GPU clouds / Google Compute Engine GPU VMs alternative

A Google Compute Engine GPU VMs alternative for the surrounding agent

Google Compute Engine GPU VMs combine accelerator hardware with VM CPU and memory. Persistent boot disks retain system software, and supported disks can grow online. GPU-attached instances cannot use the general VM suspend operation, so disk continuity and saved execution state must be evaluated separately.

Start buildingCompare capabilities ↓

Reviewed 2026-10-19 · Published by the Sandbox0 team

When Google Compute Engine GPU VMs fits

Google GPU VMs fit workloads needing accelerator choice, a self-managed operating system, cloud networking and integration with Google’s data and training services. Persistent disks and existing online disk expansion are relevant strengths.

When to consider Sandbox0

Consider Sandbox0 for compatible Linux agents that orchestrate GPU jobs and should return to a complete durable writable RootFS while compute pauses. Keep GPU execution and its infrastructure requirements on a suitable accelerator service.

Capabilities

Google Compute Engine GPU VMs and Sandbox0: what persists?

Feature comparison with documentation and verification status
Environment and runtime lifetimeGoogle Compute Engine GPU VMsNot established in this reviewSandbox0ttl and hard_ttl default to 0 (disabled). Configure idle pause and deletion policies for the same persistent sandbox identity; an idle running sandbox still accrues compute charges.Read documentation →
Persistent filesGoogle Compute Engine GPU VMsAttached durable boot and data disks remain across stop/start; Local SSD handling is separate.View evidence ↗Sandbox0The writable RootFS survives successful checkpoints, including software, system configuration, dependencies, and user data.Read documentation →
Whole RootFSGoogle Compute Engine GPU VMsWhole boot-filesystem changes can persist on the retained durable disk, excluding separate ephemeral or in-memory mounts.View evidence ↗Sandbox0Persistence extends beyond /workspace to paths such as /usr/local, /etc, and /home. Runtime tmpfs and virtual filesystems are excluded.Read documentation →
Release compute, keep stateGoogle Compute Engine GPU VMsStop/start supports retained disk and fresh execution; GPU-attached instances cannot use RAM-preserving suspend.View evidence ↗Sandbox0Pause compute between tasks; resume on available capacity. Default resume starts fresh processes from the saved filesystem.Read documentation →
Memory and process stateGoogle Compute Engine GPU VMsCompute Engine explicitly does not support suspend for instances with attached GPUs.View evidence ↗Sandbox0Experimental memory checkpoints can preserve compatible process state. Use memory: true on both pause and resume on a compatible deployment. A filesystem checkpoint alone does not preserve running processes.Read documentation →

Online RootFS growth

On growth-enabled images, Sandbox0 automatically expands the writable RootFS as free space runs low, within a configured sparse logical ceiling up to 64 TiB. Agent processes keep running. Growth is enabled in the qualified Cloud region; confirm activation for other regions. Existing images and snapshots retain their captured storage geometry. The ceiling is logical capacity, not preallocated host disk.

Google Persistent Disk already supports explicit online enlargement up to 64 TiB for supported disk types, with image-specific filesystem expansion. Sandbox0’s distinction is automatic free-space-triggered growth of an enabled durable writable RootFS, with deployment and captured-geometry qualifications.

Files while compute is paused

Sandbox0's Files API reads persistent files without resuming compute. Filesystem-only pauses also permit edits; retained-memory pauses are read-only and reject mutations with 409. SSH, SFTP, file watching, and runtime-only mounts require running compute.

Retaining a stopped Google boot disk and attaching it elsewhere is a separate infrastructure workflow. The reviewed stop and snapshot mechanisms do not establish a direct whole-RootFS file API against the original stopped guest. Sandbox0 supports offline RootFS operations while its sandbox compute stays paused, with read-only access for retained-memory generations.

See paused file operations →

Persistent boot disks retain the installed system

A stopped VM retains its attached durable disks and configuration. Those boot disks can retain installed tools and system configuration beyond a workspace directory. Local SSD is a separate lifecycle: stopping can discard it or, for supported configurations, explicitly preserve it through a Preview path that uses billable durable storage. In-guest shutdown discards Local SSD unless data was saved separately.

