gcloud alpha compute tpus tpu-vm create - create a new Cloud TPU VM node

gcloud alpha compute tpus tpu-vm create(TPU :--zone=ZONE)--version=VERSION[--async][--autocheckpoint-enabled][--boot-disk=[KEY=VALUE,…]][--data-disk=[mode=MODE],[source=SOURCE]][--description=DESCRIPTION][--image=IMAGE][--internal-ips][--labels=[KEY=VALUE,…]][--metadata=[KEY=VALUE,…]][--metadata-from-file=[KEY=VALUE,…]][--network=NETWORK; default="default"][--preemptible][--provisioning-model=PROVISIONING_MODEL][--queue-count=QUEUE_COUNT][--range=RANGE][--reservation=RESERVATION][--reserved][--scopes=[SCOPES,…]][--service-account=SERVICE_ACCOUNT][--shielded-secure-boot][--spot][--subnetwork=SUBNETWORK][--tags=[TAGS,…]][--accelerator-type=ACCELERATOR_TYPE; default="v2-8"    |--topology=TOPOLOGY--type=TYPE][GCLOUD_WIDE_FLAG …]

gcloudalphacomputetpustpu-vmcreatemy-tpu

gcloudalphacomputetpustpu-vmcreatemy-tpu--zone=us-central1-a--network=my-tf-network--description='My TPU VM'--version='v2-alpha'

To create a small TPU VM v2 pod, run:

gcloudalphacomputetpustpu-vmcreatemy-tpu--zone=us-central1-a--accelerator-type='v2-32'--description='My TPU VM'--version='v2-alpha'

Tpu resource - Name of the Cloud TPU VM node to create. The arguments in thisgroup can be used to specify the attributes of this resource. (NOTE) Someattributes are not given arguments in this group but can be set in other ways.

To set theproject attribute:

• provide the argumenttpu on the command line with a fully specifiedname;
• provide the argument--project on the command line;
• set the propertycore/project.

This must be specified.

TPU

ID of the tpu or fully qualified identifier for the tpu.

To set thetpu attribute:

• provide the argumenttpu on the command line.

This positional argument must be specified if any of the other arguments in thisgroup are specified.

--zone=ZONE

Zone of the Cloud TPU.

If not specified, will usedefault compute/zone.To set thezone attribute:

• provide the argumenttpu on the command line with a fully specifiedname;
• provide the argument--zone on the command line;
• set the propertycompute/zone.

--version=VERSION

Runtime version for the TPU, such as2.3.

For a list of available versions run:

gcloudalpha compute tpus tpu-vm versions list

--async

Return immediately, without waiting for the operation in progress to complete.

--autocheckpoint-enabled

If specified, the TPU node is created with the Autocheckpoint feature enabled.

--boot-disk=[KEY=VALUE,…]

Specifies the boot disk configuration.

gcloudalphacomputetpustpu-vmcreate--boot-diskconfidential-compute=True,kms-key=<full_kms_key_name_here>

The following keys are allowed:

confidential-compute

Create the boot disk in confidential compute mode. CMEK layer is required.

kms-key

Specifies the fully qualified Cloud KMS cryptokey name which will be used toprotect the disk. KMS cryptokey name format:projects/<kms-project>/locations/<kms-location>/keyRings/<kms-keyring>/cryptoKeys/<key-name>

--data-disk=[mode=MODE],[source=SOURCE]

Additional data disks for the TPU VM.

This flag must be repeated to provide multiple data disks. For example:

gcloudalphacomputetpustpu-vmcreate--data-disksource=projects/my-project/zones/us-central1-c/disks/my-disk,mode=read-only

The following keys are allowed:

source

Specifies the full path to an existing disk. Required. The disk must be in thesame zone.

mode

Specifies the mode in which to attach this disk. Valid options are 'read-write','read-only'. If not specified, the default is 'read-write'.

--description=DESCRIPTION

Text description of the TPU.

--image=IMAGE

Specifies the full URI of the machine image to use for creating the TPU VM'sboot disk. If specified, this will override the boot image that would normallybe used by the specified--runtime-version.

