# Hardware and hosting tiers

Match the workload to the machine and distinguish discovery from qualification.

AmpleRun supports two hosting paths. Choose by the workload your machine can actually serve, not just by the GPU vendor printed on its case.

| Tier | Workload and operating model |
| --- | --- |
| **Container / Tier A** | Linux-based GPU workloads using a compatible container runtime and published template |
| **Inference / Tier B** | Managed model serving through a host runtime and an outbound relay, including the Apple silicon path |

The public installer accepts `--tier container` or `--tier inference`. On a Mac, the supported installer path is inference; Linux is required for container hosting. Automatic tier selection is a convenience, not proof that a requested workload qualifies.

## Check compatibility in layers

1. **Discovery:** can the agent identify the hardware, memory, driver, and runtime capabilities?
2. **Preflight:** does the local environment meet the selected tier's requirements?
3. **Qualification:** does the machine pass the service's required checks?
4. **Template fit:** is this exact workload compatible with the admitted machine?
5. **Readiness:** does the workload actually start and answer its readiness checks for this rental?

These stages answer different questions. A GPU appearing in telemetry does not imply that arbitrary CUDA, ROCm, Intel, Vulkan, or Metal workloads will run on it. A runtime compatibility label is not a real-hardware test result.

## Prepare the machine

For Linux containers, verify the GPU driver, Docker, and the relevant GPU container runtime before offering capacity. For managed inference, inspect the installer's runtime and memory checks. Disk space, stable connectivity, thermals, and power settings matter alongside device memory.

Shared/unified memory and discrete GPU VRAM are not interchangeable capacity promises. Use the template's actual requirements and returned fit result. Do not publish a listing based only on a nominal device name.

## What renters should inspect

Read the offered GPU count, usable resources, backend, tier, access type, and availability class. Tier B model serving is not a general-purpose root container. Request SSH only where the chosen template and tier provide it.

Start with [host setup](https://docs.amplerun.com/start/host.md) and use the returned diagnostics to resolve incompatibilities. If a check blocks admission, [contact support](https://docs.amplerun.com/support.md) with redacted details rather than weakening it.

---
Last reviewed: 2026-10-19.
Canonical: https://docs.amplerun.com/hosting/hardware-and-tiers/