Monitor with tpu-info CLI

The tpu-info CLI is a tool for detecting Cloud TPU devices and reading runtime metrics from the libtpu library, including memory usage and duty cycle. It supports static, one-time snapshots, and live streaming to monitor metrics continuously.

Installation

Install the latest release using pip:

pip install tpu-info

Alternatively, install tpu-info from source:

pip install git+https://github.com/google/cloud-accelerator-diagnostics/#subdirectory=tpu_info

If you have already installed a version of tpu-info, make sure it is compatible with your environment and is not missing any metrics and features. For more information, see Missing features or metrics.

Access standard LibTPU metrics using the CLI

Use the following command to view the default tpu-info metrics with the CLI:

tpu-info

The output is similar to the following:

TPU Chips

| Chip         | Type         | Devices | PID    |
|--------------|--------------|---------|--------|
| /dev/vfio/0  | TPU v6e chip | 1       | 1022   |
| /dev/vfio/1  | TPU v6e chip | 1       | 1022   |
| /dev/vfio/2  | TPU v6e chip | 1       | 1022   |
| /dev/vfio/3  | TPU v6e chip | 1       | 1022   |

TPU Runtime Utilization

| Chip   | HBM Usage (GiB)          | Duty cycle |
|--------|--------------------------|------------|
| 8      | 17.26 GiB / 31.25 GiB    |    100.00% |
| 9      |  9.26 GiB / 31.25 GiB    |    100.00% |
| 12     |  9.26 GiB / 31.25 GiB    |    100.00% |
| 13     |  9.26 GiB / 31.25 GiB    |    100.00% |

TensorCore Utilization

| Core ID | TensorCore Utilization |
|---------|------------------------|
| 0       | 15.17%                 |
| 1       | 14.62%                 |
| 2       | 14.68%                 |
| 3       | 15.14%                 |

TPU Buffer Transfer Latency

| Buffer Size  | P50          | P90          | P95          | P999         |
|--------------|--------------|--------------|--------------|--------------|
| 8MB+         | 18264.03 us  | 33263.06 us  | 35990.98 us  | 53997.32 us  |

TPU Inbound Buffer Transfer Latency

| Buffer Size  | P50          | P90          | P95          | P999         |
|--------------|--------------|--------------|--------------|--------------|
| 8MB+         | 18316.95 us  | 32857.03 us  | 36501.59 us  | 58854.54 us  |

TPU Host Compute Latency

| Buffer Size  | P50          | P90          | P95          | P999         |
|--------------|--------------|--------------|--------------|--------------|
| 8MB+         | 678.33 us    | 2611.93 us   | 5258.30 us   | 11083.23 us  |

TPU gRPC TCP Minimum RTT

| P50      | P90      | P95      | P999     |
|----------|----------|----------|----------|
| 35.99 us | 52.15 us | 53.83 us | 55.51 us |

TPU gRPC TCP Delivery Rate

| P50           | P90           | P95           | P999          |
|---------------|---------------|---------------|---------------|
| 12305.96 Mbps | 18367.10 Mbps | 24872.11 Mbps | 44841.55 Mbps |

Usage

To view current TPU utilization data, tpu-info requires a running TPU workload with a supported ML framework, such as JAX or PyTorch/XLA. You can run the tpu-info command in your terminal with the following flags.

Process

Use the --process or -p flag to display information about the processes running on the TPU.

$ tpu-info --process

The output should look similar to the following:

TPU Process Info

| Chip        | PID    | Process Name |
|-------------|--------|--------------|
| /dev/vfio/0 | 799657 | python3      |
| /dev/vfio/1 | 799657 | python3      |
| /dev/vfio/2 | 799657 | python3      |
| /dev/vfio/3 | 799657 | python3      |
| /dev/vfio/4 | 799657 | python3      |
| /dev/vfio/5 | 799657 | python3      |
| /dev/vfio/6 | 799657 | python3      |
| /dev/vfio/7 | 799657 | python3      |

