How does ComputeHorde ensure miners actually perform the computation they claim to have done?

ComputeHorde employs a dual-validation approach: validators use a separate unregistered 'Trusted Miner' to pre-run part of validation tasks and establish expected results, then compare miner submissions against these ground-truth outputs. Additionally, validators can require miners to deposit collateral, enabling economic slashing for incorrect results. This combination of cryptographic verification and financial incentives prevents malicious behavior.

Can I use ComputeHorde if I don't run my own GPU hardware?

Yes. Validators do not need to own GPUs themselves—ComputeHorde is designed specifically for validators to access mining-provided GPU resources. A validator only needs a standard server running the validator software and a separate trusted miner instance (which does require an A6000 GPU for cross-validation). The bulk of computation happens on miner hardware across the network.

What GPU hardware types does ComputeHorde currently support?

ComputeHorde currently supports A6000 GPUs as the primary hardware class. A100 support is planned for future releases. The subnet is designed to eventually support all GPU types and configurations required by validators across the broader Bittensor ecosystem.

How are miner rewards calculated if multiple validators use the same miner?

Rewards are calculated stake-proportionally per miner–validator pair. Each validator receives an allowance of executor-seconds proportional to their stake. When a validator submits a job, they consume allowance blocks equal to the runtime. All validators score all miners' work, and final rewards are determined by the total compute time delivered, adjusted by hardware class weights and validator stake distribution.

What happens if a miner reduces their executor count mid-operation?

If a miner lowers their executor count, any allowance blocks tied to that reduction are invalidated and cannot be used by validators. This prevents validators from overspending or relying on capacity that no longer exists.

Can I protect my miner from DDoS attacks while mining on ComputeHorde?

Yes. ComputeHorde provides an optional DDoS shield that miners can activate by running a Docker container from the project's DDoS Shield repository. This tool reduces vulnerability to denial-of-service attacks and helps ensure stable mining performance without interruption.

Compute Horde (SN12) — Decentralized GPU Computing

Subnet validators across decentralized networks face a fundamental infrastructure challenge: validating computationally intensive tasks requires access to reliable GPU resources, but centralized cloud providers create single points of failure and dependency. Additionally, validators struggle to verify that miners actually performed work honestly without running the computation themselves, which defeats the purpose of decentralization. ComputeHorde solves this by creating a decentralized network of GPU-equipped miners whose work is cryptographically verified through a system of trusted validators. Miners compete to provide compute services fairly, with payment proportional to actual work performed, while validators confirm results using a separate trusted instance before accepting miner compensation. This network approach eliminates centralized infrastructure dependency while maintaining the ability to verify computational integrity across thousands of validators and miners through economic incentives and collateral-based trust mechanisms.

Subnet validators can offload computationally expensive validation tasks to decentralized miners without building or maintaining their own GPU infrastructure, reducing operational costs and capital expenditure.
Machine learning subnet validators can run large-scale model inference and training tasks on commodity hardware (A6000, A100) across distributed miners, scaling validation beyond single-machine constraints.
Validators requiring cryptographic verification of miner work can deposit collateral with the network to enable economic slashing, ensuring honest task completion across untrusted mining participants.
Multi-GPU intensive validation pipelines can be parallelized across hundreds of miners simultaneously, with each miner spawning multiple executors to eliminate traditional per-miner capacity limits.
Validators can integrate ComputeHorde as a primary compute layer with automatic fallback to commercial clouds (RunPod, etc.) via SkyPilot, creating hybrid infrastructure that prioritizes decentralized resources while maintaining service availability.

ComputeHorde (Subnet 12) is a specialized Bittensor subnet that provides decentralized, scalable GPU computing power to validators across the Bittensor network. It transforms untrusted GPUs from miners into trusted compute resources through innovative validation mechanisms, enabling subnet validators to perform large-scale computations without relying on centralized cloud services. The subnet uses hardware classes (currently A6000, A100 planned), collateral-based trust for organic jobs, and a sophisticated incentive mechanism that fairly distributes work and rewards across miners and validators.

ComputeHorde comprises four main components: (1) Facilitator acts as a gateway for organic requests from other subnet validators; (2) Validator receives requests via Facilitator, distributes tasks to miners, and evaluates results using a separate Trusted Miner (an unregistered GPU miner configured only to accept validator tasks) to pre-run validation tasks and establish expected results; (3) Miner accepts job requests from validators and manages executors (spawned instances performing individual containerized tasks) with optional custom ExecutorManager for optimization; (4) Executor runs in restricted environments with limited network access for task communication, docker image downloads, and job data handling. Each miner can spawn multiple executors forming a 'horde' with assigned hardware classes, removing the 256 UID limit. Miners can optionally deposit collateral to access paid organic jobs and employ a DDoS shield for protection.

The SDK enables other subnet validators to submit jobs, perform cross-validation, and integrate fallback cloud services.

Subnet validators can offload computationally expensive validation tasks to decentralized miners without building or maintaining their own GPU infrastructure, reducing operational costs and capital expenditure.
Machine learning subnet validators can run large-scale model inference and training tasks on commodity hardware (A6000, A100) across distributed miners, scaling validation beyond single-machine constraints.
Validators requiring cryptographic verification of miner work can deposit collateral with the network to enable economic slashing, ensuring honest task completion across untrusted mining participants.
Multi-GPU intensive validation pipelines can be parallelized across hundreds of miners simultaneously, with each miner spawning multiple executors to eliminate traditional per-miner capacity limits.
Validators can integrate ComputeHorde as a primary compute layer with automatic fallback to commercial clouds (RunPod, etc.) via SkyPilot, creating hybrid infrastructure that prioritizes decentralized resources while maintaining service availability.

The SDK enables other subnet validators to submit jobs, perform cross-validation, and integrate fallback cloud services.

The scoring mechanism incentivizes miners to perform organic jobs while maintaining fairness. Each block mints executor-seconds per miner-validator-executorClass tuple. The allowance pool recalculates every block as executor counts and validator stakes change.

When starting a job, a validator spends specific allowance blocks to cover runtime. During scoring, validators verify blocks were fresh, unspent, and sufficient; overspending voids incentives. If a miner lowers executor count, tied allowance blocks are invalidated.

Validators score all miners' work globally regardless of who paid, as long as jobs fit valid allowance blocks. The dancing bonus splits each coldkey's score across its hotkeys, with the declared main hotkey taking the largest share. Changing the main hotkey from the previous cycle triggers a 10% boost on the coldkey before splitting.

Each hardware class has a configurable weight that adjusts final scores based on network demand.

Compute Horde

Registration Activity

Compute Horde

Subnet Configuration

Market & Economics

For Miners

For Validators & Stakers

Registration Activity

What does Compute Horde do?

Use Cases

About

Subnet History

Frequently Asked Questions

Development Activity

Recent Activity

Hardware Requirements

What does Compute Horde do?

Use Cases

About

Subnet History

Frequently Asked Questions

Subnet Configuration

Market & Economics

For Miners

For Validators & Stakers

Hardware Requirements

Miner Setup

Validator Setup

Overview

Registration Activity

Participate

Overview

Subnet Configuration

Market & Economics

For Miners

For Validators & Stakers

Registration Activity

What does Compute Horde do?

Use Cases

About

Subnet History

Frequently Asked Questions

Participate

Development Activity

Recent Activity

Hardware Requirements

What does Compute Horde do?

Use Cases

About

Subnet History

Frequently Asked Questions

Subnet Configuration

Market & Economics

For Miners

For Validators & Stakers

Hardware Requirements

Miner Setup

Validator Setup