AI生成失败:AI返回 HTTP 403: {"error":{"message":"Access to model denied. Please make sure you are eligible for using the model.","id":"ed3d966a-0b21-42e5-9386-acf4788f4064","type":"AccessDenied.Unpurchased","code":"AccessDenied.Unpurchased"}}

闲置GPU算力回收 - 设备接入条件与参考报价 - 闲社
All systems operationalAI Agent Automation Platform
XIANSHE · SHARE IDLE COMPUTE

Connect your idle compute.
Put your hardware to work.

Have an idle GPU? Review the requirements and estimated earnings before authorizing a connection. Participation is subject to real task testing and the applicable task and settlement rules.

Estimated hourly earningsEligible task accountingYour account, your devices

This is a computing service, not a purchase of your GPU or computer. Being online does not mean you have orders. Earnings are not guaranteed.

My devices

My compute devices 0

Sign in to view your nodes

Node information and earnings are visible only to the owning account.

Sign in to Xianshe
Eligible task minutes accumulate and settle every 60 minutes

Eligible compute time accumulates across tasks on the same device, including remaining seconds. For example, 65 accumulated minutes settle as 60 minutes with 5 carried forward. The rate is fixed when the task is accepted. Online time, idle time, queues, model loading and failed tasks do not earn revenue.

Available only to single-GPU Ollama text-inference nodes that have passed device and time-tracking acceptance. H3 video, multi-GPU setups and Chinese domestic GPUs are not currently included. Settlements credit your site balance, not an automatic external payout. Existing balances remain; historical tasks are not retroactively credited with minutes.

Earnings

Review your balance before requesting a withdrawal

Sign in to view your earnings and request a withdrawal.

Share compute

Decide to share before authorizing the client

Windows client · 0.3.0

Understand what your device will do

After authorization, the client attempts to identify hardware, scan local models and bind a node. Continue only if you want to join the compute network.

  • Check VRAM, model files, disk space and network conditions first.
  • Providers are responsible for considering hardware load, electricity and network costs.
  • Installation, binding and heartbeats do not mean that you have tasks or earnings.
  • Resource limits, pausing and exit options depend on the features actually available in the client.

Version 0.3.0 accumulates eligible task minutes and does not require Python. Exit the old client before upgrading. Account settings are retained; devices must pass real-device acceptance before billing is enabled.

Connection steps

From connection to eligible tasks

01

Install the client

Download the current Windows installer and verify installation and startup on the target device.

02

Review authorization

Once you agree to bind the device, authorize it in your browser with your Xianshe account.

03

Check hardware & models

Check the detected hardware and models. Investigate missing information; registration alone does not establish task readiness.

04

Pass testing before taking tasks

Verify heartbeats and complete a real task test before participating under the enabled tasks and applicable rules. A heartbeat does not mean settlement has started.

IDLE COMPUTE · EARNING ESTIMATES

What could your device earn per hour?

Minute-based settlement · After node acceptance
Hourly estimates below are not guaranteed income

Unit: CNY per GPU per hour occupied by eligible tasks. Online time, idle time and queues do not earn revenue. Actual earnings depend on device checks, task volume, execution results and settlement rules.

Some models use internal trial estimates, subject to confirmation after testing. Accepted nodes accumulate eligible minutes at the rate fixed when a task is accepted. Every 60 minutes is credited automatically, with the remainder carried forward. Consider electricity, network costs and depreciation separately.

Estimated hourly earnings

RTX 2080 Ti · 11GB

¥0.70/ hour / GPU

Desktop hardware reference. Compatibility and task testing are required; a listed estimate does not mean task support is available.

Estimated hourly earnings

RTX A4000 · 16GB

¥0.74/ hour / GPU

Desktop hardware reference. Compatibility and task testing are required; a listed estimate does not mean task support is available.

Estimated hourly earnings

RTX 3080 Ti · 12GB

¥0.78/ hour / GPU

Desktop hardware reference. Compatibility and task testing are required; a listed estimate does not mean task support is available.

Estimated hourly earnings

RTX 3090 · 24GB

¥1.06/ hour / GPU

Desktop hardware reference. Compatibility and task testing are required; a listed estimate does not mean task support is available.

Estimated hourly earnings

RTX 4090 · 24GB

¥1.50/ hour / GPU

Desktop hardware reference. Compatibility and task testing are required; a listed estimate does not mean task support is available.

