Skip to main content

​
States & Lifecycle

States: Unverified → Verified → (potentially) Deverified → Unverified → … Lifecycle: Machines automatically move between these states based on performance and reliability factors. Once verified, a machine will remain verified unless issues arise, such as failing health checks or reliability standards, which could lead to deverification.

​
How It Works

Verification is entirely automated by proprietary algorithms that assess each machine’s operational health and performance, incorporating supply-and-demand dynamics. Only machines that meet the platform’s reliability and performance thresholds are eligible for verification. There is no manual intervention, ensuring consistency, scalability, and objectivity.
Top-tier AI GPUs are prioritized for verification because demand for them is highest. This includes datacenter GPUs such as B300, B200, H200, and H100, and dense premium builds such as RTX PRO 6000 (Server/WS/Max-Q), 8x RTX 5090, and 8x RTX 4090.

​
Host Responsibilities (Always)

  • Keep systems stable, well-cooled, and correctly configured.
  • Maintain compatible drivers/CUDA and dependable, symmetric networking.
  • Run jobs only through the Jobs tab or the create job CLI command.
  • When issues arise, fix them promptly-the automation will update status.

​
State Details & Guidance

​
Unverified

What it means: Newly added machines or machines under evaluation. The system hasn’t yet completed enough testing to confirm platform standards. This is not a judgment of quality-only that no platform guarantee exists yet. Do
  • Pass the Self-Test
  • Maintain steady uptime during evaluation.
  • Ensure drivers/CUDA and networking are correctly installed and reachable.
  • Keep the environment clean; schedule work via Create Job only.
Avoid
  • Unnecessary reboots or configuration changes.
  • Unrelated background workloads that consume GPU/CPU/IO.

​
Minimum Requirements for Verification

Your machine must meet all of the following.

​
GPU

​
CPU

On ARM64 (aarch64), CUDA 12.6 or newer is required. The CUDA 11.8 Self-Test image is built for x86_64 only, so Maxwell and Pascal GPUs cannot be verified on ARM64.

​
Memory

Formula: system RAM >= 0.95 x VRAM per GPU x number of GPUsExample: 8 x RTX PRO 6000 (96 GB) means the min system RAM >= 0.95 x 96 GB x 8 = 730 GB

​
Network

Renters connect directly to the machine over the WAN, so it needs a public IPv4 address with the port range forwarded to it. Machines behind CGNAT or a shared ISP IP cannot be used for hosting.

​
Operating system

Use a server edition. Desktop editions are not supported, and Ubuntu releases above 24.04 are not supported yet.
SSH password login must be disabled for security. SSH password login enabled will fail verification.

​
Storage

​
Reliability

Reliability starts low on a new machine and grows the longer the machine stays online. A stable machine typically reaches 90% within a few days.

​
Self-Test

You can check whether your machine is ready for verification by running a Self-Test. See How to Self-Test.
Dedicated machines only. Any personal workload, such as mining, gaming, running your own jobs, or using the machine as a desktop, will automatically fail verification.
Meeting these minimum requirements makes your machine eligible for verification, but does not automatically guarantee verification. Read More about Verification

​
Verified

What it means: The machine passed automated checks for reliability, network stability, operational health, and performance. A Verified machine consistently delivers server services to platform standards. Do
  • Monitor health (uptime, thermals, power) and respond to alerts.
  • Keep drivers/CUDA on compatible, latest stable versions.
  • Maintain stable, symmetric bandwidth.
Avoid
  • Downgrading hardware capacity (e.g., reducing GPU count, disk or RAM).
  • Allowing thermal, power, or bandwidth instability under load.

​
Deverified

What it means: A previously Verified machine no longer meets requirements. System continuous monitoring detects sustained degradation. When will deverification happen?
  • When the hosting software detects an error, your machine is automatically, but temporarily, deverified. It will appear as Unverified in search results until the underlying issue is resolved.
How should I begin fixing it?
  • A red error indicator will appear on your machine in the Machines tab. Use this message to identify and investigate the issue in your logs or metrics.
Recovery: Fix the issue and restore stability; the system will automatically transition back to Verified once system confirms healthy operation. This process may take some time, as the system ensures that the issue is fully resolved before restoring verification. Most error messages are cleared within 1-2 hours of resolution. Common causes
  • Network instability, closed ports, or low bandwidth.
  • Hardware/system errors (e.g., failing storage, insufficient PCIe bandwidth).
  • GPU issues (e.g., nvidia-smi/NVML failures, container device init errors).
  • Container launch failures or repeated runtime exceptions.
  • Detected abuse or policy violations.
Do
  • Investigate red error indicators quickly; review logs and metrics.
  • Validate thermal/power headroom and bandwidth under load.
  • Re-check health after changes to confirm resolution.
Avoid
  • Ignoring warnings or allowing instability to persist.
  • Reducing hardware below the created specification.
Note: If you’ve fixed the issue but the system doesn’t automatically detect the resolution, a Vast.ai team member may need to manually check that your machine is functioning correctly and clear the error.