Snowpark Container Services: AWS instance families¶
Instance families are grouped as current generation or previous generation. AWS instance families are available in three types:
- General Compute (GEN): Recommended for general-purpose containerized workloads.
- High Memory (MEM): High memory-to-vCPU ratio for applications that require large amounts of RAM, such as CPU-based model serving, large-scale in-memory data processing, and vector index serving.
- GPU Accelerated (GPU): For machine learning training, inference, and AI workloads requiring GPU acceleration.
For pricing information, see the Snowflake Service Consumption Table.
Note
Instance family availability varies by region and can change over time. To list the instance families available in the region of your current account, run SHOW COMPUTE POOL INSTANCE FAMILIES. If an instance family you need isn’t listed, contact your Snowflake account team.
The Node limit column shows the default maximum number of nodes of each instance family that an account can use. An account can also have at most 750 active or idle nodes in total, across all instance families. For more information, see Guidelines and limitations.
Current generation instance families¶
General compute instance families¶
Current generation instances recommended for general-purpose workloads.
x86 general compute¶
Current generation x86 instances for general-purpose workloads.
| Instance Family | vCPU | Memory (GiB) | Storage (GB) | Bandwidth limit (Gbps) | Node limit |
|---|---|---|---|---|---|
| GEN_X64_G2_2 | 1 | 6 | 100 | 12.5 | 500 |
| GEN_X64_G2_4 | 3 | 13 | 100 | 12.5 | 500 |
| GEN_X64_G2_8 | 6 | 28 | 100 | 12.5 | 500 |
| GEN_X64_G2_32 | 28 | 116 | 100 | 12.5 | 500 |
ARM general compute¶
Current generation ARM instances for general-purpose workloads.
| Instance Family | vCPU | Memory (GiB) | Storage (GB) | Bandwidth limit (Gbps) | Node limit |
|---|---|---|---|---|---|
| GEN_ARM_G1_2 | 1 | 6 | 100 | 12.5 | 150 |
| GEN_ARM_G1_4 | 3 | 13 | 100 | 12.5 | 150 |
| GEN_ARM_G1_8 | 6 | 28 | 100 | 15.0 | 150 |
| GEN_ARM_G1_16 | 14 | 58 | 100 | 15.0 | 150 |
| GEN_ARM_G1_32 | 28 | 116 | 100 | 15.0 | 150 |
Note
ARM instances require container images built for the linux/arm64 platform.
x86 instances require linux/amd64. For multi-architecture deployments, use
multi-platform container images.
High memory instance families¶
Current generation x86 instances optimized for memory-intensive workloads.
| Instance Family | vCPU | Memory (GiB) | Storage (GB) | Bandwidth limit (Gbps) | Node limit |
|---|---|---|---|---|---|
| MEM_X64_G2_8 | 6 | 58 | 100 | 12.5 | 500 |
| MEM_X64_G2_32 | 28 | 240 | 100 | 12.5 | 500 |
| MEM_X64_G2_64 | 60 | 492 | 100 | 25.0 | 500 |
| MEM_X64_G2_192 | 188 | 1436 | 100 | 50.0 | 500 |
GPU accelerated instance families¶
AWS GPU instance families feature four NVIDIA GPU architectures, each suited to different AI and ML workloads.
NVIDIA A10G¶
Mid-range Ampere GPU for ML model development and inference on small to medium models.
| Instance Family | vCPU | Memory (GiB) | NVMe Storage (GB) | Bandwidth limit (Gbps) | GPU | GPU Memory per Instance (GB) | Node limit |
|---|---|---|---|---|---|---|---|
| GPU_NV_S | 6 | 28 | 450 | 10.0 | 1 NVIDIA A10G | 24 | 150 |
| GPU_NV_M | 44 | 178 | 3800 | 40.0 | 4 NVIDIA A10G | 96 | 10 |
NVIDIA A100¶
High-throughput Ampere GPU for large-scale model training and large dataset processing.
