# Google Compute Engine GPU VMs

Official pricing: https://cloud.google.com/products/compute/pricing/accelerator-optimized  
Category: gpu-cloud · Isolation: vm

## Pricing regimes (raw)

- **a2-highgpu-1g, on-demand (×1)** (resource): $0/vCPU-h, $0/GiB-h
- **a2-ultragpu-1g, on-demand (×1)** (resource): $0/vCPU-h, $0/GiB-h
- **a3-highgpu-8g, on-demand (×8 node)** (resource): $0/vCPU-h, $0/GiB-h
- **a3-megagpu-8g, on-demand (×8 node)** (resource): $0/vCPU-h, $0/GiB-h
- **a3-ultragpu-8g, on-demand (×8 node)** (resource): $0/vCPU-h, $0/GiB-h
- **g2-standard-4, on-demand (×1)** (resource): $0/vCPU-h, $0/GiB-h
- **a2-highgpu-1g, spot (×1)** (resource): $0/vCPU-h, $0/GiB-h [spot]
- **a2-ultragpu-1g, spot (×1)** (resource): $0/vCPU-h, $0/GiB-h [spot]

## Features

Yes: snapshots, fork/clone, pause/resume, persistent disk, volumes, ≥24 h sessions, custom image (Docker or snapshot), start from your own snapshot, your own Docker/OCI image, full VM (own kernel), Docker inside, nested virtualization, root, SSH, public IPv4, HTTPS preview URLs, egress allowlist, open internet, static egress IP, private networking, SOC 2, HIPAA, SSO, GPU, Windows, Python, HTTP method/path egress rules, EU data residency, snapshots on demand, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), extra volumes, shared volumes

No: memory snapshots, idle auto-stop, browser, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, self-hosting, BYOC, open source, arm64, macOS, Node.js, wake on request, live resize, live fork (no pause), memory fork, MCP server, automatic snapshots, agent harness API, hosted agent API (their own agent), secret proxy, secret proxy for any API

Unknown: everything else. Evidence (source + quote) per feature: https://battleships.dev/data/providers/gcp-compute-gpu.json → feature_evidence

## Caveats

- VM rate includes predefined CPU/RAM/GPU and bundled SSD if applicable. 60-second minimum then second billing. The selected default region was verified in HTML aria-selected/data-value: Iowa (us-central1). Rates are regional, not global. A3 1/2/4 GPU require Spot/Flex-start, not ordinary on-demand.
- Any null count, CPU/RAM, billing increment or minimum is unverified, not unlimited/free.
- This is GPU-product research, not a claim that every feature of the entire vendor documentation was audited. Unverified features are null.
- GPU multi-count bundles, variant/region selection and fractional slices require a SKU-aware estimator; unsupported rows remain unpriced in strict cards.
- Modes with gpu=null are intentionally excluded, not free GPU modes. Never substitute 0 for a null rate. Flags alt indicate DIY/ML infrastructure, not the same thing as managed untrusted-code sandboxes.
- Unknown fee, storage or bandwidth means full monthly total may be unavailable. Rates exclude taxes.
- a3-highgpu-1g Spot ($6.62/h) and DWS Flex-start ($4.79/h) are derived as 1/8 of the published a3-highgpu-8g rows (the 1g shape is exactly 1/8 of the 8g: 26 vCPU / 234 GB / 750 GiB / 1 H100). Flex-start VMs run at most 7 days and capacity may queue.

## How this provider charges

As of 2026-09-28. Accelerator-optimized A2/A3/G2 GPU VMs.

VM rate includes predefined CPU/RAM/GPU and bundled SSD if applicable. 60-second minimum then second billing. The selected default region was verified in HTML aria-selected/data-value: Iowa (us-central1). Rates are regional, not global. A3 1/2/4 GPU require Spot/Flex-start, not ordinary on-demand.

| Regime | When applicable | Billing | Numbers ($/GPU-hour unless slice) | Source |
|---|---|---|---|---|
| on-demand  | a2-highgpu-1g ; count [1]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 3.673385; list; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| on-demand  | a2-ultragpu-1g ; count [1]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 5.06879789; list; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| on-demand  | a3-highgpu-8g ; count [8]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 11.0612500149; list; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| on-demand  | a3-megagpu-8g ; count [8]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 11.6750891009; list; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| on-demand  | a3-ultragpu-8g ; count [8]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 10.6008635616; list; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| on-demand  | g2-standard-4 ; count [1]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 0.706832276; list; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| spot  | a2-highgpu-1g ; count [1]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 2.20401; dynamic-snapshot; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| spot  | a2-ultragpu-1g ; count [1]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 3.041237534; dynamic-snapshot; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| spot  | a3-highgpu-8g ; count [8]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 6.6202870685; dynamic-snapshot; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| spot  | a4-highgpu-8g ; count [8]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 4.9542; dynamic-snapshot; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |
| spot  | g2-standard-4 ; count [1]; us-central1 (Iowa), verified selected region in source HTML. | per-second; min 60s | 0.424056; dynamic-snapshot; per_gpu_hour | [source](https://cloud.google.com/products/compute/pricing/accelerator-optimized) |

## Gotchas
- VM rate includes predefined CPU/RAM/GPU and bundled SSD if applicable. 60-second minimum then second billing. The selected default region was verified in HTML aria-selected/data-value: Iowa (us-central1). Rates are regional, not global. A3 1/2/4 GPU require Spot/Flex-start, not ordinary on-demand.
- Any null count, CPU/RAM, billing increment or minimum is unverified, not unlimited/free.
- This is GPU-product research, not a claim that every feature of the entire vendor documentation was audited. Unverified features are null.
- GPU multi-count bundles, variant/region selection and fractional slices require a SKU-aware estimator; unsupported rows remain unpriced in strict cards.

## Worked example (required protocol workload)
4 vCPU / 8 GiB, 50 concurrent × 8 hours/day × 22 days = **8,800 instance-hours/month**. 30% CPU utilization does not reduce allocated GPU uptime. 50 GiB snapshots and 100 GiB egress are additional. The requested CPU-only workload has no GPU type/count, so its full GPU-provider total is **null**, not a fictitious CPU equivalent.

Explicit GPU sensitivity example: add **1 × A100-40GB** to each of those 50 workers at the a2-highgpu-1g quoted shape. Compute component = 8,800 × $3.673385 = **$32,325.7880**. CPU/RAM are included in that SKU (not independently resizable).
This is compute-only, before platform fees, storage, egress, setup, minimum rounding, cold-start/idle tails and capacity quotas. 50 concurrent GPUs are not promised by a unit rate.

## Evidence scope
Sources read are listed in the provider/features JSON. Raw pricing pages, relevant docs and API payloads are retained under `../raw/`. All unknown feature toggles remain null.