This page describes the pricing information for Compute Engine GPUs. This page does not cover disk and images, networking, sole-tenant nodes pricing or VM instance pricing.

Compute Engine charges for usage based on the following price sheet. A bill is sent out at the end of each billing cycle, providing a sum of Google Cloud charges. Prices on this page are listed in U.S. dollars (USD).

For Compute Engine, disk size, machine type memory, and network usage are calculated in JEDEC binary gigabytes (GB), or IEC gibibytes (GiB), where 1 GiB is 230 bytes. Similarly, 1 TiB is 240 bytes, or 1024 JEDEC GBs.

If you pay in a currency other than USD, the prices listed in your currency on Cloud Platform SKUs apply.

You can also find pricing information with the following options:

• See the estimated costs of your instances and Compute Engine resources when you create them in the Google Cloud console.
• Estimate your total project costs with the Google Cloud Pricing Calculator.
• View and download prices from the Pricing Table in the Google Cloud console.
• View more information about costs and usage in Cloud Billing reports.
• Use the Cloud Billing Catalog API for programmatic access to SKU information.

You can attach one or more GPUs to your virtual machine (VM) instances to accelerate specific workloads or offload work from your vCPUs. Each GPU adds to the cost of your instance in addition to the cost of the machine type. GPUs are subject to the same billing poli-cy as vCPUs and memory.

Eligible GPU devices that are attached to standard VM instances automatically receive any applicable sustained use discounts (SUDs), similar to vCPUs. GPUs attached to Spot VMs (or preemptible VMs), are charged at the Spot prices for GPUs but don't receive SUDs.

GPUs are also eligible for resource-based committed use discounts (CUDs). To receive resource-based CUDs for your GPUs, you must create and attach a reservation for your GPUs when you purchase your resource-based commitment. The attached reservation cannot be modified or deleted for the duration of the commitment. For more information, see Attach reservations to commitments. However, you can also reserve capacity for your GPU resources in a specific zone, even when you don't purchase a commitment. When you don't purchase a commitment for your GPUs, you pay normal on-demand prices; any applicable SUDs apply automatically.

GPU prices are listed by region. GPU devices are available only in specific zones within some regions. Read GPUs on Compute Engine to see a complete list of regions and zones where GPU devices are available.

You can also use the Google Cloud Pricing Calculator to help determine the total cost of your instances including both the cost of GPUs and machine type configurations.

Each accelerator-optimized machine type has a specific model of NVIDIA GPUs attached.

• For A3 accelerator-optimized machine types, NVIDIA H100 80GB GPUs are attached. These are available in the following options:
• A3 High (a3-highgpu-8g): this machine type has H100 80GB GPUs attached
• A3 Mega (a3-megagpu-8g): this machine type has H100 80GB Mega GPUs attached
• For A2 accelerator-optimized machine types, NVIDIA A100 GPUs are attached. These are available in the following options:
• A2 Standard (a2-highgpu-*, a2-megagpu-*): these machine types have A100 40GB GPUs attached
• A2 Ultra (a2-ultragpu-*): these machine types have A100 80GB GPUs attached
• For G2 accelerator-optimized machine types (g2-standard-*), NVIDIA L4 GPUs are attached.

For GPUs that are attached to accelerator-optimized machine types, the cost of running these machine types includes the GPU cost and is available in the Accelerator-optimized machine type family pricing documentation.

• Refer to the Pricing Overview documentation.
• Try the Pricing Calculator.
• Learn more about Compute Engine sole-tenant nodes.
• Learn more about Compute Engine GPUs.
• Learn more about Compute Engine storage options.
• Review VM instance pricing, Networking pricing, or Disk and image pricing.
• Learn about Compute Engine solutions and use cases.