Understand the hardware requirements for running a successful Akash provider.
This guide covers system requirements, hardware specifications, and best practices.
System Requirements
Operating System
- Ubuntu 24.04 LTS Server (officially supported)
- Ensure all nodes are fully updated with latest security patches
CPU Architecture
- x86_64 processors only (officially supported)
- ARM processors are not currently supported
- Applies to all nodes in the cluster
Minimum Hardware Specifications
Single Server (Worker + Control Plane Combined)
Minimum:
- CPU: 8 cores
- RAM: 16GB
- Storage: 150GB SSD
Recommended:
- CPU: 12+ cores
- RAM: 48+ GB
- Storage: 500GB+ SSD
Important Considerations:
- Reserve ~4 CPU cores for Kubernetes system components (control plane + worker overhead)
- Single server setup is good for testing or small providers
- For production, separate control plane and worker nodes are recommended
Control Plane Node (per node, separate setup)
Minimum:
- CPU: 2 cores
- RAM: 4GB
- Storage: 30GB SSD
Recommended:
- CPU: 4 cores
- RAM: 8GB
- Storage: 40GB SSD
For Production (3-node control plane):
- 3x control plane nodes for high availability
- Total: 12 CPU cores, 24GB RAM, 120GB storage
Worker Node (per node, separate setup)
Minimum:
- CPU: 4 cores
- RAM: 8GB
- Storage: 100GB SSD
Recommended:
- CPU: 8+ cores
- RAM: 32+ GB
- Storage: 500GB+ SSD
Important Considerations:
- Reserve ~2 CPU cores for Kubernetes system components
- More resources = more concurrent deployments
- Example: 8 CPU node = ~6 deployments (8 CPU - 2 reserved)
GPU Requirements
Supported GPUs
- NVIDIA GPUs only currently supported
GPU Requirements & Best Practices
One GPU Type Per Node (Required):
- Each node must have only one type of GPU
- Mixing different GPU models on the same node is not supported
- Multiple identical GPUs per node is supported and recommended
One GPU Type Per Provider (Recommended):
- While technically possible to have different GPU types across nodes, it is strongly recommended to standardize on one GPU type per provider
- Having multiple GPU types in one provider (on different nodes) can complicate operations and pricing
Example Configurations:
- AI/ML Node: 4x NVIDIA A100 (all identical)
- Rendering Node: 2x RTX 4090 (all identical)
- Mixed Workload: 8x T4 (all identical)
Storage Guidelines
Root/Ephemeral Storage
- Minimum: 100GB per worker node
- Recommended: 500GB+ per worker node
- Type: SSD or NVMe for performance
Persistent Storage (Optional)
Minimum Requirements:
- Option 1: 4 SSDs across all nodes (minimum)
- Option 2: 2 NVMe SSDs across all nodes (minimum)
- All persistent storage must be same type across entire cluster
- Do not mix storage types (e.g., SSD + NVMe, or beta2 + beta3)
Dedicated Drives:
- These drives must be dedicated exclusively to persistent storage
- Cannot be used for any other purpose (no OS, no ephemeral storage)
- Recommended: Distribute dedicated drives across multiple nodes for redundancy
Storage Classes:
beta1- HDD (Hard Disk Drive)beta2- SSD (Solid State Drive)beta3- NVMe (Non-Volatile Memory Express)
Example Configurations:
- Minimum SSD: 4x 1TB SSDs across 2-4 nodes (beta2)
- Minimum NVMe: 2x 2TB NVMe drives across 2 nodes (beta3)
- Recommended: 6+ drives distributed across 3+ nodes for better redundancy
Network Requirements
Internet Connection
Bandwidth:
- Minimum: 100 Mbps symmetrical
- Recommended: 1+ Gbps symmetrical
- Critical: Upload speed matters as much as download
Latency:
- Low latency preferred (< 10ms to major hubs)
- Affects bid competitiveness
- Important for real-time workloads
IP Addressing
Required:
- Option 1: At least one static public IP
- Option 2: Dynamic DNS (DDNS) service
Optional (for IP leases feature):
- Multiple static IPs
- /29 subnet or larger
- Allows providers to offer dedicated IPs to tenants
Domain Name (Required)
You must own a domain name for your Akash provider:
- A registered domain that you control (e.g.,
yourdomain.com) - Used for provider identification and workload routing
- DNS A records will point to your provider’s public IP
Firewall Rules
Required Open Ports:
80/tcp- HTTP workloads443/tcp- HTTPS workloads8443/tcp- Manifest uploads8444/tcp- Provider gRPC30000-32767/tcp- Kubernetes NodePort range30000-32767/udp- Kubernetes NodePort range (UDP services)
Recommended:
22/tcp- SSH (restrict to your IPs)6443/tcp- Kubernetes API (control plane only, restrict to cluster nodes)
Internal Network
For Multi-Node Clusters:
- Low-latency network between nodes (< 1ms)
- 10 Gbps+ recommended for storage traffic
Cluster Sizing Examples
Small Provider (Home Lab)
Hardware:
- 1 control plane node: 4 CPU, 8GB RAM, 100GB SSD
- 1 worker node: 8 CPU, 32GB RAM, 500GB SSD
Capacity:
- ~6 concurrent deployments
- No GPU support
- 100-200GB ephemeral storage
Cost: ~\(500-1,000 initial +\)30-50/month
Medium Provider (Small Data Center)
Hardware:
- 3 control plane nodes: 4 CPU, 8GB RAM, 100GB SSD each
- 3 worker nodes: 16 CPU, 64GB RAM, 1TB SSD each
Capacity:
- ~40 concurrent deployments
- Optional: 1-2 GPUs per worker
- 1TB+ persistent storage
Cost: ~\(5,000-10,000 initial +\)200-500/month
Large Provider (Data Center)
Hardware:
- 3 control plane nodes: 8 CPU, 16GB RAM, 200GB SSD each
- 10+ worker nodes: 32 CPU, 128GB RAM, 2TB NVMe each
- Dedicated GPU nodes: 8 CPU, 64GB RAM, 4x A100 GPUs
Capacity:
- 200+ concurrent deployments
- 20+ GPU instances
- 10TB+ persistent storage
Cost: ~\(50,000-200,000 initial +\)2,000-10,000/month
Next Steps
Calculate earnings:
