SPCS GEN_ARM_G1 and GEN_X64_G1 instances are now GA on Google Cloud

The GEN_ARM_G1 and GEN_X64_G1 instance families for Snowpark Container Services (SPCS) on Google Cloud are now Generally Available (GA). Both families offer general-purpose compute across five sizes, ranging from 1 to 28 vCPUs.

Use cases

Imagine you need to run microservices or job containers compiled for ARM architecture on Google Cloud inside Snowflake, or maintain existing x86-based applications.

With this update, you can:

• Run linux/arm64 container images natively using the GEN_ARM_G1 family.
• Keep legacy x86 containers running on GEN_X64_G1 instances.
• Scale compute pools by choosing options ranging from 1 to 28 vCPUs per node.

How to use

To use these instances, specify the family name and size in the INSTANCE_FAMILY parameter when creating a compute pool.

CREATE COMPUTE POOL meu_pool_arm
  MIN_NODES = 1
  MAX_NODES = 3
  INSTANCE_FAMILY = GEN_ARM_G1_2
  AUTO_RESUME = TRUE;

What happens in this code:

1. Creates a compute pool named meu_pool_arm.
2. Sets scaling limits between a minimum of 1 node and a maximum of 3 nodes.
3. Specifies the ARM instance family size 2 (GEN_ARM_G1_2).
4. Enables auto-resume when a service or job is submitted to the pool.

Limitations and requirements

• Features available in GA only for Snowflake accounts hosted on Google Cloud (GCP).
• The executing role requires the CREATE COMPUTE POOL privilege at the account level.
• x86_64 container images will not run on ARM nodes and vice versa. Ensure container architecture compatibility with the selected family.
• The GEN_X64_G1 family is a previous generation. For new x86 workloads, Snowflake recommends using the CPU_X64 instance family instead.