Metric

Use the --metric flag to display specific metrics. You can specify multiple metrics separated by spaces. Some common supported metrics are:

hbm_usage
duty_cycle_percent
tensorcore_utilization
buffer_transfer_latency
host_to_device_transfer_latency
device_to_host_transfer_latency
collective_e2e_latency

$ tpu-info --metric duty_cycle_percent hbm_usage

The output should look similar to the following:

TPU Duty Cycle

| Core ID | Duty Cycle (%) |
|---------|----------------|
| 0       | 100.00%        |
| 1       | 100.00%        |
| 2       | 100.00%        |
| 3       | 100.00%        |
| 4       | 100.00%        |
| 5       | 100.00%        |
| 6       | 100.00%        |
| 7       | 100.00%        |

TPU HBM Usage

| Chip   | HBM Usage (GiB)       |
|--------|-----------------------|
| 0      | 29.50 GiB / 31.25 GiB |
| 1      | 21.50 GiB / 31.25 GiB |
| 2      | 21.50 GiB / 31.25 GiB |
| 3      | 21.50 GiB / 31.25 GiB |
| 4      | 21.50 GiB / 31.25 GiB |
| 5      | 21.50 GiB / 31.25 GiB |
| 6      | 21.50 GiB / 31.25 GiB |
| 7      | 21.50 GiB / 31.25 GiB |

List metrics

Use the --list_metrics flag to display all supported metrics that can be requested with the --metric flag.

$ tpu-info --list_metrics

The output should look similar to the following:

╭─ Supported Metrics ─────────────────────────────────────────────────────────────────────────────╮
│         grpc_tcp_min_rtt                                                                        │
│         host_to_device_transfer_latency                                                         │
│         grpc_tcp_delivery_rate                                                                  │
│         inbound_buffer_transfer_latency                                                         │
│         host_compute_latency                                                                    │
│         buffer_transfer_latency                                                                 │
│         collective_e2e_latency                                                                  │
│         device_to_host_transfer_latency                                                         │
│         hbm_usage                                                                               │
│         duty_cycle_percent                                                                      │
│         tensorcore_utilization                                                                  │
╰─────────────────────────────────────────────────────────────────────────────────────────────────╯

Stream metrics

Streaming mode periodically refreshes and displays up-to-date utilization statistics. To stream the LibTPU metrics, add the --streaming flag to the tpu-info command. Use the --rate flag to control the cadence of streaming in seconds.

Use the following command to stream the default tpu-info metrics with the CLI:

# Refresh metrics every 2 seconds
tpu-info --streaming --rate 2

The output is similar to the following:

Refresh rate: 0.1s
Last update: 2025-07-24 11:00:59 UTC
Libtpu version: 0.0.19.dev20250721+nightly
Accelerator type: v6e

TPU Chips

| Chip         | Type         | Devices | PID    |
|--------------|--------------|---------|--------|
| /dev/vfio/0  | TPU v6e chip | 1       | 1022   |
| /dev/vfio/1  | TPU v6e chip | 1       | 1022   |
| /dev/vfio/2  | TPU v6e chip | 1       | 1022   |
| /dev/vfio/3  | TPU v6e chip | 1       | 1022   |

TPU Runtime Utilization

| Chip   | HBM Usage (GiB)          | Duty cycle |
|--------|--------------------------|------------|
| 8      | 17.26 GiB / 31.25 GiB    |    100.00% |
| 9      |  9.26 GiB / 31.25 GiB    |    100.00% |
| 12     |  9.26 GiB / 31.25 GiB    |    100.00% |
| 13     |  9.26 GiB / 31.25 GiB    |    100.00% |

TensorCore Utilization

| Core ID | TensorCore Utilization |
|---------|------------------------|
| 0       | 15.17%                 |
| 1       | 14.62%                 |
| 2       | 14.68%                 |
| 3       | 15.14%                 |

