Running a node

Hardware Requirements

The Citrate network supports four node roles, each with different hardware demands. Validator and miner nodes handle consensus and block production, while model host and oracle nodes serve AI inference and bridge attestation duties respectively. This page details the hardware specifications for each role.

Node Role Overview

Before choosing hardware, understand which role you intend to fill:

  • Validator: Participates in BFT consensus, validates blocks, earns staking rewards
  • Miner: Produces GhostDAG blocks, competes for block rewards
  • Model Host: Serves AI inference requests, earns inference fees
  • Oracle: Attests to bridge state, requires a $SNAP NFT, earns attestation fees

A single physical machine can run multiple roles simultaneously, but resource requirements stack. Running a validator + model host on one machine requires the combined resources of both roles.

Validator Node

Validators participate in the Paraconsistent BFT finality layer. They must maintain high uptime and low latency to avoid slashing penalties. We recommend at least 16GB RAM and an NVMe SSD... spinning disks will not keep up with the consensus workload.

ComponentMinimumRecommended
CPU4 cores, 2.5 GHz (x86_64 or ARM64)8+ cores, 3.5 GHz
RAM16 GB32 GB
Storage500 GB NVMe SSD1 TB NVMe SSD
Network50 Mbps symmetric, <100ms to peers1 Gbps symmetric, <50ms
GPUNot requiredNot required
Uptime99.5% minimum99.9%+ target

Storage grows at approximately 50 GB per year under current network usage. Plan for at least two years of growth when provisioning.

Miner Node

Miners produce blocks in the GhostDAG parallel block structure. Mining is CPU-bound -- the Citrate proof-of-work algorithm (CephaloHash) is designed to be ASIC-resistant and favors general-purpose processors.

ComponentMinimumRecommended
CPU8 cores, 3.0 GHz16+ cores, 3.8 GHz (AMD Ryzen 9 / Intel i9)
RAM16 GB32 GB
Storage500 GB NVMe SSD1 TB NVMe SSD
Network100 Mbps symmetric1 Gbps symmetric
GPUNot required for miningNot required

Mining profitability depends on your hashrate relative to the network total. Monitor the network difficulty and your block production rate through the Grafana dashboards described in the Monitoring guide.

Model Host Node

Model hosts serve AI inference requests and earn fees per request. GPU requirements depend on the models you intend to host.

ComponentMinimumRecommended
CPU8 cores, 3.0 GHz16+ cores, 3.5 GHz
RAM32 GB64 GB+
Storage1 TB NVMe SSD2 TB NVMe SSD
Network100 Mbps symmetric1 Gbps symmetric
GPUNVIDIA RTX 3060 (12GB VRAM)NVIDIA A100 (40GB) or H100 (80GB)

GPU selection guide by model size:

Model ParametersMinimum GPURecommended GPU
<1B parametersRTX 3060 (12GB)RTX 4070 Ti (16GB)
1B - 7B parametersRTX 4090 (24GB)A100 (40GB)
7B - 13B parametersA100 (40GB)A100 (80GB)
13B - 70B parameters2x A100 (80GB)4x A100 (80GB) or H100
70B+ parameters4x A100 (80GB)8x H100

Citrate supports model formats including ONNX, SafeTensors, and GGUF. Quantized models (INT8, INT4) significantly reduce VRAM requirements -- a 7B parameter model quantized to INT4 can run on an RTX 4070 Ti with 16GB VRAM.

Oracle Node

Oracle nodes attest to cross-chain bridge state and require a $SNAP NFT to participate. The computational requirements are modest, but uptime and network reliability are critical.

ComponentMinimumRecommended
CPU4 cores, 2.5 GHz8 cores, 3.0 GHz
RAM8 GB16 GB
Storage250 GB SSD500 GB NVMe SSD
Network50 Mbps symmetric100 Mbps symmetric
GPUNot requiredNot required
Special$SNAP NFT in connected wallet$SNAP NFT + staked SALT

Oracle nodes must maintain connections to both the Citrate network and the bridged chains (Ethereum, etc.). Ensure your node has reliable connectivity to RPC endpoints for all supported bridge targets.

Cloud Provider Recommendations

If you prefer cloud hosting, here are equivalent instance types for the major providers. We've had good results with all three, but AWS tends to have the best GPU availability for model hosting:

RoleAWSGCPAzure
Validatorm6i.xlargen2-standard-4Standard_D4s_v5
Minerc6i.4xlargec2-standard-16Standard_F16s_v2
Model Host (small)g5.xlargea2-highgpu-1gStandard_NC4as_T4_v3
Model Host (large)p4d.24xlargea2-ultragpu-8gStandard_ND96amsr_A100_v4
Oraclet3.mediume2-mediumStandard_B2s

Further Reading