Estimated hourly earnings

RTX 5090 · 32GB

¥2.22/ hour / GPU

Desktop hardware reference. Compatibility and task testing are required; a listed estimate does not mean task support is available.

Other common GPUs · 17 estimates

Internal estimates for assessment before registration. Confirmation requires device and task testing. Orders are not guaranteed.

Device modelVRAMEstimated hourly earnings
RTX 306012GB¥0.48
RTX 3060 Ti8GB¥0.52
RTX 30708GB¥0.56
RTX 40608GB¥0.40
RTX 4060 Ti8GB¥0.52
RTX 4060 Ti16GB¥0.68
RTX 407012GB¥0.72
RTX 4070 SUPER12GB¥0.80
RTX 4070 Ti12GB¥0.88
RTX 4070 Ti SUPER16GB¥0.96
RTX 408016GB¥1.08
RTX 4080 SUPER16GB¥1.16
RTX 50608GB¥0.52
RTX 5060 Ti16GB¥0.80
RTX 507012GB¥0.88
RTX 5070 Ti16GB¥1.08
RTX 508016GB¥1.28
Professional GPUs & servers · 6 estimates

Professional GPUs, multi-GPU servers and Chinese domestic hardware require specific integration. A per-GPU estimate does not establish whole-system compatibility.

Device modelVRAMEstimated hourly earnings
V10032GB¥1.50
A80080GB¥4.47
RTX PRO 600096GB¥5.58
H80080GB¥7.98
H2096GB¥7.98
摩尔线程 S400048GB¥1.49
Other devices · 2 estimates

The configurations and conditions for these estimates remain unconfirmed. Actual task rates require separate confirmation.

Device modelVRAMEstimated hourly earnings
RTX 4090D24GBFrom ¥0.91
A100 SXM480GBFrom ¥5.58

Unlisted models, laptop GPUs, modified VRAM and power-limited devices require separate assessment. Equal VRAM does not mean equal speed; registration does not prove H3 compatibility. Account for electricity, networking and depreciation. Do not buy hardware solely on these earning estimates.

FOCUS · VIDEO GENERATION

MiniMax H3 compute

Task pipeline awaiting acceptance

Initial assessment focuses on RTX 4090 24GB and RTX 5090 32GB. Installing a model does not establish reliable video delivery. Test with the same model version, generation parameters and acceptance criteria.

Validate one task type first

The plan is to start with a roughly 5-second text-to-video task at fixed settings. Run one task per device before assessing speed, stability and cost.

RAM and storage matter too

The initial test configuration uses 64GB RAM and NVMe storage. Actual requirements vary by model version and offloading strategy. A qualifying GPU alone does not guarantee operation.

Understand the limitations

Local H3-Base 768p generation differs from the complete official 2K pipeline. Downloading the current client does not mean H3 automatic installation or video task acceptance is included.

Before commercial service, review the model license, regional applicability and content safeguards. Xianshe has not completed H3 production acceptance for the hardware combinations above.

Help

Questions before you start

Are you buying my computer or GPU?

No. You retain ownership. This is about computing services provided by your hardware. Review authorization, sharing controls and how to leave before connecting.

Will I earn the estimated amount immediately after installation?

No. Device and time-tracking acceptance come first. Eligible minutes from successful tasks settle to your site balance every 60 minutes; the remainder carries forward. Neither order volume nor fixed income is guaranteed.

Will customer data be sent to my device?

Tasks may require receiving relevant materials. Data authorization, access scope, transfer protection and cleanup rules should be defined before opening the service. Do not retain or redistribute task materials.

Do I need to share compute to use cloud models?

No. Use cloud models from the unified API center under the access and billing rules displayed there.

Why is my bound device not counted as online?

A binding record does not establish a live connection. Online counts require an enabled node, online status and a recent heartbeat. After timeout, the page shows it as offline.

Does Connected mean I can receive tasks?

Not necessarily. Model, scheduling and execution requirements must also be met. A connection status is not capability certification.

Has the client switched to the new unified API after the site upgrade?

The web entry now points to Xianshe AI, but the client retains its existing protocol. A client migration needs separate acceptance testing; changing page copy does not silently migrate it.

Just want cloud models?

No device sharing is required. Visit Xianshe AI to view models and usage prices.

Open Xianshe AI
Back to top