# Hetzner GPU servers

Official pricing: https://www.hetzner.com/dedicated-rootserver/matrix-gpu/  
Category: gpu-cloud · Isolation: dedicated-host

## Pricing regimes (raw)

- **GEX45 Blackwell SFF 24GB, on-demand (×1)** (sizes): GEX45 ? vCPU/64 GiB $0.399/h [alt]
- **GEX131 Blackwell Max-Q 96GB, on-demand (×1)** (sizes): GEX131-1 ? vCPU/256 GiB $2.2419/h [alt]
- **GEX131-2 Blackwell Max-Q 96GB** (sizes): GEX131-2 ? vCPU/512 GiB $null/h
- **GEX131-3 Blackwell Max-Q 96GB** (sizes): GEX131-3 ? vCPU/768 GiB $null/h

## Features

Yes: persistent disk, volumes, ≥24 h sessions, full VM (own kernel), Docker inside, nested virtualization, root, SSH, public IPv4, HTTPS preview URLs, egress allowlist, open internet, static egress IP, private networking, GPU, Windows, EU data residency, gVisor or VM (no shared kernel), inbound access rules, firewall inside (nftables), extra volumes, shared volumes

No: snapshots, memory snapshots, fork/clone, pause/resume, idle auto-stop, start from your own snapshot, your own Docker/OCI image, browser, desktop GUI, computer-use API, browser + desktop control, anti-bot stealth, CAPTCHA solving, residential IPs, preinstalled agents, self-hosting, BYOC, open source, SOC 2, HIPAA, SSO, arm64, macOS, Python, Node.js, HTTP method/path egress rules, wake on request, live resize, live fork (no pause), memory fork, MCP server, automatic snapshots, snapshots on demand, 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/hetzner-gpu.json → feature_evidence

## Caveats

- Dedicated bare-metal GEX, separate from CPU Hetzner Cloud. Browser price widget was blank; official website-price-api resolves USD hourly/monthly/setup. One GPU per server; monthly cap is not an on-demand hourly discount. API lists more locations than marketing matrix; check configurator stock.
- 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.

## How this provider charges

As of 2026-09-28. GEX dedicated root servers, not Hetzner Cloud VMs.

Dedicated bare-metal GEX, separate from CPU Hetzner Cloud. Browser price widget was blank; official website-price-api resolves USD hourly/monthly/setup. One GPU per server; monthly cap is not an on-demand hourly discount. API lists more locations than marketing matrix; check configurator stock.

| Regime | When applicable | Billing | Numbers ($/GPU-hour unless slice) | Source |
|---|---|---|---|---|
| on-demand  | GEX45 Blackwell SFF 24GB; count [1]; API: HEL1/FSN1/NBG1, active; order availability may differ. Prices ex VAT. | per-hour; min unknowns | 0.399; list; per_gpu_hour | [source](https://website-price-api.hetzner.com/api/v1/products/ROBOT_1768%2BROBOT_1266) |
| on-demand  | GEX131 Blackwell Max-Q 96GB; count [1]; API: HEL1/FSN1/NBG1, active; order availability may differ. Prices ex VAT. | per-hour; min unknowns | 2.2419; list; per_gpu_hour | [source](https://website-price-api.hetzner.com/api/v1/products/ROBOT_1747%2BROBOT_1266) |

## Gotchas
- Dedicated bare-metal GEX, separate from CPU Hetzner Cloud. Browser price widget was blank; official website-price-api resolves USD hourly/monthly/setup. One GPU per server; monthly cap is not an on-demand hourly discount. API lists more locations than marketing matrix; check configurator stock.
- 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 × RTX-PRO-4000** to each of those 50 workers at the GEX45 quoted shape. Compute component = 8,800 × $0.399 = **$3,511.2000**. 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.