Hardware Requirements
Bittensor TAO subnets demand solid hardware optimized for AI model inference. Miners need to serve complex models, while validators evaluate those outputs. Below are outlined hardware configurations in entry-level and high-performance categories:
Specs clarification
VRAM first: Large language and vision models need ≥ 20 GB VRAM to run in-memory to avoid penalties.
CPU scaling: More GPUs demand more cores to keep data saturated.
RAM & NVMe: Prevents swap bottlenecks when loading multi-GB checkpoints or caching batches.
Bandwidth: High-throughput links cut query latency for validators.
Power & cooling tips
Use 80 + Gold/Platinum PSUs sized for 70 % load (850 W for single-GPU; 1.5–2 kW for multi-GPU rigs). Ensure front-to-back airflow or liquid cooling; thermal throttling directly lowers TAO rewards.
Last updated