TPU Buffer Transfer Latency

| Buffer Size  | P50          | P90          | P95          | P999         |
|--------------|--------------|--------------|--------------|--------------|
| 8MB+         | 18264.03 us  | 33263.06 us  | 35990.98 us  | 53997.32 us  |

TPU Inbound Buffer Transfer Latency

| Buffer Size  | P50          | P90          | P95          | P999         |
|--------------|--------------|--------------|--------------|--------------|
| 8MB+         | 18316.95 us  | 32857.03 us  | 36501.59 us  | 58854.54 us  |

TPU Host Compute Latency

| Buffer Size  | P50          | P90          | P95          | P999         |
|--------------|--------------|--------------|--------------|--------------|
| 8MB+         | 678.33 us    | 2611.93 us   | 5258.30 us   | 11083.23 us  |

TPU gRPC TCP Minimum RTT

| P50      | P90      | P95      | P999     |
|----------|----------|----------|----------|
| 35.99 us | 52.15 us | 53.83 us | 55.51 us |

TPU gRPC TCP Delivery Rate

| P50           | P90           | P95           | P999          |
|---------------|---------------|---------------|---------------|
| 12305.96 Mbps | 18367.10 Mbps | 24872.11 Mbps | 44841.55 Mbps |

TPU-Z metrics

TPU-Z is a telemetry and debugging facility for TPUs. It provides detailed runtime status information for all TPU cores attached to a host. The functionality is provided through the tpuz module, which is part of the libtpu.sdk module in the libtpu Python SDK. The module provides a snapshot of each core's state.

The primary use case for TPU-Z is diagnosing hangs or deadlocks in distributed TPU workloads. You can query the TPU-Z service on hosts to capture the state of every core, comparing the Program Counters, HLO locations, and Run IDs across all cores to identify anomalies.

Use the following command to view TPU-Z metrics using the CLI:

tpu-info --metric core_state
tpu-info --metric sequencer_state
tpu-info --metric sequencer_state_detailed
tpu-info --metric queued_program

The output should include the core_state, sequencer_state, sequencer_state_detailed, and queued_programs tables.

Core State Information

The Core State Information (core_state) table provides information on the cores of a given chip. TPUs have either one or two cores per chip, depending on the generation.

Field	Description	Example values
Chip ID	The ID of the chip that the core belongs to.	`0`
Global Core ID	The unique ID of the core within the entire TPU system.	`1`
Core Type	The type of the TPU core.	`"TPU_CORE_TYPE_TENSOR_CORE"` `"TPU_CORE_TYPE_SPARSE_CORE"`
xdb Server Running	Indicates whether the Accelerator Debugger (XDB) server is running on a specific TPU core.	`True`

The output should look similar to the following table:

Core Information

| Chip ID | Global Core ID | Core Type                   | xdb Server |
|---------|----------------|-----------------------------|------------|
| 0       | 0              | TPU_CORE_TYPE_TENSOR_CORE   | True       |
| 0       | 1              | TPU_CORE_TYPE_SPARSE_CORE   | True       |
| 1       | 2              | TPU_CORE_TYPE_SPARSE_CORE   | False      |
| 1       | 3              | TPU_CORE_TYPE_SPARSE_CORE   | False      |
| 2       | 4              | TPU_CORE_TYPE_SPARSE_CORE   | True       |
| 2       | 5              | TPU_CORE_TYPE_SPARSE_CORE   | True       |

Sequencer State Information

The Sequencer State Information (sequencer_state) table provides information about a sequencer state on a core. A sequencer is a control unit within a TPU core responsible for fetching, decoding, and orchestrating the execution of instructions. There can be multiple sequencers for a single core.

Metric	Description	Example values
Chip ID	The ID of the chip that the core belongs to.	`0`
Global Core ID	The unique ID of the core within the entire TPU system.	`1`
Program Counter	The memory address of the instruction to be executed by the sequencer.	`15390`
Tracemark	The launch ID of the current or most recent program. This field is absent if not applicable.	`2147483647`
Program ID	The ID associated with a specific instance of a program being launched for execution on a TPU core.	`3230481660274331500`