| Instance Family | vCPU | Memory (GiB) | NVMe Storage (GB) | Bandwidth limit (Gbps) | GPU | GPU Memory per Instance (GB) | Node limit |
|---|---|---|---|---|---|---|---|
| GPU_NV_L | 92 | 1112 | 100 | 400.0 | 8 NVIDIA A100 | 320 | 10 |
NVIDIA L40S¶
Ada Lovelace GPU optimized for GenAI inference and fine-tuning.
| Instance Family | vCPU | Memory (GiB) | NVMe Storage (GB) | Bandwidth limit (Gbps) | GPU | GPU Memory per Instance (GB) | Node limit |
|---|---|---|---|---|---|---|---|
| GPU_L40S_G1_8 | 6 | 58 | 450 | 25.0 | 1 NVIDIA L40S | 48 | 5 |
| GPU_L40S_G1_16 | 14 | 116 | 600 | 25.0 | 1 NVIDIA L40S | 48 | 5 |
| GPU_L40S_G1_48 | 44 | 368 | 3800 | 100.0 | 4 NVIDIA L40S | 192 | On Request |
| GPU_L40S_G1_192 | 188 | 1436 | 3800 | 400.0 | 8 NVIDIA L40S | 384 | On Request |
NVIDIA RTX PRO 6000¶
Blackwell GPU with 96GB VRAM per GPU for large-scale inference and data-intensive workloads.
| Instance Family | vCPU | Memory (GiB) | NVMe Storage (GB) | Bandwidth limit (Gbps) | GPU | GPU Memory per Instance (GB) | Node limit |
|---|---|---|---|---|---|---|---|
| GPU_R6K_G1_8 | 6 | 58 | 1900 | 50.0 | 1 NVIDIA RTX PRO 6000 | 96 | 5 |
| GPU_R6K_G1_16 | 14 | 116 | 1900 | 50.0 | 1 NVIDIA RTX PRO 6000 | 96 | 5 |
| GPU_R6K_G1_32 | 28 | 240 | 1900 | 100.0 | 1 NVIDIA RTX PRO 6000 | 96 | 5 |
| GPU_R6K_G1_48 | 44 | 490 | 3800 | 400.0 | 2 NVIDIA RTX PRO 6000 | 192 | On Request |
| GPU_R6K_G1_96 | 92 | 984 | 7600 | 800.0 | 4 NVIDIA RTX PRO 6000 | 384 | On Request |
| GPU_R6K_G1_192 | 188 | 1843 | 11400 | 1600.0 | 8 NVIDIA RTX PRO 6000 | 768 | On Request |
Previous generation instance families¶
General compute instance families¶
Previous generation x86 instances. For new workloads, use GEN_X64_G2 instances instead.
| Instance Family | vCPU | Memory (GiB) | Storage (GB) | Bandwidth limit (Gbps) | Node limit |
|---|---|---|---|---|---|
| CPU_X64_XS | 1 | 6 | 100 | 12.5 | 500 |
| CPU_X64_S | 3 | 13 | 100 | 12.5 | 500 |
| CPU_X64_M | 6 | 28 | 100 | 12.5 | 500 |
| CPU_X64_SL | 14 | 58 | 100 | 12.5 | 500 |
| CPU_X64_L | 28 | 116 | 100 | 12.5 | 500 |
High memory instance families¶
Previous generation x86 instances optimized for memory. For new workloads, use MEM_X64_G2 instances instead.
| Instance Family | vCPU | Memory (GiB) | Storage (GB) | Bandwidth limit (Gbps) | Node limit |
|---|---|---|---|---|---|
| HIGHMEM_X64_S | 6 | 58 | 100 | 12.5 | 500 |
| HIGHMEM_X64_M | 28 | 240 | 100 | 12.5 | 500 |
| HIGHMEM_X64_L | 124 | 984 | 100 | 50.0 | 500 |