpu.multislice.network.collective_end_to_end_latencies.avg**(gauge) | Cumulative distribution of end to end collective latency for multislice traffic. A latency starts when the request for the collective is issued and ends when an acknowledgement is received that the transfer of data has completed.*Shown as microsecond* | | **gcp.tpu.multislice.network.dcn_transfer_latencies.avg**(gauge) | Cumulative distribution of network-transfer latencies for multislice traffic. A latency starts when the request for data to be transferred over the DCN is issued and ends when an acknowledgement is received that the transfer of data has completed.*Shown as microsecond* | ### Events{% #events %} The Google Cloud TPU integration does not include any events. ### Service Checks{% #service-checks %} The Google Cloud TPU integration does not include any service checks. ## Troubleshooting{% #troubleshooting %} Need help? Contact [Datadog support](https://docs.datadoghq.com/help/).