--internal-ips

Indicate that the IP addresses for the node should be internal. The default isthat external IP addresses will be associated with the TPU workers.

--labels=[KEY=VALUE,…]

Resource labels to represent user-provided metadata. Seehttps://cloud.google.com/compute/docs/labeling-resourcesfor details.

--metadata=[KEY=VALUE,…]

List of comma-separated metadata key-value pairs for the Cloud TPU VM node.Example:--metadata='key1=value1,key2=value2'

--metadata-from-file=[KEY=VALUE,…]

Same as--metadata except the value for the entry will be read froma local file. Example:--metadata-from-file='key1=value1.txt'

--network=NETWORK; default="default"

Network that this TPU will be a part of.

--preemptible

If provided, the TPU will be preemptible and time-limited. It may be preemptedto free up resources for standard TPUs, and will only be able to run for alimited amount of time.

Preemptible TPUs cannot be restarted.

--provisioning-model=PROVISIONING_MODEL

The provisioning model of the TPU node. One of [standard, spot,reservation-bound]. Setting--provisioning-model=spot isfunctionally equivalent to setting--spot.

--queue-count=QUEUE_COUNT

Specifies the networking queue count for TPU VM instances. Both Rx and Tx queueswill be set to this number. If it's not specified, a default queue count will beassigned. For Virtio-net, each interface will get min(floor(#vCPU / #vNIC), 32)queues. For gVNIC, each interface will get min(floor(#vCPU / #vNIC / 2), 16)queues.

--range=RANGE

CIDR Range for the TPU.

The IP range that the TPU will select an IP address from. Must be in CIDRnotation and a/29 range, for example192.168.0.0/29.Errors will occur if the CIDR range has already been used for a currentlyexisting TPU, the CIDR range conflicts with any networks in the user's providednetwork, or the provided network is peered with another network that is usingthat CIDR range.

--reservation=RESERVATION

The name of the reservation. This can either be the full name or just the name.However, a full name is required if the reservation is not in the defaultproject.

Full reservation name format is:

projects/<project>/reservations/<reservation-name>

--reserved

When specified, will attempt to create the TPU node under reservations made inthe current project. The reservations can be made separately but used inaggregated form. i.e., the user can make a reservation of 128 V2 TPUs and lateron make another reservation of 128 V2 TPUs then creates a v2-256 TPU instance.If there exists no reservation or not sufficient amount of reserved cores underthe project, the request will fail due to lack of capacity.

--scopes=[SCOPES,…]

List of comma-separated scopes to be made available for the service account.

--service-account=SERVICE_ACCOUNT

Email address of the service account. If empty, default Google Compute Engineservice account will be used.

--shielded-secure-boot

Specifies that the TPU instances are created with secure boot enabled. Thisimplicitly makes them Shielded VM instances.

--spot

If specified, create this VM as a spot VM. Spot VMs make unused capacityavailable at highly discounted rates. Spot VMs may be preempted at any time ifthe capacity is needed, but unless preempted there is no limit on runtimeduration. Spot VM TPUs cannot be restarted, and must be recreated again.

--subnetwork=SUBNETWORK

Subnetwork that this TPU will be a part of.

--tags=[TAGS,…]

Tags to apply to the TPU Node. Tags are used to identify valid sources ortargets for network firewalls. Seehttps://cloud.google.com/vpc/docs/add-remove-network-tagsfor more details.

Specify a TPU node in either chips or cores

At most one of these can be specified:

Core based TPU configuration

--accelerator-type=ACCELERATOR_TYPE; default="v2-8"

TPU accelerator type for the TPU. If not specified, this defaults tov2-8.

For a list of available accelerator types run:

gcloudalpha compute tpus tpu-vm accelerator-types list

Chip based TPU configuration

--topology=TOPOLOGY

Chip topology for TPU.

This flag argument must be specified if any of the other arguments in this groupare specified.

--type=TYPE

Type of TPU.TYPE must be one of:v2,v3,v4,v5litepod,v5p,v6e.

This flag argument must be specified if any of the other arguments in this groupare specified.

Run$gcloud help for details.