Supported Node Types
When launching new clusters, it is possible to specify the instance types to be used for the head node and for worker nodes launched by the autoscaler. The instance types vary depending on the cloud provider availability and nomenclature.
AWS
Anyscale currently supports all x86 instance types offered by AWS. In order to specify any AWS instance type, specify the AWS "instance size" (i.e., the name) as referenced in the AWS documentation. Note that availability of instance types may vary by region (refer to the AWS documentation for more details).
GCP
Predefined VMs
Anyscale supports all of the predefined VMs in the following families:
- N2 (general-purpose) (including N2 standard, N2 high-mem and N2 high-cpu)
- C2 (compute-optimized)
- A2 (accelerator-optimized)
Only nodes with 8 vCPUs or more are supported (this means that nodes in the N2 and C2 family should end with -8 or greater).
In order to specify any VM from the supported CPU families, specify the GCP "machine type" (i.e., the name) as referenced in the GCP documentation.
Custom VMs
Anyscale incorporates some custom VM configurations to add support for additional GPU accelerator types. These can be specified by using the Anyscale node type from the following table:
| Anyscale node type | GCP base type | vCPUs | Memory (GB) | GPU model | GPUs |
|---|---|---|---|---|---|
n1-highmem-4-nvidia-tesla-v100-1 | n1-highmem-4 | 4 | 26 | nvidia-tesla-v100 | 1 |
n1-highmem-8-nvidia-tesla-v100-1 | n1-highmem-8 | 8 | 52 | nvidia-tesla-v100 | 1 |
n1-highmem-32-nvidia-tesla-v100-4 | n1-highmem-32 | 32 | 208 | nvidia-tesla-v100 | 4 |
n1-highmem-64-nvidia-tesla-v100-8 | n1-highmem-64 | 64 | 416 | nvidia-tesla-v100 | 8 |
n1-highmem-96-nvidia-tesla-v100-8 | n1-highmem-96 | 96 | 624 | nvidia-tesla-v100 | 8 |
n1-standard-4-nvidia-tesla-t4-1 | n1-standard-4 | 4 | 15 | nvidia-tesla-t4 | 1 |
n1-standard-8-nvidia-tesla-t4-1 | n1-standard-8 | 8 | 30 | nvidia-tesla-t4 | 1 |
n1-standard-16-nvidia-tesla-t4-1 | n1-standard-16 | 16 | 60 | nvidia-tesla-t4 | 1 |