Source 1 ↗

GPU instances cannot use general memory suspend

The suspend guide explicitly excludes instances with attached GPUs or TPUs. Ordinary stop does not preserve guest RAM or application execution state. Do not apply the general CPU VM’s memory-suspend behavior to a GPU VM, or treat a disk image as a GPU-memory checkpoint.

Source 1 ↗Source 2 ↗

Google already supports online disk enlargement

Persistent Disk can be enlarged while attached to a running VM, up to the supported disk limits, with partition and filesystem steps depending on the image. Public images can automatically grow the boot filesystem; custom images may require manual work. This is real online growth, distinct from an enabled RootFS automatically growing when free space runs low.

Source 1 ↗

Pricing

Compare billing models, then price your workload

CPU allocation, memory, billing minimums, disk, plan fees, and supported workload differ. These meters are not equal-performance machine quotes. Confirm region, resource units, and included allowances before estimating a total.

Google Compute Engine GPU VMs

Pricing reviewed 2026-10-19

Iowa a2-highgpu-1g on-demand reference
$3.673385 per VM-hour
Complete predefined shape: 1 A100, 12 vCPUs and 85 GiB memory; not a standalone GPU rate.
Rate source ↗
Iowa a2-ultragpu-1g on-demand reference
$5.06879789 per VM-hour
Complete A2 Ultra shape; different A100 memory and included resources.
Rate source ↗
Stopped retained disks and other resources
Storage and applicable attached-resource charges remain
Confirm reservations and commitments separately.
Rate source ↗

References use Iowa/us-central1 and the public table’s standard on-demand Price column, not Spot, DWS or committed-use discounts. Predefined accelerator-optimized VM rates include their specified CPU, memory, GPU and applicable bundled Local SSD. Boot/data storage, licenses, network and commitments can affect the invoice.

Check current Google Compute Engine GPU VMs pricing ↗

Sandbox0

$0.015/GiB-hour compute (billed by the second at approximately $0.000004167/GiB-second), plus $0.02/GiB-month of retained incremental RootFS and memory checkpoint data. Shared base images and platform recovery retention are excluded; paused sandboxes accrue storage charges only.

Compute / GiB-hour
$0.015
Retained data / GiB-month
$0.02
Paused compute
$0.00

A 4 GiB agent, 100 running hours

$6.00 compute

4 GiB × 100 hours, billed by the second. Keeping 20 GiB of billable incremental data for a full billing month adds $0.40. CPU is derived from memory by the platform; choose a tier that fits your workload.

Understand storage billing →

Before migrating from Google Compute Engine GPU VMs

  1. Record the full GPU machine shape, durable boot/data disks, Local SSD choices, quotas, licenses and any capacity commitment.
  2. Back up indispensable local data before stopping and migrate portable Linux agent stages while keeping GPU processing and required cloud services separate.
  3. Verify installed tools, image-specific disk growth, paused-file operations, fresh-process recovery and the complete workload bill.

Common questions

Is Sandbox0 a drop-in Google Compute Engine GPU VMs replacement?

Evaluate API compatibility and product scope. These products serve different core workflows, so a change may require a different architecture.

Does keeping files also keep a running agent?

Filesystem state and volatile process state are separate. Sandbox0 requires explicit compatible memory checkpoints to preserve running processes; its default filesystem resume starts fresh processes.

Sources and review scope

The Boat team's Battleships directory is the discovery source. References below describe the product and plan discussed on this page. Check current availability, regional terms and billing units before migrating. Where the cited material does not establish a capability or price, the comparison identifies that uncertainty.

Battleships research record ↗docs.cloud.google.com/compute/docs/instances/stop-start-instance ↗docs.cloud.google.com/compute/docs/instances/suspend-resume-instance ↗docs.cloud.google.com/compute/docs/instances/suspend-stop-reset-instances-overview ↗docs.cloud.google.com/compute/docs/disks/resize-persistent-disk ↗cloud.google.com/products/compute/pricing/accelerator-optimized ↗Sandbox0 RootFS capacity →

Related comparisons

Azure GPU VMs (NC_A100_v4 reference) vs Sandbox0 →Cerebrium Serverless Compute vs Sandbox0 →Crusoe vs Sandbox0 →DigitalOcean GPU Droplets vs Sandbox0 →Fluidstack vs Sandbox0 →Hetzner GPU servers vs Sandbox0 →